diff --git a/Project_CNN.ipynb b/Project_CNN.ipynb
new file mode 100644
index 0000000000000000000000000000000000000000..6a61ed44847d21de482ca4f64af37fb430f4fe1e
--- /dev/null
+++ b/Project_CNN.ipynb
@@ -0,0 +1,499 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 0,
+ "metadata": {
+ "colab": {
+ "provenance": [],
+ "gpuType": "T4"
+ },
+ "kernelspec": {
+ "name": "python3",
+ "display_name": "Python 3"
+ },
+ "language_info": {
+ "name": "python"
+ },
+ "accelerator": "GPU"
+ },
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 0. Imports"
+ ],
+ "metadata": {
+ "id": "s88AvoLByyvb"
+ }
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {
+ "id": "ihpUQFaYvpQW"
+ },
+ "outputs": [],
+ "source": [
+ "import torch\n",
+ "import torchvision\n",
+ "import torchvision.transforms as transforms\n",
+ "from torch.utils.data import DataLoader\n",
+ "import torch.nn as nn\n",
+ "import torch.optim as optim\n",
+ "from torchvision.models import resnet18"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 1. Adat-előkészítés"
+ ],
+ "metadata": {
+ "id": "Z3fiYQZwy4wA"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "train_transform = transforms.Compose([\n",
+ " transforms.RandomHorizontalFlip(),\n",
+ " transforms.RandomCrop(32, padding=4),\n",
+ " transforms.ToTensor(),\n",
+ " transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n",
+ "])\n",
+ "\n",
+ "test_transform = transforms.Compose([\n",
+ " transforms.ToTensor(),\n",
+ " transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n",
+ "])\n",
+ "\n",
+ "trainset = torchvision.datasets.CIFAR10(\n",
+ " root='./data',\n",
+ " train=True,\n",
+ " download=True,\n",
+ " transform=train_transform\n",
+ ")\n",
+ "\n",
+ "testset = torchvision.datasets.CIFAR10(\n",
+ " root='./data',\n",
+ " train=False,\n",
+ " download=True,\n",
+ " transform=test_transform\n",
+ ")\n",
+ "\n",
+ "trainloader = DataLoader(trainset, batch_size=64, shuffle=True)\n",
+ "testloader = DataLoader(testset, batch_size=64, shuffle=False)"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "6yReCwY_zHxd",
+ "outputId": "d8810cca-c44b-4179-98ae-36e4f5bdbd4b"
+ },
+ "execution_count": 2,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stderr",
+ "text": [
+ "100%|██████████| 170M/170M [00:07<00:00, 23.5MB/s]\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 2. Modell architektúra"
+ ],
+ "metadata": {
+ "id": "F7MFKJWzL90w"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "class CIFAR10_ResNet(nn.Module):\n",
+ " def __init__(self, pretrained=True):\n",
+ " super().__init__()\n",
+ " self.model = resnet18(weights='DEFAULT' if pretrained else None)\n",
+ "\n",
+ " # CIFAR-10 kompatibilis módosítások\n",
+ " self.model.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)\n",
+ " self.model.maxpool = nn.Identity() # Eltávolítjuk a max pooling réteget\n",
+ " self.model.fc = nn.Linear(512, 10) # 10 osztály\n",
+ "\n",
+ " def forward(self, x):\n",
+ " return self.model(x)"
+ ],
+ "metadata": {
+ "id": "0QBmgWSRLzWO"
+ },
+ "execution_count": 3,
+ "outputs": []
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 3. Modell inicializálása"
+ ],
+ "metadata": {
+ "id": "fY_G06EKMCbl"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
+ "model = CIFAR10_ResNet(pretrained=True).to(device)\n",
+ "optimizer = optim.Adam(model.parameters(), lr=0.001)\n",
+ "criterion = nn.CrossEntropyLoss()"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "vjNrPunVMOsb",
+ "outputId": "f9882de9-08d1-4b7c-8e17-168690cf234d"
+ },
+ "execution_count": 4,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stderr",
+ "text": [
+ "Downloading: \"https://download.pytorch.org/models/resnet18-f37072fd.pth\" to /root/.cache/torch/hub/checkpoints/resnet18-f37072fd.pth\n",
+ "100%|██████████| 44.7M/44.7M [00:00<00:00, 189MB/s]\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 4. Tanítási ciklus"
+ ],
+ "metadata": {
+ "id": "sJlf_EPNMRqS"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "for epoch in range(30):\n",
+ " model.train()\n",
+ " running_loss = 0.0\n",
+ "\n",
+ " for i, (inputs, labels) in enumerate(trainloader):\n",
+ " inputs, labels = inputs.to(device), labels.to(device)\n",
+ "\n",
+ " optimizer.zero_grad()\n",
+ " outputs = model(inputs)\n",
+ " loss = criterion(outputs, labels)\n",
+ " loss.backward()\n",
+ " optimizer.step()\n",
+ "\n",
+ " running_loss += loss.item()\n",
+ " if i % 100 == 99:\n",
+ " print(f'Epoch: {epoch+1}, Batch: {i+1}, Loss: {running_loss/100:.4f}')\n",
+ " running_loss = 0.0\n",
+ "\n",
+ " # Tesztelés epoch végén\n",
+ " model.eval()\n",
+ " correct = 0\n",
+ " total = 0\n",
+ "\n",
+ " with torch.no_grad():\n",
+ " for inputs, labels in testloader:\n",
+ " inputs, labels = inputs.to(device), labels.to(device)\n",
+ " outputs = model(inputs)\n",
+ " _, predicted = torch.max(outputs.data, 1)\n",
+ " total += labels.size(0)\n",
+ " correct += (predicted == labels).sum().item()\n",
+ "\n",
+ " print(f'Epoch {epoch+1} Tesztpontosság: {100 * correct / total:.2f}%')"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "0D9j4DQIMR_8",
+ "outputId": "f10739b6-fbe0-4481-91b7-df7b3848ffc7"
+ },
+ "execution_count": 5,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Epoch: 1, Batch: 100, Loss: 1.4680\n",
+ "Epoch: 1, Batch: 200, Loss: 0.9620\n",
+ "Epoch: 1, Batch: 300, Loss: 0.8665\n",
+ "Epoch: 1, Batch: 400, Loss: 0.7700\n",
+ "Epoch: 1, Batch: 500, Loss: 0.7240\n",
+ "Epoch: 1, Batch: 600, Loss: 0.6732\n",
+ "Epoch: 1, Batch: 700, Loss: 0.6179\n",
+ "Epoch 1 Tesztpontosság: 77.67%\n",
+ "Epoch: 2, Batch: 100, Loss: 0.5333\n",
+ "Epoch: 2, Batch: 200, Loss: 0.5511\n",
+ "Epoch: 2, Batch: 300, Loss: 0.5397\n",
+ "Epoch: 2, Batch: 400, Loss: 0.5328\n",
+ "Epoch: 2, Batch: 500, Loss: 0.4696\n",
+ "Epoch: 2, Batch: 600, Loss: 0.4975\n",
+ "Epoch: 2, Batch: 700, Loss: 0.4947\n",
+ "Epoch 2 Tesztpontosság: 82.47%\n",
+ "Epoch: 3, Batch: 100, Loss: 0.4422\n",
+ "Epoch: 3, Batch: 200, Loss: 0.4207\n",
+ "Epoch: 3, Batch: 300, Loss: 0.4368\n",
+ "Epoch: 3, Batch: 400, Loss: 0.4378\n",
+ "Epoch: 3, Batch: 500, Loss: 0.4034\n",
+ "Epoch: 3, Batch: 600, Loss: 0.4169\n",
+ "Epoch: 3, Batch: 700, Loss: 0.4098\n",
+ "Epoch 3 Tesztpontosság: 87.10%\n",
+ "Epoch: 4, Batch: 100, Loss: 0.3536\n",
+ "Epoch: 4, Batch: 200, Loss: 0.3616\n",
+ "Epoch: 4, Batch: 300, Loss: 0.3752\n",
+ "Epoch: 4, Batch: 400, Loss: 0.3411\n",
+ "Epoch: 4, Batch: 500, Loss: 0.3639\n",
+ "Epoch: 4, Batch: 600, Loss: 0.3431\n",
+ "Epoch: 4, Batch: 700, Loss: 0.3496\n",
+ "Epoch 4 Tesztpontosság: 86.70%\n",
+ "Epoch: 5, Batch: 100, Loss: 0.3108\n",
+ "Epoch: 5, Batch: 200, Loss: 0.3237\n",
+ "Epoch: 5, Batch: 300, Loss: 0.3030\n",
+ "Epoch: 5, Batch: 400, Loss: 0.3255\n",
+ "Epoch: 5, Batch: 500, Loss: 0.3062\n",
+ "Epoch: 5, Batch: 600, Loss: 0.3203\n",
+ "Epoch: 5, Batch: 700, Loss: 0.3354\n",
+ "Epoch 5 Tesztpontosság: 88.46%\n",
+ "Epoch: 6, Batch: 100, Loss: 0.2724\n",
+ "Epoch: 6, Batch: 200, Loss: 0.2630\n",
+ "Epoch: 6, Batch: 300, Loss: 0.2843\n",
+ "Epoch: 6, Batch: 400, Loss: 0.2824\n",
+ "Epoch: 6, Batch: 500, Loss: 0.2750\n",
+ "Epoch: 6, Batch: 600, Loss: 0.2951\n",
+ "Epoch: 6, Batch: 700, Loss: 0.2883\n",
+ "Epoch 6 Tesztpontosság: 88.35%\n",
+ "Epoch: 7, Batch: 100, Loss: 0.2395\n",
+ "Epoch: 7, Batch: 200, Loss: 0.2622\n",
+ "Epoch: 7, Batch: 300, Loss: 0.2538\n",
+ "Epoch: 7, Batch: 400, Loss: 0.2578\n",
+ "Epoch: 7, Batch: 500, Loss: 0.2514\n",
+ "Epoch: 7, Batch: 600, Loss: 0.2480\n",
+ "Epoch: 7, Batch: 700, Loss: 0.2719\n",
+ "Epoch 7 Tesztpontosság: 89.58%\n",
+ "Epoch: 8, Batch: 100, Loss: 0.2339\n",
+ "Epoch: 8, Batch: 200, Loss: 0.2230\n",
+ "Epoch: 8, Batch: 300, Loss: 0.2099\n",
+ "Epoch: 8, Batch: 400, Loss: 0.2184\n",
+ "Epoch: 8, Batch: 500, Loss: 0.2406\n",
+ "Epoch: 8, Batch: 600, Loss: 0.2465\n",
+ "Epoch: 8, Batch: 700, Loss: 0.2453\n",
+ "Epoch 8 Tesztpontosság: 89.00%\n",
+ "Epoch: 9, Batch: 100, Loss: 0.2055\n",
+ "Epoch: 9, Batch: 200, Loss: 0.1891\n",
+ "Epoch: 9, Batch: 300, Loss: 0.2235\n",
+ "Epoch: 9, Batch: 400, Loss: 0.2157\n",
+ "Epoch: 9, Batch: 500, Loss: 0.2056\n",
+ "Epoch: 9, Batch: 600, Loss: 0.2027\n",
+ "Epoch: 9, Batch: 700, Loss: 0.2140\n",
+ "Epoch 9 Tesztpontosság: 91.07%\n",
+ "Epoch: 10, Batch: 100, Loss: 0.1896\n",
+ "Epoch: 10, Batch: 200, Loss: 0.1978\n",
+ "Epoch: 10, Batch: 300, Loss: 0.1876\n",
+ "Epoch: 10, Batch: 400, Loss: 0.1940\n",
+ "Epoch: 10, Batch: 500, Loss: 0.1812\n",
+ "Epoch: 10, Batch: 600, Loss: 0.2067\n",
+ "Epoch: 10, Batch: 700, Loss: 0.2033\n",
+ "Epoch 10 Tesztpontosság: 90.68%\n",
+ "Epoch: 11, Batch: 100, Loss: 0.1759\n",
+ "Epoch: 11, Batch: 200, Loss: 0.1731\n",
+ "Epoch: 11, Batch: 300, Loss: 0.1723\n",
+ "Epoch: 11, Batch: 400, Loss: 0.1741\n",
+ "Epoch: 11, Batch: 500, Loss: 0.1707\n",
+ "Epoch: 11, Batch: 600, Loss: 0.1888\n",
+ "Epoch: 11, Batch: 700, Loss: 0.1947\n",
+ "Epoch 11 Tesztpontosság: 88.09%\n",
+ "Epoch: 12, Batch: 100, Loss: 0.1695\n",
+ "Epoch: 12, Batch: 200, Loss: 0.1626\n",
+ "Epoch: 12, Batch: 300, Loss: 0.1641\n",
+ "Epoch: 12, Batch: 400, Loss: 0.1712\n",
+ "Epoch: 12, Batch: 500, Loss: 0.1614\n",
+ "Epoch: 12, Batch: 600, Loss: 0.1699\n",
+ "Epoch: 12, Batch: 700, Loss: 0.1678\n",
+ "Epoch 12 Tesztpontosság: 90.86%\n",
+ "Epoch: 13, Batch: 100, Loss: 0.1418\n",
+ "Epoch: 13, Batch: 200, Loss: 0.1360\n",
+ "Epoch: 13, Batch: 300, Loss: 0.1517\n",
+ "Epoch: 13, Batch: 400, Loss: 0.1409\n",
+ "Epoch: 13, Batch: 500, Loss: 0.1710\n",
+ "Epoch: 13, Batch: 600, Loss: 0.1737\n",
+ "Epoch: 13, Batch: 700, Loss: 0.1523\n",
+ "Epoch 13 Tesztpontosság: 91.47%\n",
+ "Epoch: 14, Batch: 100, Loss: 0.1171\n",
+ "Epoch: 14, Batch: 200, Loss: 0.1287\n",
+ "Epoch: 14, Batch: 300, Loss: 0.1397\n",
+ "Epoch: 14, Batch: 400, Loss: 0.1522\n",
+ "Epoch: 14, Batch: 500, Loss: 0.1401\n",
+ "Epoch: 14, Batch: 600, Loss: 0.1544\n",
+ "Epoch: 14, Batch: 700, Loss: 0.1376\n",
+ "Epoch 14 Tesztpontosság: 91.91%\n",
+ "Epoch: 15, Batch: 100, Loss: 0.1405\n",
+ "Epoch: 15, Batch: 200, Loss: 0.1327\n",
+ "Epoch: 15, Batch: 300, Loss: 0.1288\n",
+ "Epoch: 15, Batch: 400, Loss: 0.1285\n",
+ "Epoch: 15, Batch: 500, Loss: 0.1299\n",
+ "Epoch: 15, Batch: 600, Loss: 0.1522\n",
+ "Epoch: 15, Batch: 700, Loss: 0.1458\n",
+ "Epoch 15 Tesztpontosság: 91.92%\n",
+ "Epoch: 16, Batch: 100, Loss: 0.1140\n",
+ "Epoch: 16, Batch: 200, Loss: 0.1281\n",
+ "Epoch: 16, Batch: 300, Loss: 0.1214\n",
+ "Epoch: 16, Batch: 400, Loss: 0.1279\n",
+ "Epoch: 16, Batch: 500, Loss: 0.1150\n",
+ "Epoch: 16, Batch: 600, Loss: 0.1293\n",
+ "Epoch: 16, Batch: 700, Loss: 0.1251\n",
+ "Epoch 16 Tesztpontosság: 91.35%\n",
+ "Epoch: 17, Batch: 100, Loss: 0.1142\n",
+ "Epoch: 17, Batch: 200, Loss: 0.1135\n",
+ "Epoch: 17, Batch: 300, Loss: 0.1085\n",
+ "Epoch: 17, Batch: 400, Loss: 0.1151\n",
+ "Epoch: 17, Batch: 500, Loss: 0.1299\n",
+ "Epoch: 17, Batch: 600, Loss: 0.1201\n",
+ "Epoch: 17, Batch: 700, Loss: 0.1175\n",
+ "Epoch 17 Tesztpontosság: 91.71%\n",
+ "Epoch: 18, Batch: 100, Loss: 0.1007\n",
+ "Epoch: 18, Batch: 200, Loss: 0.1121\n",
+ "Epoch: 18, Batch: 300, Loss: 0.1097\n",
+ "Epoch: 18, Batch: 400, Loss: 0.1106\n",
+ "Epoch: 18, Batch: 500, Loss: 0.0980\n",
+ "Epoch: 18, Batch: 600, Loss: 0.1032\n",
+ "Epoch: 18, Batch: 700, Loss: 0.1204\n",
+ "Epoch 18 Tesztpontosság: 91.99%\n",
+ "Epoch: 19, Batch: 100, Loss: 0.1125\n",
+ "Epoch: 19, Batch: 200, Loss: 0.0860\n",
+ "Epoch: 19, Batch: 300, Loss: 0.1014\n",
+ "Epoch: 19, Batch: 400, Loss: 0.1171\n",
+ "Epoch: 19, Batch: 500, Loss: 0.1103\n",
+ "Epoch: 19, Batch: 600, Loss: 0.1121\n",
+ "Epoch: 19, Batch: 700, Loss: 0.1062\n",
+ "Epoch 19 Tesztpontosság: 92.58%\n",
+ "Epoch: 20, Batch: 100, Loss: 0.0826\n",
+ "Epoch: 20, Batch: 200, Loss: 0.0860\n",
+ "Epoch: 20, Batch: 300, Loss: 0.0905\n",
+ "Epoch: 20, Batch: 400, Loss: 0.0881\n",
+ "Epoch: 20, Batch: 500, Loss: 0.0938\n",
+ "Epoch: 20, Batch: 600, Loss: 0.0906\n",
+ "Epoch: 20, Batch: 700, Loss: 0.1004\n",
+ "Epoch 20 Tesztpontosság: 90.97%\n",
+ "Epoch: 21, Batch: 100, Loss: 0.0998\n",
+ "Epoch: 21, Batch: 200, Loss: 0.1026\n",
+ "Epoch: 21, Batch: 300, Loss: 0.0826\n",
+ "Epoch: 21, Batch: 400, Loss: 0.0920\n",
+ "Epoch: 21, Batch: 500, Loss: 0.1117\n",
+ "Epoch: 21, Batch: 600, Loss: 0.0951\n",
+ "Epoch: 21, Batch: 700, Loss: 0.0983\n",
+ "Epoch 21 Tesztpontosság: 92.63%\n",
+ "Epoch: 22, Batch: 100, Loss: 0.0791\n",
+ "Epoch: 22, Batch: 200, Loss: 0.0790\n",
+ "Epoch: 22, Batch: 300, Loss: 0.0890\n",
+ "Epoch: 22, Batch: 400, Loss: 0.0817\n",
+ "Epoch: 22, Batch: 500, Loss: 0.0875\n",
+ "Epoch: 22, Batch: 600, Loss: 0.0988\n",
+ "Epoch: 22, Batch: 700, Loss: 0.0929\n",
+ "Epoch 22 Tesztpontosság: 91.48%\n",
+ "Epoch: 23, Batch: 100, Loss: 0.0729\n",
+ "Epoch: 23, Batch: 200, Loss: 0.0762\n",
+ "Epoch: 23, Batch: 300, Loss: 0.0842\n",
+ "Epoch: 23, Batch: 400, Loss: 0.0796\n",
+ "Epoch: 23, Batch: 500, Loss: 0.0856\n",
+ "Epoch: 23, Batch: 600, Loss: 0.0948\n",
+ "Epoch: 23, Batch: 700, Loss: 0.0938\n",
+ "Epoch 23 Tesztpontosság: 92.47%\n",
+ "Epoch: 24, Batch: 100, Loss: 0.0694\n",
+ "Epoch: 24, Batch: 200, Loss: 0.0818\n",
+ "Epoch: 24, Batch: 300, Loss: 0.0998\n",
+ "Epoch: 24, Batch: 400, Loss: 0.0775\n",
+ "Epoch: 24, Batch: 500, Loss: 0.0724\n",
+ "Epoch: 24, Batch: 600, Loss: 0.0851\n",
+ "Epoch: 24, Batch: 700, Loss: 0.0796\n",
+ "Epoch 24 Tesztpontosság: 91.97%\n",
+ "Epoch: 25, Batch: 100, Loss: 0.0762\n",
+ "Epoch: 25, Batch: 200, Loss: 0.0717\n",
+ "Epoch: 25, Batch: 300, Loss: 0.0770\n",
+ "Epoch: 25, Batch: 400, Loss: 0.0848\n",
+ "Epoch: 25, Batch: 500, Loss: 0.0819\n",
+ "Epoch: 25, Batch: 600, Loss: 0.0759\n",
+ "Epoch: 25, Batch: 700, Loss: 0.0759\n",
+ "Epoch 25 Tesztpontosság: 92.07%\n",
+ "Epoch: 26, Batch: 100, Loss: 0.0739\n",
+ "Epoch: 26, Batch: 200, Loss: 0.0572\n",
+ "Epoch: 26, Batch: 300, Loss: 0.0765\n",
+ "Epoch: 26, Batch: 400, Loss: 0.0729\n",
+ "Epoch: 26, Batch: 500, Loss: 0.0710\n",
+ "Epoch: 26, Batch: 600, Loss: 0.0701\n",
+ "Epoch: 26, Batch: 700, Loss: 0.0768\n",
+ "Epoch 26 Tesztpontosság: 92.46%\n",
+ "Epoch: 27, Batch: 100, Loss: 0.0641\n",
+ "Epoch: 27, Batch: 200, Loss: 0.0715\n",
+ "Epoch: 27, Batch: 300, Loss: 0.0572\n",
+ "Epoch: 27, Batch: 400, Loss: 0.0766\n",
+ "Epoch: 27, Batch: 500, Loss: 0.0809\n",
+ "Epoch: 27, Batch: 600, Loss: 0.0635\n",
+ "Epoch: 27, Batch: 700, Loss: 0.0622\n",
+ "Epoch 27 Tesztpontosság: 92.40%\n",
+ "Epoch: 28, Batch: 100, Loss: 0.0683\n",
+ "Epoch: 28, Batch: 200, Loss: 0.0676\n",
+ "Epoch: 28, Batch: 300, Loss: 0.0878\n",
+ "Epoch: 28, Batch: 400, Loss: 0.0681\n",
+ "Epoch: 28, Batch: 500, Loss: 0.0660\n",
+ "Epoch: 28, Batch: 600, Loss: 0.0616\n",
+ "Epoch: 28, Batch: 700, Loss: 0.0702\n",
+ "Epoch 28 Tesztpontosság: 92.49%\n",
+ "Epoch: 29, Batch: 100, Loss: 0.0736\n",
+ "Epoch: 29, Batch: 200, Loss: 0.0594\n",
+ "Epoch: 29, Batch: 300, Loss: 0.0706\n",
+ "Epoch: 29, Batch: 400, Loss: 0.0659\n",
+ "Epoch: 29, Batch: 500, Loss: 0.0609\n",
+ "Epoch: 29, Batch: 600, Loss: 0.0652\n",
+ "Epoch: 29, Batch: 700, Loss: 0.0598\n",
+ "Epoch 29 Tesztpontosság: 91.98%\n",
+ "Epoch: 30, Batch: 100, Loss: 0.0463\n",
+ "Epoch: 30, Batch: 200, Loss: 0.0531\n",
+ "Epoch: 30, Batch: 300, Loss: 0.0564\n",
+ "Epoch: 30, Batch: 400, Loss: 0.0621\n",
+ "Epoch: 30, Batch: 500, Loss: 0.0658\n",
+ "Epoch: 30, Batch: 600, Loss: 0.0604\n",
+ "Epoch: 30, Batch: 700, Loss: 0.0585\n",
+ "Epoch 30 Tesztpontosság: 92.40%\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# Mentés"
+ ],
+ "metadata": {
+ "id": "NXIh9r4mXKti"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "torch.save({\n",
+ " 'model_state_dict': model.state_dict(),\n",
+ " 'conv1_weights': model.model.conv1.weight,\n",
+ " 'fc_weights': model.model.fc.weight,\n",
+ " 'fc_bias': model.model.fc.bias\n",
+ "}, 'cnn_model.pth')"
+ ],
+ "metadata": {
+ "id": "wC2gba2EXRjg"
+ },
+ "execution_count": 6,
+ "outputs": []
+ }
+ ]
+}
\ No newline at end of file
diff --git a/Project_CeNN.ipynb b/Project_CeNN.ipynb
new file mode 100644
index 0000000000000000000000000000000000000000..b6e1a45e32ba3c238b7fb5fa0c210d31244f3347
--- /dev/null
+++ b/Project_CeNN.ipynb
@@ -0,0 +1,742 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 0,
+ "metadata": {
+ "colab": {
+ "provenance": [],
+ "gpuType": "T4"
+ },
+ "kernelspec": {
+ "name": "python3",
+ "display_name": "Python 3"
+ },
+ "language_info": {
+ "name": "python"
+ },
+ "accelerator": "GPU"
+ },
+ "cells": [
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {
+ "id": "8UDCqMDw_PWm"
+ },
+ "outputs": [],
+ "source": [
+ "import torch\n",
+ "import torchvision\n",
+ "import torchvision.transforms as transforms\n",
+ "from torch.utils.data import DataLoader\n",
+ "import torch.nn as nn"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
+ "print(f\"Használt eszköz: {device}\")"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "iMNqDPPNIzIq",
+ "outputId": "b9b27e02-9489-4998-b28e-f67df0fd2bd1"
+ },
+ "execution_count": 2,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Használt eszköz: cuda:0\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 1. Adathalmaz betöltése"
+ ],
+ "metadata": {
+ "id": "JyR4J1W0ALuy"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "transform = transforms.Compose([\n",
+ " transforms.ToTensor(),\n",
+ " transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n",
+ "])\n",
+ "\n",
+ "trainset = torchvision.datasets.CIFAR10(\n",
+ " root='./data',\n",
+ " train=True,\n",
+ " download=True,\n",
+ " transform=transform\n",
+ ")\n",
+ "\n",
+ "trainloader = DataLoader(\n",
+ " trainset,\n",
+ " batch_size=64,\n",
+ " shuffle=True,\n",
+ " num_workers=2\n",
+ ")"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "JGYQXMp6_hzN",
+ "outputId": "4fdd0075-1820-4bcb-884e-cfb46ca7a5af"
+ },
+ "execution_count": 3,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stderr",
+ "text": [
+ "100%|██████████| 170M/170M [00:12<00:00, 13.3MB/s]\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 2. CeNN modell osztály"
+ ],
+ "metadata": {
+ "id": "hw8Y_wi_AaDM"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "class CeNN(nn.Module):\n",
+ " def __init__(self, h=0.1, iter_num=20, alpha=0.01):\n",
+ " super().__init__()\n",
+ " self.h = h\n",
+ " self.iter_num = iter_num\n",
+ " self.alpha = alpha\n",
+ "\n",
+ " # A: visszacsatoló kernel\n",
+ " self.A = nn.Parameter(torch.randn(1, 3, 3, 3)) # 3 input csatorna (RGB)\n",
+ " # B: előrecsatoló kernel\n",
+ " self.B = nn.Parameter(torch.randn(1, 3, 3, 3)) # 3 input csatorna\n",
+ " # Z: bias\n",
+ " self.Z = nn.Parameter(torch.randn(1))\n",
+ "\n",
+ " def forward(self, U):\n",
+ " x = U # Kezdeti állapot\n",
+ " for _ in range(self.iter_num):\n",
+ " # Aktivációs függvény\n",
+ " y = torch.minimum(x, 1 + self.alpha * x)\n",
+ " y = torch.maximum(y, -1 + self.alpha * y)\n",
+ "\n",
+ " # Konvolúciós számítások\n",
+ " fwd = torch.nn.functional.conv2d(U, self.B, padding=1) + self.Z\n",
+ " bwd = torch.nn.functional.conv2d(y, self.A, padding=1)\n",
+ "\n",
+ " # Euler-lépés\n",
+ " x = x + self.h * (-x + bwd + fwd)\n",
+ "\n",
+ " # Végső aktiváció\n",
+ " out = torch.minimum(x, 1 + self.alpha * x)\n",
+ " out = torch.maximum(out, -1 + self.alpha * out)\n",
+ " return out"
+ ],
+ "metadata": {
+ "id": "gYY4Wi04AZrw"
+ },
+ "execution_count": 4,
+ "outputs": []
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 3. Modell és optimalizáló inicializálása"
+ ],
+ "metadata": {
+ "id": "xUgxBaQ8BgQK"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "model = CeNN(h=0.1, iter_num=20).to(device)\n",
+ "optimizer = torch.optim.Adam([model.A, model.B, model.Z], lr=0.01)"
+ ],
+ "metadata": {
+ "id": "pGveFopjBhKc"
+ },
+ "execution_count": 5,
+ "outputs": []
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 4. Tanítási ciklus"
+ ],
+ "metadata": {
+ "id": "HR1VbUxSCGWI"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "for epoch in range(30): # 30 epoch\n",
+ " for batch_id, (data, labels) in enumerate(trainloader):\n",
+ " data = data.to(device)\n",
+ " labels = labels.to(device)\n",
+ " optimizer.zero_grad()\n",
+ "\n",
+ " # Előrehaladás\n",
+ " outputs = model(data)\n",
+ "\n",
+ " # Célértékek generálása (bináris eset: 0. osztály = -1, egyéb = 1)\n",
+ " expected_out = torch.ones_like(outputs)\n",
+ " expected_out[labels == 0] = -1\n",
+ "\n",
+ " # Hibaszámítás\n",
+ " loss = torch.mean((outputs - expected_out)**2)\n",
+ "\n",
+ " # Visszaterjesztés\n",
+ " loss.backward()\n",
+ " optimizer.step()\n",
+ "\n",
+ " # Accuracy számítás\n",
+ " correct_dist = torch.mean((outputs - expected_out)**2, dim=[1,2,3])\n",
+ " incorrect_dist = torch.mean((outputs - (-expected_out))**2, dim=[1,2,3])\n",
+ " accuracy = (correct_dist < incorrect_dist).float().mean()\n",
+ "\n",
+ " if batch_id % 50 == 0:\n",
+ " print(f\"Epoch: {epoch+1}, Batch: {batch_id}, Loss: {loss.item():.4f}, Accuracy: {accuracy.item():.4f}\")"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "h1U7kJnfBjjF",
+ "outputId": "ac4b384b-541f-4d0b-8af9-0afb6dcdaea4"
+ },
+ "execution_count": 6,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Epoch: 1, Batch: 0, Loss: 3.2504, Accuracy: 0.0781\n",
+ "Epoch: 1, Batch: 50, Loss: 1.7928, Accuracy: 0.0938\n",
+ "Epoch: 1, Batch: 100, Loss: 1.0809, Accuracy: 0.4062\n",
+ "Epoch: 1, Batch: 150, Loss: 0.9643, Accuracy: 0.6406\n",
+ "Epoch: 1, Batch: 200, Loss: 0.9468, Accuracy: 0.7344\n",
+ "Epoch: 1, Batch: 250, Loss: 0.9145, Accuracy: 0.6719\n",
+ "Epoch: 1, Batch: 300, Loss: 0.7937, Accuracy: 0.7500\n",
+ "Epoch: 1, Batch: 350, Loss: 0.3365, Accuracy: 0.9219\n",
+ "Epoch: 1, Batch: 400, Loss: 0.2891, Accuracy: 0.9219\n",
+ "Epoch: 1, Batch: 450, Loss: 0.6049, Accuracy: 0.7812\n",
+ "Epoch: 1, Batch: 500, Loss: 0.2799, Accuracy: 0.9219\n",
+ "Epoch: 1, Batch: 550, Loss: 0.3563, Accuracy: 0.8906\n",
+ "Epoch: 1, Batch: 600, Loss: 0.4664, Accuracy: 0.8438\n",
+ "Epoch: 1, Batch: 650, Loss: 0.2913, Accuracy: 0.9219\n",
+ "Epoch: 1, Batch: 700, Loss: 0.3457, Accuracy: 0.9062\n",
+ "Epoch: 1, Batch: 750, Loss: 0.3231, Accuracy: 0.8906\n",
+ "Epoch: 2, Batch: 0, Loss: 0.3062, Accuracy: 0.9219\n",
+ "Epoch: 2, Batch: 50, Loss: 0.2287, Accuracy: 0.9375\n",
+ "Epoch: 2, Batch: 100, Loss: 0.1790, Accuracy: 0.9375\n",
+ "Epoch: 2, Batch: 150, Loss: 0.2975, Accuracy: 0.9219\n",
+ "Epoch: 2, Batch: 200, Loss: 0.4369, Accuracy: 0.8594\n",
+ "Epoch: 2, Batch: 250, Loss: 0.2556, Accuracy: 0.9062\n",
+ "Epoch: 2, Batch: 300, Loss: 0.3841, Accuracy: 0.8906\n",
+ "Epoch: 2, Batch: 350, Loss: 0.2146, Accuracy: 0.9531\n",
+ "Epoch: 2, Batch: 400, Loss: 0.3204, Accuracy: 0.9062\n",
+ "Epoch: 2, Batch: 450, Loss: 0.4400, Accuracy: 0.8438\n",
+ "Epoch: 2, Batch: 500, Loss: 0.3326, Accuracy: 0.9062\n",
+ "Epoch: 2, Batch: 550, Loss: 0.3200, Accuracy: 0.9062\n",
+ "Epoch: 2, Batch: 600, Loss: 0.3158, Accuracy: 0.8906\n",
+ "Epoch: 2, Batch: 650, Loss: 0.4220, Accuracy: 0.8594\n",
+ "Epoch: 2, Batch: 700, Loss: 0.4758, Accuracy: 0.8594\n",
+ "Epoch: 2, Batch: 750, Loss: 0.4251, Accuracy: 0.8750\n",
+ "Epoch: 3, Batch: 0, Loss: 0.5105, Accuracy: 0.8438\n",
+ "Epoch: 3, Batch: 50, Loss: 0.1849, Accuracy: 0.9531\n",
+ "Epoch: 3, Batch: 100, Loss: 0.2802, Accuracy: 0.9219\n",
+ "Epoch: 3, Batch: 150, Loss: 0.1867, Accuracy: 0.9531\n",
+ "Epoch: 3, Batch: 200, Loss: 0.2983, Accuracy: 0.9219\n",
+ "Epoch: 3, Batch: 250, Loss: 0.2900, Accuracy: 0.9219\n",
+ "Epoch: 3, Batch: 300, Loss: 0.2088, Accuracy: 0.9375\n",
+ "Epoch: 3, Batch: 350, Loss: 0.2820, Accuracy: 0.9219\n",
+ "Epoch: 3, Batch: 400, Loss: 0.5264, Accuracy: 0.8438\n",
+ "Epoch: 3, Batch: 450, Loss: 0.3403, Accuracy: 0.9062\n",
+ "Epoch: 3, Batch: 500, Loss: 0.4824, Accuracy: 0.8438\n",
+ "Epoch: 3, Batch: 550, Loss: 0.2307, Accuracy: 0.9219\n",
+ "Epoch: 3, Batch: 600, Loss: 0.3859, Accuracy: 0.8906\n",
+ "Epoch: 3, Batch: 650, Loss: 0.3170, Accuracy: 0.9062\n",
+ "Epoch: 3, Batch: 700, Loss: 0.4403, Accuracy: 0.8438\n",
+ "Epoch: 3, Batch: 750, Loss: 0.3150, Accuracy: 0.9219\n",
+ "Epoch: 4, Batch: 0, Loss: 0.3866, Accuracy: 0.8750\n",
+ "Epoch: 4, Batch: 50, Loss: 0.3570, Accuracy: 0.8906\n",
+ "Epoch: 4, Batch: 100, Loss: 0.3416, Accuracy: 0.8906\n",
+ "Epoch: 4, Batch: 150, Loss: 0.4609, Accuracy: 0.8594\n",
+ "Epoch: 4, Batch: 200, Loss: 0.2462, Accuracy: 0.9531\n",
+ "Epoch: 4, Batch: 250, Loss: 0.3144, Accuracy: 0.9062\n",
+ "Epoch: 4, Batch: 300, Loss: 0.2060, Accuracy: 0.9375\n",
+ "Epoch: 4, Batch: 350, Loss: 0.5053, Accuracy: 0.8438\n",
+ "Epoch: 4, Batch: 400, Loss: 0.3569, Accuracy: 0.8750\n",
+ "Epoch: 4, Batch: 450, Loss: 0.3021, Accuracy: 0.9219\n",
+ "Epoch: 4, Batch: 500, Loss: 0.4802, Accuracy: 0.8594\n",
+ "Epoch: 4, Batch: 550, Loss: 0.2447, Accuracy: 0.9375\n",
+ "Epoch: 4, Batch: 600, Loss: 0.3382, Accuracy: 0.9062\n",
+ "Epoch: 4, Batch: 650, Loss: 0.1233, Accuracy: 0.9688\n",
+ "Epoch: 4, Batch: 700, Loss: 0.3592, Accuracy: 0.8750\n",
+ "Epoch: 4, Batch: 750, Loss: 0.2131, Accuracy: 0.9375\n",
+ "Epoch: 5, Batch: 0, Loss: 0.1955, Accuracy: 0.9531\n",
+ "Epoch: 5, Batch: 50, Loss: 0.2461, Accuracy: 0.9219\n",
+ "Epoch: 5, Batch: 100, Loss: 0.2585, Accuracy: 0.9375\n",
+ "Epoch: 5, Batch: 150, Loss: 0.3313, Accuracy: 0.9062\n",
+ "Epoch: 5, Batch: 200, Loss: 0.1465, Accuracy: 0.9688\n",
+ "Epoch: 5, Batch: 250, Loss: 0.2710, Accuracy: 0.9219\n",
+ "Epoch: 5, Batch: 300, Loss: 0.1964, Accuracy: 0.9531\n",
+ "Epoch: 5, Batch: 350, Loss: 0.5045, Accuracy: 0.8438\n",
+ "Epoch: 5, Batch: 400, Loss: 0.3234, Accuracy: 0.9062\n",
+ "Epoch: 5, Batch: 450, Loss: 0.4351, Accuracy: 0.8594\n",
+ "Epoch: 5, Batch: 500, Loss: 0.4981, Accuracy: 0.8438\n",
+ "Epoch: 5, Batch: 550, Loss: 0.3685, Accuracy: 0.8906\n",
+ "Epoch: 5, Batch: 600, Loss: 0.3898, Accuracy: 0.8906\n",
+ "Epoch: 5, Batch: 650, Loss: 0.2581, Accuracy: 0.9219\n",
+ "Epoch: 5, Batch: 700, Loss: 0.3540, Accuracy: 0.9062\n",
+ "Epoch: 5, Batch: 750, Loss: 0.6129, Accuracy: 0.8125\n",
+ "Epoch: 6, Batch: 0, Loss: 0.2730, Accuracy: 0.9219\n",
+ "Epoch: 6, Batch: 50, Loss: 0.3072, Accuracy: 0.9219\n",
+ "Epoch: 6, Batch: 100, Loss: 0.3900, Accuracy: 0.8750\n",
+ "Epoch: 6, Batch: 150, Loss: 0.2398, Accuracy: 0.9375\n",
+ "Epoch: 6, Batch: 200, Loss: 0.2062, Accuracy: 0.9531\n",
+ "Epoch: 6, Batch: 250, Loss: 0.5174, Accuracy: 0.8438\n",
+ "Epoch: 6, Batch: 300, Loss: 0.2991, Accuracy: 0.9062\n",
+ "Epoch: 6, Batch: 350, Loss: 0.4451, Accuracy: 0.8594\n",
+ "Epoch: 6, Batch: 400, Loss: 0.2166, Accuracy: 0.9531\n",
+ "Epoch: 6, Batch: 450, Loss: 0.4295, Accuracy: 0.8594\n",
+ "Epoch: 6, Batch: 500, Loss: 0.2640, Accuracy: 0.9219\n",
+ "Epoch: 6, Batch: 550, Loss: 0.3824, Accuracy: 0.8906\n",
+ "Epoch: 6, Batch: 600, Loss: 0.3471, Accuracy: 0.8906\n",
+ "Epoch: 6, Batch: 650, Loss: 0.3431, Accuracy: 0.9219\n",
+ "Epoch: 6, Batch: 700, Loss: 0.2478, Accuracy: 0.9375\n",
+ "Epoch: 6, Batch: 750, Loss: 0.2856, Accuracy: 0.9219\n",
+ "Epoch: 7, Batch: 0, Loss: 0.2095, Accuracy: 0.9375\n",
+ "Epoch: 7, Batch: 50, Loss: 0.2560, Accuracy: 0.9219\n",
+ "Epoch: 7, Batch: 100, Loss: 0.4533, Accuracy: 0.8594\n",
+ "Epoch: 7, Batch: 150, Loss: 0.1734, Accuracy: 0.9531\n",
+ "Epoch: 7, Batch: 200, Loss: 0.3265, Accuracy: 0.9062\n",
+ "Epoch: 7, Batch: 250, Loss: 0.2912, Accuracy: 0.9062\n",
+ "Epoch: 7, Batch: 300, Loss: 0.2897, Accuracy: 0.9062\n",
+ "Epoch: 7, Batch: 350, Loss: 0.3915, Accuracy: 0.8906\n",
+ "Epoch: 7, Batch: 400, Loss: 0.4599, Accuracy: 0.8594\n",
+ "Epoch: 7, Batch: 450, Loss: 0.2889, Accuracy: 0.9219\n",
+ "Epoch: 7, Batch: 500, Loss: 0.3306, Accuracy: 0.9062\n",
+ "Epoch: 7, Batch: 550, Loss: 0.2967, Accuracy: 0.9375\n",
+ "Epoch: 7, Batch: 600, Loss: 0.3698, Accuracy: 0.8906\n",
+ "Epoch: 7, Batch: 650, Loss: 0.2918, Accuracy: 0.9062\n",
+ "Epoch: 7, Batch: 700, Loss: 0.1363, Accuracy: 0.9688\n",
+ "Epoch: 7, Batch: 750, Loss: 0.2358, Accuracy: 0.9375\n",
+ "Epoch: 8, Batch: 0, Loss: 0.4078, Accuracy: 0.8906\n",
+ "Epoch: 8, Batch: 50, Loss: 0.2609, Accuracy: 0.9219\n",
+ "Epoch: 8, Batch: 100, Loss: 0.3632, Accuracy: 0.9062\n",
+ "Epoch: 8, Batch: 150, Loss: 0.2642, Accuracy: 0.9219\n",
+ "Epoch: 8, Batch: 200, Loss: 0.3594, Accuracy: 0.9062\n",
+ "Epoch: 8, Batch: 250, Loss: 0.2832, Accuracy: 0.9219\n",
+ "Epoch: 8, Batch: 300, Loss: 0.2071, Accuracy: 0.9531\n",
+ "Epoch: 8, Batch: 350, Loss: 0.3322, Accuracy: 0.8906\n",
+ "Epoch: 8, Batch: 400, Loss: 0.1634, Accuracy: 0.9688\n",
+ "Epoch: 8, Batch: 450, Loss: 0.3029, Accuracy: 0.9062\n",
+ "Epoch: 8, Batch: 500, Loss: 0.2814, Accuracy: 0.9219\n",
+ "Epoch: 8, Batch: 550, Loss: 0.3699, Accuracy: 0.8906\n",
+ "Epoch: 8, Batch: 600, Loss: 0.3284, Accuracy: 0.9062\n",
+ "Epoch: 8, Batch: 650, Loss: 0.3867, Accuracy: 0.8750\n",
+ "Epoch: 8, Batch: 700, Loss: 0.1809, Accuracy: 0.9531\n",
+ "Epoch: 8, Batch: 750, Loss: 0.3607, Accuracy: 0.8906\n",
+ "Epoch: 9, Batch: 0, Loss: 0.2970, Accuracy: 0.9219\n",
+ "Epoch: 9, Batch: 50, Loss: 0.6045, Accuracy: 0.8125\n",
+ "Epoch: 9, Batch: 100, Loss: 0.4143, Accuracy: 0.8750\n",
+ "Epoch: 9, Batch: 150, Loss: 0.3316, Accuracy: 0.8906\n",
+ "Epoch: 9, Batch: 200, Loss: 0.2525, Accuracy: 0.9375\n",
+ "Epoch: 9, Batch: 250, Loss: 0.3518, Accuracy: 0.9062\n",
+ "Epoch: 9, Batch: 300, Loss: 0.3512, Accuracy: 0.8906\n",
+ "Epoch: 9, Batch: 350, Loss: 0.3825, Accuracy: 0.8750\n",
+ "Epoch: 9, Batch: 400, Loss: 0.2434, Accuracy: 0.9375\n",
+ "Epoch: 9, Batch: 450, Loss: 0.5630, Accuracy: 0.8281\n",
+ "Epoch: 9, Batch: 500, Loss: 0.2942, Accuracy: 0.9219\n",
+ "Epoch: 9, Batch: 550, Loss: 0.1786, Accuracy: 0.9531\n",
+ "Epoch: 9, Batch: 600, Loss: 0.2020, Accuracy: 0.9375\n",
+ "Epoch: 9, Batch: 650, Loss: 0.4735, Accuracy: 0.8594\n",
+ "Epoch: 9, Batch: 700, Loss: 0.3972, Accuracy: 0.8906\n",
+ "Epoch: 9, Batch: 750, Loss: 0.4706, Accuracy: 0.8594\n",
+ "Epoch: 10, Batch: 0, Loss: 0.3030, Accuracy: 0.9219\n",
+ "Epoch: 10, Batch: 50, Loss: 0.3658, Accuracy: 0.8906\n",
+ "Epoch: 10, Batch: 100, Loss: 0.3322, Accuracy: 0.8906\n",
+ "Epoch: 10, Batch: 150, Loss: 0.3684, Accuracy: 0.8906\n",
+ "Epoch: 10, Batch: 200, Loss: 0.4174, Accuracy: 0.8750\n",
+ "Epoch: 10, Batch: 250, Loss: 0.3050, Accuracy: 0.9062\n",
+ "Epoch: 10, Batch: 300, Loss: 0.3083, Accuracy: 0.9062\n",
+ "Epoch: 10, Batch: 350, Loss: 0.4319, Accuracy: 0.8750\n",
+ "Epoch: 10, Batch: 400, Loss: 0.6263, Accuracy: 0.7969\n",
+ "Epoch: 10, Batch: 450, Loss: 0.2850, Accuracy: 0.9219\n",
+ "Epoch: 10, Batch: 500, Loss: 0.3495, Accuracy: 0.8906\n",
+ "Epoch: 10, Batch: 550, Loss: 0.2538, Accuracy: 0.9375\n",
+ "Epoch: 10, Batch: 600, Loss: 0.3938, Accuracy: 0.8750\n",
+ "Epoch: 10, Batch: 650, Loss: 0.4767, Accuracy: 0.8438\n",
+ "Epoch: 10, Batch: 700, Loss: 0.2534, Accuracy: 0.9375\n",
+ "Epoch: 10, Batch: 750, Loss: 0.3790, Accuracy: 0.8906\n",
+ "Epoch: 11, Batch: 0, Loss: 0.4360, Accuracy: 0.8594\n",
+ "Epoch: 11, Batch: 50, Loss: 0.3722, Accuracy: 0.8906\n",
+ "Epoch: 11, Batch: 100, Loss: 0.4026, Accuracy: 0.8906\n",
+ "Epoch: 11, Batch: 150, Loss: 0.1270, Accuracy: 0.9688\n",
+ "Epoch: 11, Batch: 200, Loss: 0.3616, Accuracy: 0.8906\n",
+ "Epoch: 11, Batch: 250, Loss: 0.2308, Accuracy: 0.9375\n",
+ "Epoch: 11, Batch: 300, Loss: 0.4520, Accuracy: 0.8594\n",
+ "Epoch: 11, Batch: 350, Loss: 0.4075, Accuracy: 0.8750\n",
+ "Epoch: 11, Batch: 400, Loss: 0.3222, Accuracy: 0.9062\n",
+ "Epoch: 11, Batch: 450, Loss: 0.2663, Accuracy: 0.9219\n",
+ "Epoch: 11, Batch: 500, Loss: 0.1387, Accuracy: 0.9531\n",
+ "Epoch: 11, Batch: 550, Loss: 0.4702, Accuracy: 0.8594\n",
+ "Epoch: 11, Batch: 600, Loss: 0.4505, Accuracy: 0.8594\n",
+ "Epoch: 11, Batch: 650, Loss: 0.4182, Accuracy: 0.8750\n",
+ "Epoch: 11, Batch: 700, Loss: 0.3478, Accuracy: 0.8906\n",
+ "Epoch: 11, Batch: 750, Loss: 0.3621, Accuracy: 0.8906\n",
+ "Epoch: 12, Batch: 0, Loss: 0.2849, Accuracy: 0.9219\n",
+ "Epoch: 12, Batch: 50, Loss: 0.3110, Accuracy: 0.9062\n",
+ "Epoch: 12, Batch: 100, Loss: 0.3051, Accuracy: 0.9062\n",
+ "Epoch: 12, Batch: 150, Loss: 0.2737, Accuracy: 0.9219\n",
+ "Epoch: 12, Batch: 200, Loss: 0.3742, Accuracy: 0.8750\n",
+ "Epoch: 12, Batch: 250, Loss: 0.4212, Accuracy: 0.8594\n",
+ "Epoch: 12, Batch: 300, Loss: 0.2280, Accuracy: 0.9375\n",
+ "Epoch: 12, Batch: 350, Loss: 0.4784, Accuracy: 0.8594\n",
+ "Epoch: 12, Batch: 400, Loss: 0.2956, Accuracy: 0.9062\n",
+ "Epoch: 12, Batch: 450, Loss: 0.2589, Accuracy: 0.9219\n",
+ "Epoch: 12, Batch: 500, Loss: 0.4105, Accuracy: 0.8906\n",
+ "Epoch: 12, Batch: 550, Loss: 0.3742, Accuracy: 0.8750\n",
+ "Epoch: 12, Batch: 600, Loss: 0.4328, Accuracy: 0.8594\n",
+ "Epoch: 12, Batch: 650, Loss: 0.4948, Accuracy: 0.8438\n",
+ "Epoch: 12, Batch: 700, Loss: 0.3466, Accuracy: 0.8906\n",
+ "Epoch: 12, Batch: 750, Loss: 0.2288, Accuracy: 0.9375\n",
+ "Epoch: 13, Batch: 0, Loss: 0.2461, Accuracy: 0.9375\n",
+ "Epoch: 13, Batch: 50, Loss: 0.1805, Accuracy: 0.9531\n",
+ "Epoch: 13, Batch: 100, Loss: 0.2795, Accuracy: 0.9062\n",
+ "Epoch: 13, Batch: 150, Loss: 0.3512, Accuracy: 0.9062\n",
+ "Epoch: 13, Batch: 200, Loss: 0.5196, Accuracy: 0.8125\n",
+ "Epoch: 13, Batch: 250, Loss: 0.4389, Accuracy: 0.8594\n",
+ "Epoch: 13, Batch: 300, Loss: 0.3456, Accuracy: 0.8906\n",
+ "Epoch: 13, Batch: 350, Loss: 0.2473, Accuracy: 0.9375\n",
+ "Epoch: 13, Batch: 400, Loss: 0.3316, Accuracy: 0.9062\n",
+ "Epoch: 13, Batch: 450, Loss: 0.3315, Accuracy: 0.9062\n",
+ "Epoch: 13, Batch: 500, Loss: 0.3654, Accuracy: 0.8906\n",
+ "Epoch: 13, Batch: 550, Loss: 0.2489, Accuracy: 0.9375\n",
+ "Epoch: 13, Batch: 600, Loss: 0.3257, Accuracy: 0.9062\n",
+ "Epoch: 13, Batch: 650, Loss: 0.2387, Accuracy: 0.9375\n",
+ "Epoch: 13, Batch: 700, Loss: 0.5489, Accuracy: 0.8281\n",
+ "Epoch: 13, Batch: 750, Loss: 0.2914, Accuracy: 0.9219\n",
+ "Epoch: 14, Batch: 0, Loss: 0.4353, Accuracy: 0.8594\n",
+ "Epoch: 14, Batch: 50, Loss: 0.3106, Accuracy: 0.9062\n",
+ "Epoch: 14, Batch: 100, Loss: 0.2361, Accuracy: 0.9219\n",
+ "Epoch: 14, Batch: 150, Loss: 0.3523, Accuracy: 0.9062\n",
+ "Epoch: 14, Batch: 200, Loss: 0.3012, Accuracy: 0.9219\n",
+ "Epoch: 14, Batch: 250, Loss: 0.2116, Accuracy: 0.9375\n",
+ "Epoch: 14, Batch: 300, Loss: 0.3845, Accuracy: 0.8750\n",
+ "Epoch: 14, Batch: 350, Loss: 0.4540, Accuracy: 0.8594\n",
+ "Epoch: 14, Batch: 400, Loss: 0.2243, Accuracy: 0.9375\n",
+ "Epoch: 14, Batch: 450, Loss: 0.3854, Accuracy: 0.8750\n",
+ "Epoch: 14, Batch: 500, Loss: 0.2479, Accuracy: 0.9219\n",
+ "Epoch: 14, Batch: 550, Loss: 0.5586, Accuracy: 0.8281\n",
+ "Epoch: 14, Batch: 600, Loss: 0.3884, Accuracy: 0.8750\n",
+ "Epoch: 14, Batch: 650, Loss: 0.5463, Accuracy: 0.8281\n",
+ "Epoch: 14, Batch: 700, Loss: 0.3759, Accuracy: 0.8906\n",
+ "Epoch: 14, Batch: 750, Loss: 0.3011, Accuracy: 0.9062\n",
+ "Epoch: 15, Batch: 0, Loss: 0.6202, Accuracy: 0.8125\n",
+ "Epoch: 15, Batch: 50, Loss: 0.4044, Accuracy: 0.8906\n",
+ "Epoch: 15, Batch: 100, Loss: 0.3505, Accuracy: 0.8906\n",
+ "Epoch: 15, Batch: 150, Loss: 0.2088, Accuracy: 0.9531\n",
+ "Epoch: 15, Batch: 200, Loss: 0.1554, Accuracy: 0.9688\n",
+ "Epoch: 15, Batch: 250, Loss: 0.3466, Accuracy: 0.9062\n",
+ "Epoch: 15, Batch: 300, Loss: 0.1943, Accuracy: 0.9531\n",
+ "Epoch: 15, Batch: 350, Loss: 0.5770, Accuracy: 0.8281\n",
+ "Epoch: 15, Batch: 400, Loss: 0.2344, Accuracy: 0.9375\n",
+ "Epoch: 15, Batch: 450, Loss: 0.4119, Accuracy: 0.8750\n",
+ "Epoch: 15, Batch: 500, Loss: 0.3551, Accuracy: 0.9062\n",
+ "Epoch: 15, Batch: 550, Loss: 0.2234, Accuracy: 0.9375\n",
+ "Epoch: 15, Batch: 600, Loss: 0.2032, Accuracy: 0.9531\n",
+ "Epoch: 15, Batch: 650, Loss: 0.1824, Accuracy: 0.9531\n",
+ "Epoch: 15, Batch: 700, Loss: 0.2137, Accuracy: 0.9375\n",
+ "Epoch: 15, Batch: 750, Loss: 0.3519, Accuracy: 0.8750\n",
+ "Epoch: 16, Batch: 0, Loss: 0.4850, Accuracy: 0.8594\n",
+ "Epoch: 16, Batch: 50, Loss: 0.2885, Accuracy: 0.9219\n",
+ "Epoch: 16, Batch: 100, Loss: 0.2115, Accuracy: 0.9375\n",
+ "Epoch: 16, Batch: 150, Loss: 0.1637, Accuracy: 0.9531\n",
+ "Epoch: 16, Batch: 200, Loss: 0.3442, Accuracy: 0.8906\n",
+ "Epoch: 16, Batch: 250, Loss: 0.4721, Accuracy: 0.8594\n",
+ "Epoch: 16, Batch: 300, Loss: 0.2668, Accuracy: 0.9219\n",
+ "Epoch: 16, Batch: 350, Loss: 0.2947, Accuracy: 0.9062\n",
+ "Epoch: 16, Batch: 400, Loss: 0.4097, Accuracy: 0.8594\n",
+ "Epoch: 16, Batch: 450, Loss: 0.5181, Accuracy: 0.8438\n",
+ "Epoch: 16, Batch: 500, Loss: 0.2716, Accuracy: 0.9062\n",
+ "Epoch: 16, Batch: 550, Loss: 0.2996, Accuracy: 0.9062\n",
+ "Epoch: 16, Batch: 600, Loss: 0.1031, Accuracy: 0.9844\n",
+ "Epoch: 16, Batch: 650, Loss: 0.4878, Accuracy: 0.8594\n",
+ "Epoch: 16, Batch: 700, Loss: 0.2796, Accuracy: 0.9219\n",
+ "Epoch: 16, Batch: 750, Loss: 0.2015, Accuracy: 0.9531\n",
+ "Epoch: 17, Batch: 0, Loss: 0.2510, Accuracy: 0.9219\n",
+ "Epoch: 17, Batch: 50, Loss: 0.2685, Accuracy: 0.9219\n",
+ "Epoch: 17, Batch: 100, Loss: 0.2255, Accuracy: 0.9375\n",
+ "Epoch: 17, Batch: 150, Loss: 0.2705, Accuracy: 0.9219\n",
+ "Epoch: 17, Batch: 200, Loss: 0.3725, Accuracy: 0.8750\n",
+ "Epoch: 17, Batch: 250, Loss: 0.4283, Accuracy: 0.8750\n",
+ "Epoch: 17, Batch: 300, Loss: 0.3770, Accuracy: 0.8906\n",
+ "Epoch: 17, Batch: 350, Loss: 0.2558, Accuracy: 0.9375\n",
+ "Epoch: 17, Batch: 400, Loss: 0.2779, Accuracy: 0.9219\n",
+ "Epoch: 17, Batch: 450, Loss: 0.3149, Accuracy: 0.8906\n",
+ "Epoch: 17, Batch: 500, Loss: 0.1625, Accuracy: 0.9531\n",
+ "Epoch: 17, Batch: 550, Loss: 0.3953, Accuracy: 0.8594\n",
+ "Epoch: 17, Batch: 600, Loss: 0.3675, Accuracy: 0.8906\n",
+ "Epoch: 17, Batch: 650, Loss: 0.4696, Accuracy: 0.8594\n",
+ "Epoch: 17, Batch: 700, Loss: 0.3609, Accuracy: 0.8906\n",
+ "Epoch: 17, Batch: 750, Loss: 0.3258, Accuracy: 0.9219\n",
+ "Epoch: 18, Batch: 0, Loss: 0.3999, Accuracy: 0.8750\n",
+ "Epoch: 18, Batch: 50, Loss: 0.4544, Accuracy: 0.8906\n",
+ "Epoch: 18, Batch: 100, Loss: 0.4615, Accuracy: 0.8438\n",
+ "Epoch: 18, Batch: 150, Loss: 0.2437, Accuracy: 0.9219\n",
+ "Epoch: 18, Batch: 200, Loss: 0.2197, Accuracy: 0.9531\n",
+ "Epoch: 18, Batch: 250, Loss: 0.4701, Accuracy: 0.8594\n",
+ "Epoch: 18, Batch: 300, Loss: 0.3827, Accuracy: 0.8750\n",
+ "Epoch: 18, Batch: 350, Loss: 0.2714, Accuracy: 0.9219\n",
+ "Epoch: 18, Batch: 400, Loss: 0.2385, Accuracy: 0.9375\n",
+ "Epoch: 18, Batch: 450, Loss: 0.2662, Accuracy: 0.9062\n",
+ "Epoch: 18, Batch: 500, Loss: 0.5095, Accuracy: 0.8438\n",
+ "Epoch: 18, Batch: 550, Loss: 0.4597, Accuracy: 0.8750\n",
+ "Epoch: 18, Batch: 600, Loss: 0.2917, Accuracy: 0.9219\n",
+ "Epoch: 18, Batch: 650, Loss: 0.2518, Accuracy: 0.9219\n",
+ "Epoch: 18, Batch: 700, Loss: 0.2892, Accuracy: 0.9219\n",
+ "Epoch: 18, Batch: 750, Loss: 0.3718, Accuracy: 0.8906\n",
+ "Epoch: 19, Batch: 0, Loss: 0.4317, Accuracy: 0.8750\n",
+ "Epoch: 19, Batch: 50, Loss: 0.3516, Accuracy: 0.8906\n",
+ "Epoch: 19, Batch: 100, Loss: 0.4401, Accuracy: 0.8750\n",
+ "Epoch: 19, Batch: 150, Loss: 0.3609, Accuracy: 0.8906\n",
+ "Epoch: 19, Batch: 200, Loss: 0.3140, Accuracy: 0.9062\n",
+ "Epoch: 19, Batch: 250, Loss: 0.4060, Accuracy: 0.8750\n",
+ "Epoch: 19, Batch: 300, Loss: 0.2689, Accuracy: 0.9219\n",
+ "Epoch: 19, Batch: 350, Loss: 0.1900, Accuracy: 0.9375\n",
+ "Epoch: 19, Batch: 400, Loss: 0.3847, Accuracy: 0.8750\n",
+ "Epoch: 19, Batch: 450, Loss: 0.3713, Accuracy: 0.8906\n",
+ "Epoch: 19, Batch: 500, Loss: 0.2912, Accuracy: 0.9219\n",
+ "Epoch: 19, Batch: 550, Loss: 0.3616, Accuracy: 0.8906\n",
+ "Epoch: 19, Batch: 600, Loss: 0.3331, Accuracy: 0.9062\n",
+ "Epoch: 19, Batch: 650, Loss: 0.2643, Accuracy: 0.9219\n",
+ "Epoch: 19, Batch: 700, Loss: 0.3527, Accuracy: 0.9062\n",
+ "Epoch: 19, Batch: 750, Loss: 0.3679, Accuracy: 0.8750\n",
+ "Epoch: 20, Batch: 0, Loss: 0.4632, Accuracy: 0.8438\n",
+ "Epoch: 20, Batch: 50, Loss: 0.3168, Accuracy: 0.8906\n",
+ "Epoch: 20, Batch: 100, Loss: 0.3277, Accuracy: 0.9219\n",
+ "Epoch: 20, Batch: 150, Loss: 0.1503, Accuracy: 0.9688\n",
+ "Epoch: 20, Batch: 200, Loss: 0.3160, Accuracy: 0.9062\n",
+ "Epoch: 20, Batch: 250, Loss: 0.2083, Accuracy: 0.9531\n",
+ "Epoch: 20, Batch: 300, Loss: 0.4475, Accuracy: 0.8594\n",
+ "Epoch: 20, Batch: 350, Loss: 0.4463, Accuracy: 0.8594\n",
+ "Epoch: 20, Batch: 400, Loss: 0.1941, Accuracy: 0.9531\n",
+ "Epoch: 20, Batch: 450, Loss: 0.2940, Accuracy: 0.9219\n",
+ "Epoch: 20, Batch: 500, Loss: 0.4440, Accuracy: 0.8594\n",
+ "Epoch: 20, Batch: 550, Loss: 0.5096, Accuracy: 0.8594\n",
+ "Epoch: 20, Batch: 600, Loss: 0.2695, Accuracy: 0.9219\n",
+ "Epoch: 20, Batch: 650, Loss: 0.4495, Accuracy: 0.8594\n",
+ "Epoch: 20, Batch: 700, Loss: 0.3003, Accuracy: 0.9062\n",
+ "Epoch: 20, Batch: 750, Loss: 0.2347, Accuracy: 0.9375\n",
+ "Epoch: 21, Batch: 0, Loss: 0.3034, Accuracy: 0.9219\n",
+ "Epoch: 21, Batch: 50, Loss: 0.3877, Accuracy: 0.8750\n",
+ "Epoch: 21, Batch: 100, Loss: 0.4309, Accuracy: 0.8750\n",
+ "Epoch: 21, Batch: 150, Loss: 0.3924, Accuracy: 0.8750\n",
+ "Epoch: 21, Batch: 200, Loss: 0.4508, Accuracy: 0.8594\n",
+ "Epoch: 21, Batch: 250, Loss: 0.2740, Accuracy: 0.9219\n",
+ "Epoch: 21, Batch: 300, Loss: 0.4891, Accuracy: 0.8438\n",
+ "Epoch: 21, Batch: 350, Loss: 0.2856, Accuracy: 0.9219\n",
+ "Epoch: 21, Batch: 400, Loss: 0.0455, Accuracy: 1.0000\n",
+ "Epoch: 21, Batch: 450, Loss: 0.4910, Accuracy: 0.8438\n",
+ "Epoch: 21, Batch: 500, Loss: 0.2694, Accuracy: 0.9062\n",
+ "Epoch: 21, Batch: 550, Loss: 0.3082, Accuracy: 0.9062\n",
+ "Epoch: 21, Batch: 600, Loss: 0.2260, Accuracy: 0.9531\n",
+ "Epoch: 21, Batch: 650, Loss: 0.1631, Accuracy: 0.9531\n",
+ "Epoch: 21, Batch: 700, Loss: 0.3069, Accuracy: 0.9219\n",
+ "Epoch: 21, Batch: 750, Loss: 0.4678, Accuracy: 0.8438\n",
+ "Epoch: 22, Batch: 0, Loss: 0.4563, Accuracy: 0.8594\n",
+ "Epoch: 22, Batch: 50, Loss: 0.2691, Accuracy: 0.9375\n",
+ "Epoch: 22, Batch: 100, Loss: 0.4520, Accuracy: 0.8438\n",
+ "Epoch: 22, Batch: 150, Loss: 0.3972, Accuracy: 0.8750\n",
+ "Epoch: 22, Batch: 200, Loss: 0.3320, Accuracy: 0.8906\n",
+ "Epoch: 22, Batch: 250, Loss: 0.4119, Accuracy: 0.8906\n",
+ "Epoch: 22, Batch: 300, Loss: 0.2597, Accuracy: 0.9219\n",
+ "Epoch: 22, Batch: 350, Loss: 0.2030, Accuracy: 0.9531\n",
+ "Epoch: 22, Batch: 400, Loss: 0.2151, Accuracy: 0.9375\n",
+ "Epoch: 22, Batch: 450, Loss: 0.2871, Accuracy: 0.9219\n",
+ "Epoch: 22, Batch: 500, Loss: 0.2892, Accuracy: 0.9062\n",
+ "Epoch: 22, Batch: 550, Loss: 0.4443, Accuracy: 0.8594\n",
+ "Epoch: 22, Batch: 600, Loss: 0.2177, Accuracy: 0.9219\n",
+ "Epoch: 22, Batch: 650, Loss: 0.3408, Accuracy: 0.9062\n",
+ "Epoch: 22, Batch: 700, Loss: 0.4698, Accuracy: 0.8594\n",
+ "Epoch: 22, Batch: 750, Loss: 0.4157, Accuracy: 0.8594\n",
+ "Epoch: 23, Batch: 0, Loss: 0.3918, Accuracy: 0.8906\n",
+ "Epoch: 23, Batch: 50, Loss: 0.1003, Accuracy: 0.9844\n",
+ "Epoch: 23, Batch: 100, Loss: 0.3034, Accuracy: 0.9062\n",
+ "Epoch: 23, Batch: 150, Loss: 0.2433, Accuracy: 0.9219\n",
+ "Epoch: 23, Batch: 200, Loss: 0.4006, Accuracy: 0.8750\n",
+ "Epoch: 23, Batch: 250, Loss: 0.4214, Accuracy: 0.8594\n",
+ "Epoch: 23, Batch: 300, Loss: 0.4411, Accuracy: 0.8594\n",
+ "Epoch: 23, Batch: 350, Loss: 0.2651, Accuracy: 0.9219\n",
+ "Epoch: 23, Batch: 400, Loss: 0.2992, Accuracy: 0.9219\n",
+ "Epoch: 23, Batch: 450, Loss: 0.3926, Accuracy: 0.8750\n",
+ "Epoch: 23, Batch: 500, Loss: 0.1348, Accuracy: 0.9688\n",
+ "Epoch: 23, Batch: 550, Loss: 0.2408, Accuracy: 0.9375\n",
+ "Epoch: 23, Batch: 600, Loss: 0.2532, Accuracy: 0.9219\n",
+ "Epoch: 23, Batch: 650, Loss: 0.3790, Accuracy: 0.8906\n",
+ "Epoch: 23, Batch: 700, Loss: 0.3529, Accuracy: 0.8906\n",
+ "Epoch: 23, Batch: 750, Loss: 0.4256, Accuracy: 0.8750\n",
+ "Epoch: 24, Batch: 0, Loss: 0.5058, Accuracy: 0.8438\n",
+ "Epoch: 24, Batch: 50, Loss: 0.2562, Accuracy: 0.9375\n",
+ "Epoch: 24, Batch: 100, Loss: 0.1293, Accuracy: 0.9688\n",
+ "Epoch: 24, Batch: 150, Loss: 0.2718, Accuracy: 0.9219\n",
+ "Epoch: 24, Batch: 200, Loss: 0.1990, Accuracy: 0.9375\n",
+ "Epoch: 24, Batch: 250, Loss: 0.5353, Accuracy: 0.8281\n",
+ "Epoch: 24, Batch: 300, Loss: 0.4225, Accuracy: 0.8750\n",
+ "Epoch: 24, Batch: 350, Loss: 0.5562, Accuracy: 0.8281\n",
+ "Epoch: 24, Batch: 400, Loss: 0.4328, Accuracy: 0.8750\n",
+ "Epoch: 24, Batch: 450, Loss: 0.5232, Accuracy: 0.8438\n",
+ "Epoch: 24, Batch: 500, Loss: 0.4480, Accuracy: 0.8750\n",
+ "Epoch: 24, Batch: 550, Loss: 0.2664, Accuracy: 0.9219\n",
+ "Epoch: 24, Batch: 600, Loss: 0.1258, Accuracy: 0.9844\n",
+ "Epoch: 24, Batch: 650, Loss: 0.2008, Accuracy: 0.9375\n",
+ "Epoch: 24, Batch: 700, Loss: 0.3194, Accuracy: 0.9062\n",
+ "Epoch: 24, Batch: 750, Loss: 0.2823, Accuracy: 0.9219\n",
+ "Epoch: 25, Batch: 0, Loss: 0.3141, Accuracy: 0.9062\n",
+ "Epoch: 25, Batch: 50, Loss: 0.4506, Accuracy: 0.8594\n",
+ "Epoch: 25, Batch: 100, Loss: 0.3840, Accuracy: 0.8906\n",
+ "Epoch: 25, Batch: 150, Loss: 0.3327, Accuracy: 0.8906\n",
+ "Epoch: 25, Batch: 200, Loss: 0.2974, Accuracy: 0.9219\n",
+ "Epoch: 25, Batch: 250, Loss: 0.1352, Accuracy: 0.9688\n",
+ "Epoch: 25, Batch: 300, Loss: 0.2615, Accuracy: 0.9219\n",
+ "Epoch: 25, Batch: 350, Loss: 0.2592, Accuracy: 0.9219\n",
+ "Epoch: 25, Batch: 400, Loss: 0.1725, Accuracy: 0.9531\n",
+ "Epoch: 25, Batch: 450, Loss: 0.2702, Accuracy: 0.9375\n",
+ "Epoch: 25, Batch: 500, Loss: 0.2383, Accuracy: 0.9375\n",
+ "Epoch: 25, Batch: 550, Loss: 0.3933, Accuracy: 0.8750\n",
+ "Epoch: 25, Batch: 600, Loss: 0.3011, Accuracy: 0.9219\n",
+ "Epoch: 25, Batch: 650, Loss: 0.2424, Accuracy: 0.9375\n",
+ "Epoch: 25, Batch: 700, Loss: 0.3150, Accuracy: 0.9062\n",
+ "Epoch: 25, Batch: 750, Loss: 0.4006, Accuracy: 0.8750\n",
+ "Epoch: 26, Batch: 0, Loss: 0.1929, Accuracy: 0.9531\n",
+ "Epoch: 26, Batch: 50, Loss: 0.5716, Accuracy: 0.8125\n",
+ "Epoch: 26, Batch: 100, Loss: 0.2425, Accuracy: 0.9375\n",
+ "Epoch: 26, Batch: 150, Loss: 0.3950, Accuracy: 0.8750\n",
+ "Epoch: 26, Batch: 200, Loss: 0.3319, Accuracy: 0.9062\n",
+ "Epoch: 26, Batch: 250, Loss: 0.1654, Accuracy: 0.9688\n",
+ "Epoch: 26, Batch: 300, Loss: 0.5397, Accuracy: 0.8438\n",
+ "Epoch: 26, Batch: 350, Loss: 0.2591, Accuracy: 0.9219\n",
+ "Epoch: 26, Batch: 400, Loss: 0.2968, Accuracy: 0.9219\n",
+ "Epoch: 26, Batch: 450, Loss: 0.5532, Accuracy: 0.8125\n",
+ "Epoch: 26, Batch: 500, Loss: 0.1760, Accuracy: 0.9531\n",
+ "Epoch: 26, Batch: 550, Loss: 0.3264, Accuracy: 0.8906\n",
+ "Epoch: 26, Batch: 600, Loss: 0.2803, Accuracy: 0.9219\n",
+ "Epoch: 26, Batch: 650, Loss: 0.2472, Accuracy: 0.9375\n",
+ "Epoch: 26, Batch: 700, Loss: 0.2606, Accuracy: 0.9219\n",
+ "Epoch: 26, Batch: 750, Loss: 0.2745, Accuracy: 0.9219\n",
+ "Epoch: 27, Batch: 0, Loss: 0.3512, Accuracy: 0.8906\n",
+ "Epoch: 27, Batch: 50, Loss: 0.2941, Accuracy: 0.9062\n",
+ "Epoch: 27, Batch: 100, Loss: 0.2187, Accuracy: 0.9375\n",
+ "Epoch: 27, Batch: 150, Loss: 0.2879, Accuracy: 0.9219\n",
+ "Epoch: 27, Batch: 200, Loss: 0.2874, Accuracy: 0.9219\n",
+ "Epoch: 27, Batch: 250, Loss: 0.5574, Accuracy: 0.8125\n",
+ "Epoch: 27, Batch: 300, Loss: 0.4487, Accuracy: 0.8594\n",
+ "Epoch: 27, Batch: 350, Loss: 0.2447, Accuracy: 0.9375\n",
+ "Epoch: 27, Batch: 400, Loss: 0.1519, Accuracy: 0.9688\n",
+ "Epoch: 27, Batch: 450, Loss: 0.3609, Accuracy: 0.8906\n",
+ "Epoch: 27, Batch: 500, Loss: 0.2143, Accuracy: 0.9375\n",
+ "Epoch: 27, Batch: 550, Loss: 0.3586, Accuracy: 0.9062\n",
+ "Epoch: 27, Batch: 600, Loss: 0.1934, Accuracy: 0.9375\n",
+ "Epoch: 27, Batch: 650, Loss: 0.3593, Accuracy: 0.8906\n",
+ "Epoch: 27, Batch: 700, Loss: 0.3218, Accuracy: 0.9062\n",
+ "Epoch: 27, Batch: 750, Loss: 0.3630, Accuracy: 0.8906\n",
+ "Epoch: 28, Batch: 0, Loss: 0.3300, Accuracy: 0.8906\n",
+ "Epoch: 28, Batch: 50, Loss: 0.3491, Accuracy: 0.8906\n",
+ "Epoch: 28, Batch: 100, Loss: 0.2256, Accuracy: 0.9375\n",
+ "Epoch: 28, Batch: 150, Loss: 0.3707, Accuracy: 0.8750\n",
+ "Epoch: 28, Batch: 200, Loss: 0.4507, Accuracy: 0.8438\n",
+ "Epoch: 28, Batch: 250, Loss: 0.3541, Accuracy: 0.8906\n",
+ "Epoch: 28, Batch: 300, Loss: 0.3883, Accuracy: 0.8750\n",
+ "Epoch: 28, Batch: 350, Loss: 0.2863, Accuracy: 0.9219\n",
+ "Epoch: 28, Batch: 400, Loss: 0.2894, Accuracy: 0.9219\n",
+ "Epoch: 28, Batch: 450, Loss: 0.3787, Accuracy: 0.8906\n",
+ "Epoch: 28, Batch: 500, Loss: 0.3526, Accuracy: 0.9062\n",
+ "Epoch: 28, Batch: 550, Loss: 0.3710, Accuracy: 0.8906\n",
+ "Epoch: 28, Batch: 600, Loss: 0.4035, Accuracy: 0.8750\n",
+ "Epoch: 28, Batch: 650, Loss: 0.2845, Accuracy: 0.9062\n",
+ "Epoch: 28, Batch: 700, Loss: 0.3513, Accuracy: 0.8906\n",
+ "Epoch: 28, Batch: 750, Loss: 0.3650, Accuracy: 0.8906\n",
+ "Epoch: 29, Batch: 0, Loss: 0.4340, Accuracy: 0.8594\n",
+ "Epoch: 29, Batch: 50, Loss: 0.3820, Accuracy: 0.8906\n",
+ "Epoch: 29, Batch: 100, Loss: 0.4393, Accuracy: 0.8750\n",
+ "Epoch: 29, Batch: 150, Loss: 0.5519, Accuracy: 0.8281\n",
+ "Epoch: 29, Batch: 200, Loss: 0.2818, Accuracy: 0.9219\n",
+ "Epoch: 29, Batch: 250, Loss: 0.4767, Accuracy: 0.8438\n",
+ "Epoch: 29, Batch: 300, Loss: 0.2885, Accuracy: 0.9219\n",
+ "Epoch: 29, Batch: 350, Loss: 0.4896, Accuracy: 0.8594\n",
+ "Epoch: 29, Batch: 400, Loss: 0.3687, Accuracy: 0.8906\n",
+ "Epoch: 29, Batch: 450, Loss: 0.2621, Accuracy: 0.9219\n",
+ "Epoch: 29, Batch: 500, Loss: 0.3818, Accuracy: 0.8906\n",
+ "Epoch: 29, Batch: 550, Loss: 0.3212, Accuracy: 0.9062\n",
+ "Epoch: 29, Batch: 600, Loss: 0.3866, Accuracy: 0.8906\n",
+ "Epoch: 29, Batch: 650, Loss: 0.3533, Accuracy: 0.9062\n",
+ "Epoch: 29, Batch: 700, Loss: 0.3396, Accuracy: 0.9062\n",
+ "Epoch: 29, Batch: 750, Loss: 0.3327, Accuracy: 0.9219\n",
+ "Epoch: 30, Batch: 0, Loss: 0.1774, Accuracy: 0.9531\n",
+ "Epoch: 30, Batch: 50, Loss: 0.1823, Accuracy: 0.9531\n",
+ "Epoch: 30, Batch: 100, Loss: 0.2740, Accuracy: 0.9219\n",
+ "Epoch: 30, Batch: 150, Loss: 0.2063, Accuracy: 0.9531\n",
+ "Epoch: 30, Batch: 200, Loss: 0.3956, Accuracy: 0.8906\n",
+ "Epoch: 30, Batch: 250, Loss: 0.2873, Accuracy: 0.9062\n",
+ "Epoch: 30, Batch: 300, Loss: 0.3908, Accuracy: 0.8906\n",
+ "Epoch: 30, Batch: 350, Loss: 0.2038, Accuracy: 0.9531\n",
+ "Epoch: 30, Batch: 400, Loss: 0.2114, Accuracy: 0.9531\n",
+ "Epoch: 30, Batch: 450, Loss: 0.1781, Accuracy: 0.9531\n",
+ "Epoch: 30, Batch: 500, Loss: 0.5614, Accuracy: 0.8281\n",
+ "Epoch: 30, Batch: 550, Loss: 0.3801, Accuracy: 0.8906\n",
+ "Epoch: 30, Batch: 600, Loss: 0.2480, Accuracy: 0.9375\n",
+ "Epoch: 30, Batch: 650, Loss: 0.2774, Accuracy: 0.9219\n",
+ "Epoch: 30, Batch: 700, Loss: 0.4436, Accuracy: 0.8594\n",
+ "Epoch: 30, Batch: 750, Loss: 0.2464, Accuracy: 0.9375\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 5. Modell mentése"
+ ],
+ "metadata": {
+ "id": "LAii5r1LFyJ1"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "torch.save({\n",
+ " 'model_state_dict': model.state_dict(),\n",
+ " 'A': model.A,\n",
+ " 'B': model.B,\n",
+ " 'Z': model.Z\n",
+ "}, 'cenn_model.pth')"
+ ],
+ "metadata": {
+ "id": "yyG9r3whFxuQ"
+ },
+ "execution_count": 7,
+ "outputs": []
+ }
+ ]
+}
\ No newline at end of file
diff --git a/Project_CeNN_CNN.ipynb b/Project_CeNN_CNN.ipynb
new file mode 100644
index 0000000000000000000000000000000000000000..ff3d6f653762f1cca72404b1d7c43c03489906ab
--- /dev/null
+++ b/Project_CeNN_CNN.ipynb
@@ -0,0 +1,515 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 0,
+ "metadata": {
+ "colab": {
+ "provenance": [],
+ "gpuType": "T4"
+ },
+ "kernelspec": {
+ "name": "python3",
+ "display_name": "Python 3"
+ },
+ "language_info": {
+ "name": "python"
+ },
+ "accelerator": "GPU"
+ },
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 0. Importok"
+ ],
+ "metadata": {
+ "id": "Lil58l51RzsM"
+ }
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {
+ "id": "jc3M1FaFQNsp"
+ },
+ "outputs": [],
+ "source": [
+ "import torch\n",
+ "import torchvision\n",
+ "import torchvision.transforms as transforms\n",
+ "from torch.utils.data import DataLoader\n",
+ "import torch.nn as nn\n",
+ "from torchvision.models import resnet18"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [],
+ "metadata": {
+ "id": "SEMlMaM1R92Y"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "# 1. Adatelőkészítés\n",
+ "transform = transforms.Compose([\n",
+ " transforms.RandomHorizontalFlip(),\n",
+ " transforms.RandomCrop(32, padding=4),\n",
+ " transforms.ToTensor(),\n",
+ " transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n",
+ "])\n",
+ "\n",
+ "trainset = torchvision.datasets.CIFAR10(\n",
+ " root='./data',\n",
+ " train=True,\n",
+ " download=True,\n",
+ " transform=transform\n",
+ ")\n",
+ "\n",
+ "trainloader = DataLoader(trainset, batch_size=64, shuffle=True)"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "ybVCnWi4R-NS",
+ "outputId": "71d7e5f2-b80b-44a4-b368-1b16c6d70e65"
+ },
+ "execution_count": 2,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to ./data/cifar-10-python.tar.gz\n"
+ ]
+ },
+ {
+ "output_type": "stream",
+ "name": "stderr",
+ "text": [
+ "100%|██████████| 170M/170M [00:06<00:00, 28.1MB/s]\n"
+ ]
+ },
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Extracting ./data/cifar-10-python.tar.gz to ./data\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 2. Hibrid modell osztály"
+ ],
+ "metadata": {
+ "id": "MAF_1TgfSYHJ"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "class HybridCeNN_CNN(nn.Module):\n",
+ " def __init__(self, h=0.1, iter_num=10):\n",
+ " super().__init__()\n",
+ " self.h = h\n",
+ " self.iter_num = iter_num\n",
+ "\n",
+ " # CeNN rész\n",
+ " self.A = nn.Parameter(torch.randn(1, 3, 3, 3))\n",
+ " self.B = nn.Parameter(torch.randn(1, 3, 3, 3))\n",
+ " self.Z = nn.Parameter(torch.randn(1))\n",
+ "\n",
+ " # CNN rész (módosított ResNet)\n",
+ " self.cnn = resnet18(weights=None)\n",
+ " self.cnn.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)\n",
+ " self.cnn.maxpool = nn.Identity()\n",
+ " self.cnn.fc = nn.Linear(512, 10)\n",
+ "\n",
+ " def forward(self, U):\n",
+ " # CeNN előfeldolgozás\n",
+ " x = U\n",
+ " for _ in range(self.iter_num):\n",
+ " y = torch.clamp(x, -1, 1) # Egyszerűsített aktiváció\n",
+ " fwd = torch.nn.functional.conv2d(U, self.B, padding=1) + self.Z\n",
+ " bwd = torch.nn.functional.conv2d(y, self.A, padding=1)\n",
+ " x = x + self.h * (-x + bwd + fwd)\n",
+ "\n",
+ " # CNN osztályozás\n",
+ " return self.cnn(x)"
+ ],
+ "metadata": {
+ "id": "iMuSg8YNSXxi"
+ },
+ "execution_count": 3,
+ "outputs": []
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 3. Modell és optimalizáló"
+ ],
+ "metadata": {
+ "id": "AgTBXU42T8pe"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
+ "print(f\"Használt eszköz: {device}\")"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "GWjVItvj_IJF",
+ "outputId": "1088e026-9db3-4638-ce19-55c2477190e3"
+ },
+ "execution_count": 5,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Használt eszköz: cuda\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "model = HybridCeNN_CNN().to(device)\n",
+ "optimizer = torch.optim.Adam([\n",
+ " {'params': model.A},\n",
+ " {'params': model.B},\n",
+ " {'params': model.Z},\n",
+ " {'params': model.cnn.parameters()}\n",
+ "], lr=0.001)\n",
+ "\n",
+ "criterion = nn.CrossEntropyLoss()"
+ ],
+ "metadata": {
+ "id": "H0hd36GPT8dA"
+ },
+ "execution_count": 6,
+ "outputs": []
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 4. Tanítási ciklus"
+ ],
+ "metadata": {
+ "id": "eZPvhqalUNGG"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "for epoch in range(30):\n",
+ " model.train()\n",
+ " for batch_idx, (data, targets) in enumerate(trainloader):\n",
+ " data, targets = data.to(device), targets.to(device)\n",
+ "\n",
+ " optimizer.zero_grad()\n",
+ " outputs = model(data)\n",
+ " loss = criterion(outputs, targets)\n",
+ " loss.backward()\n",
+ " optimizer.step()\n",
+ "\n",
+ " if batch_idx % 100 == 0:\n",
+ " print(f'Epoch: {epoch+1}, Batch: {batch_idx}, Loss: {loss.item():.4f}')"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "uuVVLhpBUSr9",
+ "outputId": "3c219607-0135-42cb-fec6-8000744c33a3"
+ },
+ "execution_count": 7,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Epoch: 1, Batch: 0, Loss: 2.3162\n",
+ "Epoch: 1, Batch: 100, Loss: 1.8707\n",
+ "Epoch: 1, Batch: 200, Loss: 1.6191\n",
+ "Epoch: 1, Batch: 300, Loss: 1.4384\n",
+ "Epoch: 1, Batch: 400, Loss: 1.5578\n",
+ "Epoch: 1, Batch: 500, Loss: 1.3173\n",
+ "Epoch: 1, Batch: 600, Loss: 1.4486\n",
+ "Epoch: 1, Batch: 700, Loss: 0.9764\n",
+ "Epoch: 2, Batch: 0, Loss: 1.1860\n",
+ "Epoch: 2, Batch: 100, Loss: 0.9394\n",
+ "Epoch: 2, Batch: 200, Loss: 0.9662\n",
+ "Epoch: 2, Batch: 300, Loss: 0.8775\n",
+ "Epoch: 2, Batch: 400, Loss: 0.8830\n",
+ "Epoch: 2, Batch: 500, Loss: 1.0138\n",
+ "Epoch: 2, Batch: 600, Loss: 0.8941\n",
+ "Epoch: 2, Batch: 700, Loss: 0.6419\n",
+ "Epoch: 3, Batch: 0, Loss: 0.7703\n",
+ "Epoch: 3, Batch: 100, Loss: 0.6211\n",
+ "Epoch: 3, Batch: 200, Loss: 0.8289\n",
+ "Epoch: 3, Batch: 300, Loss: 0.9120\n",
+ "Epoch: 3, Batch: 400, Loss: 0.8239\n",
+ "Epoch: 3, Batch: 500, Loss: 0.8767\n",
+ "Epoch: 3, Batch: 600, Loss: 0.6739\n",
+ "Epoch: 3, Batch: 700, Loss: 0.4059\n",
+ "Epoch: 4, Batch: 0, Loss: 0.6486\n",
+ "Epoch: 4, Batch: 100, Loss: 0.6528\n",
+ "Epoch: 4, Batch: 200, Loss: 0.5334\n",
+ "Epoch: 4, Batch: 300, Loss: 0.5487\n",
+ "Epoch: 4, Batch: 400, Loss: 0.5126\n",
+ "Epoch: 4, Batch: 500, Loss: 0.6614\n",
+ "Epoch: 4, Batch: 600, Loss: 0.5521\n",
+ "Epoch: 4, Batch: 700, Loss: 0.5489\n",
+ "Epoch: 5, Batch: 0, Loss: 0.5379\n",
+ "Epoch: 5, Batch: 100, Loss: 0.4864\n",
+ "Epoch: 5, Batch: 200, Loss: 0.5838\n",
+ "Epoch: 5, Batch: 300, Loss: 0.7733\n",
+ "Epoch: 5, Batch: 400, Loss: 0.3621\n",
+ "Epoch: 5, Batch: 500, Loss: 0.6030\n",
+ "Epoch: 5, Batch: 600, Loss: 0.4617\n",
+ "Epoch: 5, Batch: 700, Loss: 0.4785\n",
+ "Epoch: 6, Batch: 0, Loss: 0.6555\n",
+ "Epoch: 6, Batch: 100, Loss: 0.6239\n",
+ "Epoch: 6, Batch: 200, Loss: 0.4051\n",
+ "Epoch: 6, Batch: 300, Loss: 0.4033\n",
+ "Epoch: 6, Batch: 400, Loss: 0.3421\n",
+ "Epoch: 6, Batch: 500, Loss: 0.4544\n",
+ "Epoch: 6, Batch: 600, Loss: 0.4110\n",
+ "Epoch: 6, Batch: 700, Loss: 0.4839\n",
+ "Epoch: 7, Batch: 0, Loss: 0.4884\n",
+ "Epoch: 7, Batch: 100, Loss: 0.4682\n",
+ "Epoch: 7, Batch: 200, Loss: 0.4639\n",
+ "Epoch: 7, Batch: 300, Loss: 0.5159\n",
+ "Epoch: 7, Batch: 400, Loss: 0.2979\n",
+ "Epoch: 7, Batch: 500, Loss: 0.4135\n",
+ "Epoch: 7, Batch: 600, Loss: 0.4771\n",
+ "Epoch: 7, Batch: 700, Loss: 0.3821\n",
+ "Epoch: 8, Batch: 0, Loss: 0.3357\n",
+ "Epoch: 8, Batch: 100, Loss: 0.4008\n",
+ "Epoch: 8, Batch: 200, Loss: 0.3050\n",
+ "Epoch: 8, Batch: 300, Loss: 0.2159\n",
+ "Epoch: 8, Batch: 400, Loss: 0.4112\n",
+ "Epoch: 8, Batch: 500, Loss: 0.4043\n",
+ "Epoch: 8, Batch: 600, Loss: 0.2823\n",
+ "Epoch: 8, Batch: 700, Loss: 0.3153\n",
+ "Epoch: 9, Batch: 0, Loss: 0.3715\n",
+ "Epoch: 9, Batch: 100, Loss: 0.4164\n",
+ "Epoch: 9, Batch: 200, Loss: 0.2845\n",
+ "Epoch: 9, Batch: 300, Loss: 0.3807\n",
+ "Epoch: 9, Batch: 400, Loss: 0.5043\n",
+ "Epoch: 9, Batch: 500, Loss: 0.2632\n",
+ "Epoch: 9, Batch: 600, Loss: 0.3511\n",
+ "Epoch: 9, Batch: 700, Loss: 0.3831\n",
+ "Epoch: 10, Batch: 0, Loss: 0.3084\n",
+ "Epoch: 10, Batch: 100, Loss: 0.1943\n",
+ "Epoch: 10, Batch: 200, Loss: 0.2923\n",
+ "Epoch: 10, Batch: 300, Loss: 0.1154\n",
+ "Epoch: 10, Batch: 400, Loss: 0.1852\n",
+ "Epoch: 10, Batch: 500, Loss: 0.1287\n",
+ "Epoch: 10, Batch: 600, Loss: 0.3440\n",
+ "Epoch: 10, Batch: 700, Loss: 0.2018\n",
+ "Epoch: 11, Batch: 0, Loss: 0.1535\n",
+ "Epoch: 11, Batch: 100, Loss: 0.2198\n",
+ "Epoch: 11, Batch: 200, Loss: 0.3176\n",
+ "Epoch: 11, Batch: 300, Loss: 0.1781\n",
+ "Epoch: 11, Batch: 400, Loss: 0.2402\n",
+ "Epoch: 11, Batch: 500, Loss: 0.3165\n",
+ "Epoch: 11, Batch: 600, Loss: 0.1474\n",
+ "Epoch: 11, Batch: 700, Loss: 0.3127\n",
+ "Epoch: 12, Batch: 0, Loss: 0.2592\n",
+ "Epoch: 12, Batch: 100, Loss: 0.1840\n",
+ "Epoch: 12, Batch: 200, Loss: 0.3848\n",
+ "Epoch: 12, Batch: 300, Loss: 0.2294\n",
+ "Epoch: 12, Batch: 400, Loss: 0.2759\n",
+ "Epoch: 12, Batch: 500, Loss: 0.2546\n",
+ "Epoch: 12, Batch: 600, Loss: 0.2649\n",
+ "Epoch: 12, Batch: 700, Loss: 0.4619\n",
+ "Epoch: 13, Batch: 0, Loss: 0.1814\n",
+ "Epoch: 13, Batch: 100, Loss: 0.3335\n",
+ "Epoch: 13, Batch: 200, Loss: 0.1933\n",
+ "Epoch: 13, Batch: 300, Loss: 0.2852\n",
+ "Epoch: 13, Batch: 400, Loss: 0.2613\n",
+ "Epoch: 13, Batch: 500, Loss: 0.2625\n",
+ "Epoch: 13, Batch: 600, Loss: 0.1588\n",
+ "Epoch: 13, Batch: 700, Loss: 0.2675\n",
+ "Epoch: 14, Batch: 0, Loss: 0.1518\n",
+ "Epoch: 14, Batch: 100, Loss: 0.1809\n",
+ "Epoch: 14, Batch: 200, Loss: 0.1823\n",
+ "Epoch: 14, Batch: 300, Loss: 0.2635\n",
+ "Epoch: 14, Batch: 400, Loss: 0.4669\n",
+ "Epoch: 14, Batch: 500, Loss: 0.1534\n",
+ "Epoch: 14, Batch: 600, Loss: 0.0958\n",
+ "Epoch: 14, Batch: 700, Loss: 0.2517\n",
+ "Epoch: 15, Batch: 0, Loss: 0.1085\n",
+ "Epoch: 15, Batch: 100, Loss: 0.1393\n",
+ "Epoch: 15, Batch: 200, Loss: 0.1517\n",
+ "Epoch: 15, Batch: 300, Loss: 0.1721\n",
+ "Epoch: 15, Batch: 400, Loss: 0.0785\n",
+ "Epoch: 15, Batch: 500, Loss: 0.2807\n",
+ "Epoch: 15, Batch: 600, Loss: 0.0837\n",
+ "Epoch: 15, Batch: 700, Loss: 0.1786\n",
+ "Epoch: 16, Batch: 0, Loss: 0.0274\n",
+ "Epoch: 16, Batch: 100, Loss: 0.1039\n",
+ "Epoch: 16, Batch: 200, Loss: 0.2527\n",
+ "Epoch: 16, Batch: 300, Loss: 0.1042\n",
+ "Epoch: 16, Batch: 400, Loss: 0.1901\n",
+ "Epoch: 16, Batch: 500, Loss: 0.2128\n",
+ "Epoch: 16, Batch: 600, Loss: 0.1787\n",
+ "Epoch: 16, Batch: 700, Loss: 0.1444\n",
+ "Epoch: 17, Batch: 0, Loss: 0.0999\n",
+ "Epoch: 17, Batch: 100, Loss: 0.2429\n",
+ "Epoch: 17, Batch: 200, Loss: 0.0779\n",
+ "Epoch: 17, Batch: 300, Loss: 0.1294\n",
+ "Epoch: 17, Batch: 400, Loss: 0.1542\n",
+ "Epoch: 17, Batch: 500, Loss: 0.2249\n",
+ "Epoch: 17, Batch: 600, Loss: 0.1600\n",
+ "Epoch: 17, Batch: 700, Loss: 0.3770\n",
+ "Epoch: 18, Batch: 0, Loss: 0.1151\n",
+ "Epoch: 18, Batch: 100, Loss: 0.0851\n",
+ "Epoch: 18, Batch: 200, Loss: 0.1027\n",
+ "Epoch: 18, Batch: 300, Loss: 0.1325\n",
+ "Epoch: 18, Batch: 400, Loss: 0.1680\n",
+ "Epoch: 18, Batch: 500, Loss: 0.2423\n",
+ "Epoch: 18, Batch: 600, Loss: 0.1666\n",
+ "Epoch: 18, Batch: 700, Loss: 0.2186\n",
+ "Epoch: 19, Batch: 0, Loss: 0.1842\n",
+ "Epoch: 19, Batch: 100, Loss: 0.1486\n",
+ "Epoch: 19, Batch: 200, Loss: 0.1381\n",
+ "Epoch: 19, Batch: 300, Loss: 0.0629\n",
+ "Epoch: 19, Batch: 400, Loss: 0.0521\n",
+ "Epoch: 19, Batch: 500, Loss: 0.0506\n",
+ "Epoch: 19, Batch: 600, Loss: 0.2298\n",
+ "Epoch: 19, Batch: 700, Loss: 0.0978\n",
+ "Epoch: 20, Batch: 0, Loss: 0.1976\n",
+ "Epoch: 20, Batch: 100, Loss: 0.0710\n",
+ "Epoch: 20, Batch: 200, Loss: 0.1511\n",
+ "Epoch: 20, Batch: 300, Loss: 0.1046\n",
+ "Epoch: 20, Batch: 400, Loss: 0.2077\n",
+ "Epoch: 20, Batch: 500, Loss: 0.1461\n",
+ "Epoch: 20, Batch: 600, Loss: 0.1101\n",
+ "Epoch: 20, Batch: 700, Loss: 0.2041\n",
+ "Epoch: 21, Batch: 0, Loss: 0.0737\n",
+ "Epoch: 21, Batch: 100, Loss: 0.1191\n",
+ "Epoch: 21, Batch: 200, Loss: 0.1232\n",
+ "Epoch: 21, Batch: 300, Loss: 0.1898\n",
+ "Epoch: 21, Batch: 400, Loss: 0.1074\n",
+ "Epoch: 21, Batch: 500, Loss: 0.0719\n",
+ "Epoch: 21, Batch: 600, Loss: 0.0778\n",
+ "Epoch: 21, Batch: 700, Loss: 0.2157\n",
+ "Epoch: 22, Batch: 0, Loss: 0.1120\n",
+ "Epoch: 22, Batch: 100, Loss: 0.0175\n",
+ "Epoch: 22, Batch: 200, Loss: 0.1115\n",
+ "Epoch: 22, Batch: 300, Loss: 0.1243\n",
+ "Epoch: 22, Batch: 400, Loss: 0.0250\n",
+ "Epoch: 22, Batch: 500, Loss: 0.0922\n",
+ "Epoch: 22, Batch: 600, Loss: 0.2028\n",
+ "Epoch: 22, Batch: 700, Loss: 0.0796\n",
+ "Epoch: 23, Batch: 0, Loss: 0.1341\n",
+ "Epoch: 23, Batch: 100, Loss: 0.0497\n",
+ "Epoch: 23, Batch: 200, Loss: 0.2450\n",
+ "Epoch: 23, Batch: 300, Loss: 0.0892\n",
+ "Epoch: 23, Batch: 400, Loss: 0.0652\n",
+ "Epoch: 23, Batch: 500, Loss: 0.2296\n",
+ "Epoch: 23, Batch: 600, Loss: 0.0666\n",
+ "Epoch: 23, Batch: 700, Loss: 0.1161\n",
+ "Epoch: 24, Batch: 0, Loss: 0.1039\n",
+ "Epoch: 24, Batch: 100, Loss: 0.1519\n",
+ "Epoch: 24, Batch: 200, Loss: 0.1032\n",
+ "Epoch: 24, Batch: 300, Loss: 0.0345\n",
+ "Epoch: 24, Batch: 400, Loss: 0.1313\n",
+ "Epoch: 24, Batch: 500, Loss: 0.1651\n",
+ "Epoch: 24, Batch: 600, Loss: 0.1424\n",
+ "Epoch: 24, Batch: 700, Loss: 0.1687\n",
+ "Epoch: 25, Batch: 0, Loss: 0.0630\n",
+ "Epoch: 25, Batch: 100, Loss: 0.0532\n",
+ "Epoch: 25, Batch: 200, Loss: 0.1020\n",
+ "Epoch: 25, Batch: 300, Loss: 0.1182\n",
+ "Epoch: 25, Batch: 400, Loss: 0.0716\n",
+ "Epoch: 25, Batch: 500, Loss: 0.0289\n",
+ "Epoch: 25, Batch: 600, Loss: 0.1287\n",
+ "Epoch: 25, Batch: 700, Loss: 0.1860\n",
+ "Epoch: 26, Batch: 0, Loss: 0.0946\n",
+ "Epoch: 26, Batch: 100, Loss: 0.0702\n",
+ "Epoch: 26, Batch: 200, Loss: 0.1193\n",
+ "Epoch: 26, Batch: 300, Loss: 0.0780\n",
+ "Epoch: 26, Batch: 400, Loss: 0.0149\n",
+ "Epoch: 26, Batch: 500, Loss: 0.0486\n",
+ "Epoch: 26, Batch: 600, Loss: 0.0951\n",
+ "Epoch: 26, Batch: 700, Loss: 0.1454\n",
+ "Epoch: 27, Batch: 0, Loss: 0.0472\n",
+ "Epoch: 27, Batch: 100, Loss: 0.0720\n",
+ "Epoch: 27, Batch: 200, Loss: 0.0521\n",
+ "Epoch: 27, Batch: 300, Loss: 0.0654\n",
+ "Epoch: 27, Batch: 400, Loss: 0.0928\n",
+ "Epoch: 27, Batch: 500, Loss: 0.0249\n",
+ "Epoch: 27, Batch: 600, Loss: 0.1215\n",
+ "Epoch: 27, Batch: 700, Loss: 0.0526\n",
+ "Epoch: 28, Batch: 0, Loss: 0.0769\n",
+ "Epoch: 28, Batch: 100, Loss: 0.1254\n",
+ "Epoch: 28, Batch: 200, Loss: 0.1225\n",
+ "Epoch: 28, Batch: 300, Loss: 0.0684\n",
+ "Epoch: 28, Batch: 400, Loss: 0.1082\n",
+ "Epoch: 28, Batch: 500, Loss: 0.0621\n",
+ "Epoch: 28, Batch: 600, Loss: 0.0434\n",
+ "Epoch: 28, Batch: 700, Loss: 0.0655\n",
+ "Epoch: 29, Batch: 0, Loss: 0.0395\n",
+ "Epoch: 29, Batch: 100, Loss: 0.0380\n",
+ "Epoch: 29, Batch: 200, Loss: 0.0291\n",
+ "Epoch: 29, Batch: 300, Loss: 0.0632\n",
+ "Epoch: 29, Batch: 400, Loss: 0.0638\n",
+ "Epoch: 29, Batch: 500, Loss: 0.1050\n",
+ "Epoch: 29, Batch: 600, Loss: 0.0911\n",
+ "Epoch: 29, Batch: 700, Loss: 0.0645\n",
+ "Epoch: 30, Batch: 0, Loss: 0.0548\n",
+ "Epoch: 30, Batch: 100, Loss: 0.0504\n",
+ "Epoch: 30, Batch: 200, Loss: 0.0496\n",
+ "Epoch: 30, Batch: 300, Loss: 0.0294\n",
+ "Epoch: 30, Batch: 400, Loss: 0.0203\n",
+ "Epoch: 30, Batch: 500, Loss: 0.0696\n",
+ "Epoch: 30, Batch: 600, Loss: 0.1455\n",
+ "Epoch: 30, Batch: 700, Loss: 0.0719\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# Mentés"
+ ],
+ "metadata": {
+ "id": "fJOZXRqVUoqo"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "torch.save({\n",
+ " 'model_state_dict': model.state_dict(),\n",
+ " 'A': model.A,\n",
+ " 'B': model.B,\n",
+ " 'Z': model.Z,\n",
+ " 'cnn_conv1': model.cnn.conv1.weight,\n",
+ " 'cnn_fc': model.cnn.fc.state_dict(),\n",
+ " 'h': model.h,\n",
+ " 'iter_num': model.iter_num\n",
+ "}, 'hybrid_model.pth')"
+ ],
+ "metadata": {
+ "id": "WObN5P2rXhJg"
+ },
+ "execution_count": 9,
+ "outputs": []
+ }
+ ]
+}
\ No newline at end of file
diff --git a/Project_Test.ipynb b/Project_Test.ipynb
new file mode 100644
index 0000000000000000000000000000000000000000..7877a2114e31719948c0dac4770cabbce74c817f
--- /dev/null
+++ b/Project_Test.ipynb
@@ -0,0 +1,598 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 0,
+ "metadata": {
+ "colab": {
+ "provenance": []
+ },
+ "kernelspec": {
+ "name": "python3",
+ "display_name": "Python 3"
+ },
+ "language_info": {
+ "name": "python"
+ }
+ },
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 0. Importok, device"
+ ],
+ "metadata": {
+ "id": "tDLgxyVKaDl_"
+ }
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 29,
+ "metadata": {
+ "id": "6l13ZBfQZLuf"
+ },
+ "outputs": [],
+ "source": [
+ "import torch\n",
+ "import torchvision\n",
+ "import torchvision.transforms as transforms\n",
+ "from torch.utils.data import DataLoader\n",
+ "import torch.nn as nn\n",
+ "from torchvision.models import resnet18"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "# GPU ellenőrzés\n",
+ "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
+ "print(f\"Használt eszköz: {device}\")"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "icTgUWEAZV6b",
+ "outputId": "7ddc398d-1396-4e2f-ae0f-7775dcd256eb"
+ },
+ "execution_count": 30,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Használt eszköz: cpu\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 1. Dataset"
+ ],
+ "metadata": {
+ "id": "rnDrSHuyaJ3I"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "transform = transforms.Compose([\n",
+ " transforms.ToTensor(),\n",
+ " transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n",
+ "])\n",
+ "\n",
+ "testset = torchvision.datasets.CIFAR10(\n",
+ " root='./data',\n",
+ " train=False,\n",
+ " download=True,\n",
+ " transform=transform\n",
+ ")\n",
+ "\n",
+ "testloader = DataLoader(\n",
+ " testset,\n",
+ " batch_size=64,\n",
+ " shuffle=False\n",
+ ")"
+ ],
+ "metadata": {
+ "id": "7PXRQQNGaRXF"
+ },
+ "execution_count": 31,
+ "outputs": []
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 2. Model definitions"
+ ],
+ "metadata": {
+ "id": "MdqLUOlQb6Z3"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "# CeNN modell\n",
+ "class CeNN(nn.Module):\n",
+ " def __init__(self, h=0.1, iter_num=20):\n",
+ " super().__init__()\n",
+ " self.h = h\n",
+ " self.iter_num = iter_num\n",
+ " self.alpha = 0.01\n",
+ " self.A = nn.Parameter(torch.randn(1, 3, 3, 3))\n",
+ " self.B = nn.Parameter(torch.randn(1, 3, 3, 3))\n",
+ " self.Z = nn.Parameter(torch.randn(1))\n",
+ "\n",
+ " def forward(self, U):\n",
+ " x = U\n",
+ " for _ in range(self.iter_num):\n",
+ " y = torch.minimum(x, 1 + self.alpha * x)\n",
+ " y = torch.maximum(y, -1 + self.alpha * y)\n",
+ " fwd = torch.nn.functional.conv2d(U, self.B, padding=1) + self.Z\n",
+ " bwd = torch.nn.functional.conv2d(y, self.A, padding=1)\n",
+ " x = x + self.h * (-x + bwd + fwd)\n",
+ " out = torch.minimum(x, 1 + self.alpha * x)\n",
+ " out = torch.maximum(out, -1 + self.alpha * out)\n",
+ " return out"
+ ],
+ "metadata": {
+ "id": "zgm3MJOMqlIC"
+ },
+ "execution_count": 4,
+ "outputs": []
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "# CNN modell\n",
+ "class CIFAR10_ResNet(nn.Module):\n",
+ " def __init__(self):\n",
+ " super().__init__()\n",
+ " self.model = resnet18()\n",
+ " self.model.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)\n",
+ " self.model.maxpool = nn.Identity()\n",
+ " self.model.fc = nn.Linear(512, 10)\n",
+ "\n",
+ " def forward(self, x):\n",
+ " return self.model(x)"
+ ],
+ "metadata": {
+ "id": "1Muj5ySpqk73"
+ },
+ "execution_count": 20,
+ "outputs": []
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "# Hibrid modell\n",
+ "class HybridCeNN_CNN(nn.Module):\n",
+ " def __init__(self, h=0.1, iter_num=10):\n",
+ " super().__init__()\n",
+ " self.h = h\n",
+ " self.iter_num = iter_num\n",
+ "\n",
+ " # CeNN rész\n",
+ " self.A = nn.Parameter(torch.randn(1, 3, 3, 3))\n",
+ " self.B = nn.Parameter(torch.randn(1, 3, 3, 3))\n",
+ " self.Z = nn.Parameter(torch.randn(1))\n",
+ "\n",
+ " # CNN rész (módosított ResNet)\n",
+ " self.cnn = resnet18(weights=None)\n",
+ " self.cnn.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)\n",
+ " self.cnn.maxpool = nn.Identity()\n",
+ " self.cnn.fc = nn.Linear(512, 10)\n",
+ "\n",
+ " def forward(self, U):\n",
+ " # CeNN előfeldolgozás\n",
+ " x = U\n",
+ " for _ in range(self.iter_num):\n",
+ " y = torch.clamp(x, -1, 1) # Egyszerűsített aktiváció\n",
+ " fwd = torch.nn.functional.conv2d(U, self.B, padding=1) + self.Z\n",
+ " bwd = torch.nn.functional.conv2d(y, self.A, padding=1)\n",
+ " x = x + self.h * (-x + bwd + fwd)\n",
+ "\n",
+ " # CNN osztályozás\n",
+ " return self.cnn(x)"
+ ],
+ "metadata": {
+ "id": "r-3L0H3Obzcm"
+ },
+ "execution_count": 32,
+ "outputs": []
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 3. Modellek betöltése"
+ ],
+ "metadata": {
+ "id": "-W8u5L-tc4Ot"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "# CeNN betöltése\n",
+ "cenn_model = CeNN().to(device)\n",
+ "checkpoint_cenn = torch.load('cenn_model.pth', map_location=device)\n",
+ "cenn_model.load_state_dict(checkpoint_cenn['model_state_dict'])\n",
+ "cenn_model.A.data = checkpoint_cenn['A']\n",
+ "cenn_model.B.data = checkpoint_cenn['B']\n",
+ "cenn_model.Z.data = checkpoint_cenn['Z']\n",
+ "cenn_model.eval()"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "k9EGUbpNqsu1",
+ "outputId": "64996546-d8bf-40ea-ab05-81b49da6338e"
+ },
+ "execution_count": 21,
+ "outputs": [
+ {
+ "output_type": "execute_result",
+ "data": {
+ "text/plain": [
+ "CeNN()"
+ ]
+ },
+ "metadata": {},
+ "execution_count": 21
+ }
+ ]
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "# CNN betöltése\n",
+ "cnn_model = CIFAR10_ResNet().to(device)\n",
+ "checkpoint_cnn = torch.load('cnn_model.pth', map_location=device)\n",
+ "cnn_model.load_state_dict(checkpoint_cnn['model_state_dict'])\n",
+ "cnn_model.model.conv1.weight = checkpoint_cnn['conv1_weights']\n",
+ "cnn_model.model.fc.weight = checkpoint_cnn['fc_weights']\n",
+ "cnn_model.model.fc.bias = checkpoint_cnn['fc_bias']\n",
+ "cnn_model.eval()"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "S9-ZoV2Zc7ih",
+ "outputId": "42c4c6ab-e083-4767-bd65-e0572106a9de"
+ },
+ "execution_count": 22,
+ "outputs": [
+ {
+ "output_type": "execute_result",
+ "data": {
+ "text/plain": [
+ "CIFAR10_ResNet(\n",
+ " (model): ResNet(\n",
+ " (conv1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (maxpool): Identity()\n",
+ " (layer1): Sequential(\n",
+ " (0): BasicBlock(\n",
+ " (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " (1): BasicBlock(\n",
+ " (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (layer2): Sequential(\n",
+ " (0): BasicBlock(\n",
+ " (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (downsample): Sequential(\n",
+ " (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
+ " (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (1): BasicBlock(\n",
+ " (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (layer3): Sequential(\n",
+ " (0): BasicBlock(\n",
+ " (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (downsample): Sequential(\n",
+ " (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
+ " (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (1): BasicBlock(\n",
+ " (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (layer4): Sequential(\n",
+ " (0): BasicBlock(\n",
+ " (conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (downsample): Sequential(\n",
+ " (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
+ " (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (1): BasicBlock(\n",
+ " (conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))\n",
+ " (fc): Linear(in_features=512, out_features=10, bias=True)\n",
+ " )\n",
+ ")"
+ ]
+ },
+ "metadata": {},
+ "execution_count": 22
+ }
+ ]
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "# Hibrid modell betöltése\n",
+ "hybrid_model = HybridCeNN_CNN().to(device)\n",
+ "checkpoint_h = torch.load('hybrid_model.pth', map_location=device)\n",
+ "hybrid_model.load_state_dict(checkpoint_h['model_state_dict'])\n",
+ "hybrid_model.A.data = checkpoint_h['A']\n",
+ "hybrid_model.B.data = checkpoint_h['B']\n",
+ "hybrid_model.Z.data = checkpoint_h['Z']\n",
+ "hybrid_model.cnn.conv1.weight = checkpoint_h['cnn_conv1']\n",
+ "hybrid_model.cnn.fc.load_state_dict(checkpoint_h['cnn_fc'])\n",
+ "hybrid_model.eval()"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "oukhJeGVqsT8",
+ "outputId": "42d1af54-c5d6-4c12-e1b7-a06e41056d98"
+ },
+ "execution_count": 33,
+ "outputs": [
+ {
+ "output_type": "execute_result",
+ "data": {
+ "text/plain": [
+ "HybridCeNN_CNN(\n",
+ " (cnn): ResNet(\n",
+ " (conv1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (maxpool): Identity()\n",
+ " (layer1): Sequential(\n",
+ " (0): BasicBlock(\n",
+ " (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " (1): BasicBlock(\n",
+ " (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (layer2): Sequential(\n",
+ " (0): BasicBlock(\n",
+ " (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (downsample): Sequential(\n",
+ " (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
+ " (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (1): BasicBlock(\n",
+ " (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (layer3): Sequential(\n",
+ " (0): BasicBlock(\n",
+ " (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (downsample): Sequential(\n",
+ " (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
+ " (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (1): BasicBlock(\n",
+ " (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (layer4): Sequential(\n",
+ " (0): BasicBlock(\n",
+ " (conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (downsample): Sequential(\n",
+ " (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
+ " (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (1): BasicBlock(\n",
+ " (conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " (relu): ReLU(inplace=True)\n",
+ " (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
+ " (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
+ " )\n",
+ " )\n",
+ " (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))\n",
+ " (fc): Linear(in_features=512, out_features=10, bias=True)\n",
+ " )\n",
+ ")"
+ ]
+ },
+ "metadata": {},
+ "execution_count": 33
+ }
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "source": [
+ "# 4. Kiértékelés\n",
+ "\n"
+ ],
+ "metadata": {
+ "id": "1AjDZUsWf63_"
+ }
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "def evaluate_model(model, model_name):\n",
+ " correct = 0\n",
+ " total = 0\n",
+ " with torch.no_grad():\n",
+ " for images, labels in testloader:\n",
+ " images, labels = images.to(device), labels.to(device)\n",
+ " outputs = model(images)\n",
+ "\n",
+ " # CeNN speciális értékelés\n",
+ " if isinstance(model, CeNN):\n",
+ " expected = torch.ones_like(outputs)\n",
+ " expected[labels == 0] = -1\n",
+ " correct_dist = torch.mean((outputs - expected)**2, [1,2,3])\n",
+ " incorrect_dist = torch.mean((outputs - (-expected))**2, [1,2,3])\n",
+ " correct += (correct_dist < incorrect_dist).sum().item()\n",
+ "\n",
+ " # CNN/Hibrid standard értékelés\n",
+ " else:\n",
+ " _, predicted = torch.max(outputs.data, 1)\n",
+ " correct += (predicted == labels).sum().item()\n",
+ "\n",
+ " total += labels.size(0)\n",
+ "\n",
+ " print(f\"{model_name} pontossága: {100 * correct / total:.2f}%\")"
+ ],
+ "metadata": {
+ "id": "GyRfDVwugAd3"
+ },
+ "execution_count": 23,
+ "outputs": []
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "evaluate_model(cenn_model, \"CeNN\")"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "Qsbc5sXjq_Op",
+ "outputId": "84d1f18e-f5c3-44a1-8306-05b2bff06915"
+ },
+ "execution_count": 10,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "CeNN pontossága: 90.02%\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "evaluate_model(cnn_model, \"CNN\")"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "PviP-1Y-rAMW",
+ "outputId": "c3f456a1-c434-4fd2-faca-814487815524"
+ },
+ "execution_count": 24,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "CNN pontossága: 92.40%\n"
+ ]
+ }
+ ]
+ },
+ {
+ "cell_type": "code",
+ "source": [
+ "evaluate_model(hybrid_model, \"Hibrid modell\")"
+ ],
+ "metadata": {
+ "colab": {
+ "base_uri": "https://localhost:8080/"
+ },
+ "id": "CJTPURkcq_p9",
+ "outputId": "9a53ec7f-3d78-45e4-d8bb-30906f11633a"
+ },
+ "execution_count": 34,
+ "outputs": [
+ {
+ "output_type": "stream",
+ "name": "stdout",
+ "text": [
+ "Hibrid modell pontossága: 91.45%\n"
+ ]
+ }
+ ]
+ }
+ ]
+}
\ No newline at end of file
diff --git a/cenn_model.pth b/cenn_model.pth
new file mode 100644
index 0000000000000000000000000000000000000000..08622191c943f0013ea6bbc5714263894ffe191d
Binary files /dev/null and b/cenn_model.pth differ
diff --git a/cnn-project b/cnn-project
new file mode 160000
index 0000000000000000000000000000000000000000..aeaa420605938f77f59aba8231cd39a09ac9e9a0
--- /dev/null
+++ b/cnn-project
@@ -0,0 +1 @@
+Subproject commit aeaa420605938f77f59aba8231cd39a09ac9e9a0
diff --git a/cnn_model.pth b/cnn_model.pth
new file mode 100644
index 0000000000000000000000000000000000000000..352b21cb2c8d02d12c5740931b97364efafce4c4
Binary files /dev/null and b/cnn_model.pth differ
diff --git a/hybrid_model.pth b/hybrid_model.pth
new file mode 100644
index 0000000000000000000000000000000000000000..9ac070cf831e5bf129fab4af33cd5caca5d8a2c4
Binary files /dev/null and b/hybrid_model.pth differ