{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "4f403af3",
"metadata": {},
"outputs": [],
"source": [
"#Source: https://medium.com/dataseries/convolutional-autoencoder-in-pytorch-on-mnist-dataset-d65145c132ac"
]
},
{
"cell_type": "code",
"execution_count": 46,
"id": "add961d3",
"metadata": {},
"outputs": [],
"source": [
"import matplotlib.pyplot as plt # plotting library\n",
"from sklearn.model_selection import train_test_split\n",
"import numpy as np # this module is useful to work with numerical arrays\n",
"import pandas as pd \n",
"import random \n",
"import os\n",
"import torch\n",
"import torchvision\n",
"from torchvision import transforms, datasets\n",
"from torch.utils.data import DataLoader,random_split\n",
"from torch import nn\n",
"import torch.nn.functional as F\n",
"import torch.optim as optim"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "7f5313b5",
"metadata": {},
"outputs": [],
"source": [
"def find_candidate_images(images_path):\n",
" \"\"\"\n",
" Finds all candidate images in the given folder and its sub-folders.\n",
"\n",
" Returns:\n",
" images: a list of absolute paths to the discovered images.\n",
" \"\"\"\n",
" images = []\n",
" for root, dirs, files in os.walk(images_path):\n",
" for name in files:\n",
" file_path = os.path.abspath(os.path.join(root, name))\n",
" if ((os.path.splitext(name)[1]).lower() in ['.jpg','.png','.jpeg']):\n",
" images.append(file_path)\n",
" return images"
]
},
{
"cell_type": "code",
"execution_count": 49,
"id": "1e7f0096",
"metadata": {},
"outputs": [],
"source": [
"class MyDataset(torch.utils.data.Dataset):\n",
" def __init__(self, img_list, augmentations):\n",
" super(MyDataset, self).__init__()\n",
" self.img_list = img_list\n",
" self.augmentations = augmentations\n",
"\n",
" def __len__(self):\n",
" return len(self.img_list)\n",
"\n",
" def __getitem__(self, idx):\n",
" img = self.img_list[idx]\n",
" return self.augmentations(img)"
]
},
{
"cell_type": "code",
"execution_count": 51,
"id": "f846b86c",
"metadata": {},
"outputs": [],
"source": [
"images = find_candidate_images('../SD_sample_f_m_pt2')"
]
},
{
"cell_type": "code",
"execution_count": 43,
"id": "da000292",
"metadata": {},
"outputs": [],
"source": [
"transform = transforms.Compose([\n",
"transforms.ToTensor(),\n",
"])"
]
},
{
"cell_type": "code",
"execution_count": 55,
"id": "d8f46911",
"metadata": {},
"outputs": [],
"source": [
"data = MyDataset(images, transform)\n",
"dataset_iterator = DataLoader(data, batch_size=1)"
]
},
{
"cell_type": "code",
"execution_count": 56,
"id": "05504c87",
"metadata": {},
"outputs": [
{
"ename": "TypeError",
"evalue": "pic should be PIL Image or ndarray. Got <class 'str'>",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mTypeError\u001b[0m Traceback (most recent call last)",
"Input \u001b[0;32mIn [56]\u001b[0m, in \u001b[0;36m<cell line: 1>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0m train_images, test_images \u001b[38;5;241m=\u001b[39m \u001b[43mtrain_test_split\u001b[49m\u001b[43m(\u001b[49m\u001b[43mdata\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtest_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m0.33\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mrandom_state\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m42\u001b[39;49m\u001b[43m)\u001b[49m\n\u001b[1;32m 2\u001b[0m \u001b[38;5;28mprint\u001b[39m(\u001b[38;5;28mlen\u001b[39m(train_images))\n\u001b[1;32m 3\u001b[0m \u001b[38;5;28mprint\u001b[39m(\u001b[38;5;28mlen\u001b[39m(test_images))\n",
"File \u001b[0;32m~/miniconda3/envs/stablediffusion/lib/python3.9/site-packages/sklearn/model_selection/_split.py:2471\u001b[0m, in \u001b[0;36mtrain_test_split\u001b[0;34m(test_size, train_size, random_state, shuffle, stratify, *arrays)\u001b[0m\n\u001b[1;32m 2467\u001b[0m cv \u001b[38;5;241m=\u001b[39m CVClass(test_size\u001b[38;5;241m=\u001b[39mn_test, train_size\u001b[38;5;241m=\u001b[39mn_train, random_state\u001b[38;5;241m=\u001b[39mrandom_state)\n\u001b[1;32m 2469\u001b[0m train, test \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mnext\u001b[39m(cv\u001b[38;5;241m.\u001b[39msplit(X\u001b[38;5;241m=\u001b[39marrays[\u001b[38;5;241m0\u001b[39m], y\u001b[38;5;241m=\u001b[39mstratify))\n\u001b[0;32m-> 2471\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mlist\u001b[39;49m\u001b[43m(\u001b[49m\n\u001b[1;32m 2472\u001b[0m \u001b[43m \u001b[49m\u001b[43mchain\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfrom_iterable\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 2473\u001b[0m \u001b[43m \u001b[49m\u001b[43m(\u001b[49m\u001b[43m_safe_indexing\u001b[49m\u001b[43m(\u001b[49m\u001b[43ma\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtrain\u001b[49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43m_safe_indexing\u001b[49m\u001b[43m(\u001b[49m\u001b[43ma\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtest\u001b[49m\u001b[43m)\u001b[49m\u001b[43m)\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mfor\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43ma\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;129;43;01min\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43marrays\u001b[49m\n\u001b[1;32m 2474\u001b[0m \u001b[43m \u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 2475\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n",
"File \u001b[0;32m~/miniconda3/envs/stablediffusion/lib/python3.9/site-packages/sklearn/model_selection/_split.py:2473\u001b[0m, in \u001b[0;36m<genexpr>\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m 2467\u001b[0m cv \u001b[38;5;241m=\u001b[39m CVClass(test_size\u001b[38;5;241m=\u001b[39mn_test, train_size\u001b[38;5;241m=\u001b[39mn_train, random_state\u001b[38;5;241m=\u001b[39mrandom_state)\n\u001b[1;32m 2469\u001b[0m train, test \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mnext\u001b[39m(cv\u001b[38;5;241m.\u001b[39msplit(X\u001b[38;5;241m=\u001b[39marrays[\u001b[38;5;241m0\u001b[39m], y\u001b[38;5;241m=\u001b[39mstratify))\n\u001b[1;32m 2471\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mlist\u001b[39m(\n\u001b[1;32m 2472\u001b[0m chain\u001b[38;5;241m.\u001b[39mfrom_iterable(\n\u001b[0;32m-> 2473\u001b[0m (\u001b[43m_safe_indexing\u001b[49m\u001b[43m(\u001b[49m\u001b[43ma\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtrain\u001b[49m\u001b[43m)\u001b[49m, _safe_indexing(a, test)) \u001b[38;5;28;01mfor\u001b[39;00m a \u001b[38;5;129;01min\u001b[39;00m arrays\n\u001b[1;32m 2474\u001b[0m )\n\u001b[1;32m 2475\u001b[0m )\n",
"File \u001b[0;32m~/miniconda3/envs/stablediffusion/lib/python3.9/site-packages/sklearn/utils/__init__.py:363\u001b[0m, in \u001b[0;36m_safe_indexing\u001b[0;34m(X, indices, axis)\u001b[0m\n\u001b[1;32m 361\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m _array_indexing(X, indices, indices_dtype, axis\u001b[38;5;241m=\u001b[39maxis)\n\u001b[1;32m 362\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 363\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43m_list_indexing\u001b[49m\u001b[43m(\u001b[49m\u001b[43mX\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mindices\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mindices_dtype\u001b[49m\u001b[43m)\u001b[49m\n",
"File \u001b[0;32m~/miniconda3/envs/stablediffusion/lib/python3.9/site-packages/sklearn/utils/__init__.py:217\u001b[0m, in \u001b[0;36m_list_indexing\u001b[0;34m(X, key, key_dtype)\u001b[0m\n\u001b[1;32m 215\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mlist\u001b[39m(compress(X, key))\n\u001b[1;32m 216\u001b[0m \u001b[38;5;66;03m# key is a integer array-like of key\u001b[39;00m\n\u001b[0;32m--> 217\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m [X[idx] \u001b[38;5;28;01mfor\u001b[39;00m idx \u001b[38;5;129;01min\u001b[39;00m key]\n",
"File \u001b[0;32m~/miniconda3/envs/stablediffusion/lib/python3.9/site-packages/sklearn/utils/__init__.py:217\u001b[0m, in \u001b[0;36m<listcomp>\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m 215\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mlist\u001b[39m(compress(X, key))\n\u001b[1;32m 216\u001b[0m \u001b[38;5;66;03m# key is a integer array-like of key\u001b[39;00m\n\u001b[0;32m--> 217\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m [\u001b[43mX\u001b[49m\u001b[43m[\u001b[49m\u001b[43midx\u001b[49m\u001b[43m]\u001b[49m \u001b[38;5;28;01mfor\u001b[39;00m idx \u001b[38;5;129;01min\u001b[39;00m key]\n",
"Input \u001b[0;32mIn [49]\u001b[0m, in \u001b[0;36mMyDataset.__getitem__\u001b[0;34m(self, idx)\u001b[0m\n\u001b[1;32m 10\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21m__getitem__\u001b[39m(\u001b[38;5;28mself\u001b[39m, idx):\n\u001b[1;32m 11\u001b[0m img \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mimg_list[idx]\n\u001b[0;32m---> 12\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43maugmentations\u001b[49m\u001b[43m(\u001b[49m\u001b[43mimg\u001b[49m\u001b[43m)\u001b[49m\n",
"File \u001b[0;32m~/miniconda3/envs/stablediffusion/lib/python3.9/site-packages/torchvision/transforms/transforms.py:95\u001b[0m, in \u001b[0;36mCompose.__call__\u001b[0;34m(self, img)\u001b[0m\n\u001b[1;32m 93\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21m__call__\u001b[39m(\u001b[38;5;28mself\u001b[39m, img):\n\u001b[1;32m 94\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m t \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtransforms:\n\u001b[0;32m---> 95\u001b[0m img \u001b[38;5;241m=\u001b[39m \u001b[43mt\u001b[49m\u001b[43m(\u001b[49m\u001b[43mimg\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 96\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m img\n",
"File \u001b[0;32m~/miniconda3/envs/stablediffusion/lib/python3.9/site-packages/torchvision/transforms/transforms.py:135\u001b[0m, in \u001b[0;36mToTensor.__call__\u001b[0;34m(self, pic)\u001b[0m\n\u001b[1;32m 127\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21m__call__\u001b[39m(\u001b[38;5;28mself\u001b[39m, pic):\n\u001b[1;32m 128\u001b[0m \u001b[38;5;124;03m\"\"\"\u001b[39;00m\n\u001b[1;32m 129\u001b[0m \u001b[38;5;124;03m Args:\u001b[39;00m\n\u001b[1;32m 130\u001b[0m \u001b[38;5;124;03m pic (PIL Image or numpy.ndarray): Image to be converted to tensor.\u001b[39;00m\n\u001b[0;32m (...)\u001b[0m\n\u001b[1;32m 133\u001b[0m \u001b[38;5;124;03m Tensor: Converted image.\u001b[39;00m\n\u001b[1;32m 134\u001b[0m \u001b[38;5;124;03m \"\"\"\u001b[39;00m\n\u001b[0;32m--> 135\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mF\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mto_tensor\u001b[49m\u001b[43m(\u001b[49m\u001b[43mpic\u001b[49m\u001b[43m)\u001b[49m\n",
"File \u001b[0;32m~/miniconda3/envs/stablediffusion/lib/python3.9/site-packages/torchvision/transforms/functional.py:137\u001b[0m, in \u001b[0;36mto_tensor\u001b[0;34m(pic)\u001b[0m\n\u001b[1;32m 135\u001b[0m _log_api_usage_once(to_tensor)\n\u001b[1;32m 136\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (F_pil\u001b[38;5;241m.\u001b[39m_is_pil_image(pic) \u001b[38;5;129;01mor\u001b[39;00m _is_numpy(pic)):\n\u001b[0;32m--> 137\u001b[0m \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mTypeError\u001b[39;00m(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mpic should be PIL Image or ndarray. Got \u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mtype\u001b[39m(pic)\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m 139\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m _is_numpy(pic) \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m _is_numpy_image(pic):\n\u001b[1;32m 140\u001b[0m \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mValueError\u001b[39;00m(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mpic should be 2/3 dimensional. Got \u001b[39m\u001b[38;5;132;01m{\u001b[39;00mpic\u001b[38;5;241m.\u001b[39mndim\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m dimensions.\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n",
"\u001b[0;31mTypeError\u001b[0m: pic should be PIL Image or ndarray. Got <class 'str'>"
]
}
],
"source": [
"train_images, test_images = train_test_split(data, test_size=0.33, random_state=42)\n",
"print(len(train_images))\n",
"print(len(test_images))"
]
},
{
"cell_type": "code",
"execution_count": 16,
"id": "669f82ab",
"metadata": {},
"outputs": [],
"source": [
"m=len(train_images)"
]
},
{
"cell_type": "code",
"execution_count": 23,
"id": "e962953c",
"metadata": {},
"outputs": [],
"source": [
"train_data, val_data = random_split(train_images, [int(m-m*0.2), int(m*0.2)])\n",
"test_dataset = test_images"
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "16a8e2a1",
"metadata": {},
"outputs": [],
"source": [
"train_loader = torch.utils.data.DataLoader(train_data, batch_size=batch_size)\n",
"valid_loader = torch.utils.data.DataLoader(val_data, batch_size=batch_size)\n",
"test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size,shuffle=True)"
]
},
{
"cell_type": "code",
"execution_count": 25,
"id": "07403239",
"metadata": {},
"outputs": [],
"source": [
"class Encoder(nn.Module):\n",
" \n",
" def __init__(self, encoded_space_dim,fc2_input_dim):\n",
" super().__init__()\n",
" \n",
" ### Convolutional section\n",
" self.encoder_cnn = nn.Sequential(\n",
" nn.Conv2d(1, 8, 3, stride=2, padding=1),\n",
" nn.ReLU(True),\n",
" nn.Conv2d(8, 16, 3, stride=2, padding=1),\n",
" nn.BatchNorm2d(16),\n",
" nn.ReLU(True),\n",
" nn.Conv2d(16, 32, 3, stride=2, padding=0),\n",
" nn.ReLU(True)\n",
" )\n",
" \n",
" ### Flatten layer\n",
" self.flatten = nn.Flatten(start_dim=1)\n",
"### Linear section\n",
" self.encoder_lin = nn.Sequential(\n",
" nn.Linear(3 * 3 * 32, 128),\n",
" nn.ReLU(True),\n",
" nn.Linear(128, encoded_space_dim)\n",
" )\n",
" \n",
" def forward(self, x):\n",
" x = self.encoder_cnn(x)\n",
" x = self.flatten(x)\n",
" x = self.encoder_lin(x)\n",
" return x\n",
"class Decoder(nn.Module):\n",
" \n",
" def __init__(self, encoded_space_dim,fc2_input_dim):\n",
" super().__init__()\n",
" self.decoder_lin = nn.Sequential(\n",
" nn.Linear(encoded_space_dim, 128),\n",
" nn.ReLU(True),\n",
" nn.Linear(128, 3 * 3 * 32),\n",
" nn.ReLU(True)\n",
" )\n",
"\n",
" self.unflatten = nn.Unflatten(dim=1, \n",
" unflattened_size=(32, 3, 3))\n",
"\n",
" self.decoder_conv = nn.Sequential(\n",
" nn.ConvTranspose2d(32, 16, 3, \n",
" stride=2, output_padding=0),\n",
" nn.BatchNorm2d(16),\n",
" nn.ReLU(True),\n",
" nn.ConvTranspose2d(16, 8, 3, stride=2, \n",
" padding=1, output_padding=1),\n",
" nn.BatchNorm2d(8),\n",
" nn.ReLU(True),\n",
" nn.ConvTranspose2d(8, 1, 3, stride=2, \n",
" padding=1, output_padding=1)\n",
" )\n",
" \n",
" def forward(self, x):\n",
" x = self.decoder_lin(x)\n",
" x = self.unflatten(x)\n",
" x = self.decoder_conv(x)\n",
" x = torch.sigmoid(x)\n",
" return x"
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "fedfd708",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Selected device: cuda\n"
]
},
{
"data": {
"text/plain": [
"Decoder(\n",
" (decoder_lin): Sequential(\n",
" (0): Linear(in_features=4, out_features=128, bias=True)\n",
" (1): ReLU(inplace=True)\n",
" (2): Linear(in_features=128, out_features=288, bias=True)\n",
" (3): ReLU(inplace=True)\n",
" )\n",
" (unflatten): Unflatten(dim=1, unflattened_size=(32, 3, 3))\n",
" (decoder_conv): Sequential(\n",
" (0): ConvTranspose2d(32, 16, kernel_size=(3, 3), stride=(2, 2))\n",
" (1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
" (2): ReLU(inplace=True)\n",
" (3): ConvTranspose2d(16, 8, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))\n",
" (4): BatchNorm2d(8, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
" (5): ReLU(inplace=True)\n",
" (6): ConvTranspose2d(8, 1, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))\n",
" )\n",
")"
]
},
"execution_count": 26,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"### Define the loss function\n",
"loss_fn = torch.nn.MSELoss()\n",
"\n",
"### Define an optimizer (both for the encoder and the decoder!)\n",
"lr= 0.001\n",
"\n",
"### Set the random seed for reproducible results\n",
"torch.manual_seed(0)\n",
"\n",
"### Initialize the two networks\n",
"d = 4\n",
"\n",
"#model = Autoencoder(encoded_space_dim=encoded_space_dim)\n",
"encoder = Encoder(encoded_space_dim=d,fc2_input_dim=128)\n",
"decoder = Decoder(encoded_space_dim=d,fc2_input_dim=128)\n",
"params_to_optimize = [\n",
" {'params': encoder.parameters()},\n",
" {'params': decoder.parameters()}\n",
"]\n",
"\n",
"optim = torch.optim.Adam(params_to_optimize, lr=lr, weight_decay=1e-05)\n",
"\n",
"# Check if the GPU is available\n",
"device = torch.device(\"cuda\") if torch.cuda.is_available() else torch.device(\"cpu\")\n",
"print(f'Selected device: {device}')\n",
"\n",
"# Move both the encoder and the decoder to the selected device\n",
"encoder.to(device)\n",
"decoder.to(device)"
]
},
{
"cell_type": "code",
"execution_count": 33,
"id": "bae32de2",
"metadata": {},
"outputs": [],
"source": [
"### Training function\n",
"def train_epoch(encoder, decoder, device, dataloader, loss_fn, optimizer):\n",
" # Set train mode for both the encoder and the decoder\n",
" encoder.train()\n",
" decoder.train()\n",
" train_loss = []\n",
" # Iterate the dataloader (we do not need the label values, this is unsupervised learning)\n",
" for image_batch, _ in dataloader: # with \"_\" we just ignore the labels (the second element of the dataloader tuple)\n",
" # Move tensor to the proper device\n",
" image_batch = image_batch.to(device)\n",
" # Encode data\n",
" encoded_data = encoder(image_batch)\n",
" # Decode data\n",
" decoded_data = decoder(encoded_data)\n",
" # Evaluate loss\n",
" loss = loss_fn(decoded_data, image_batch)\n",
" # Backward pass\n",
" optimizer.zero_grad()\n",
" loss.backward()\n",
" optimizer.step()\n",
" # Print batch loss\n",
" print('\\t partial train loss (single batch): %f' % (loss.data))\n",
" train_loss.append(loss.detach().cpu().numpy())\n",
"\n",
" return np.mean(train_loss)"
]
},
{
"cell_type": "code",
"execution_count": 28,
"id": "ff2ec5fd",
"metadata": {},
"outputs": [],
"source": [
"### Testing function\n",
"def test_epoch(encoder, decoder, device, dataloader, loss_fn):\n",
" # Set evaluation mode for encoder and decoder\n",
" encoder.eval()\n",
" decoder.eval()\n",
" with torch.no_grad(): # No need to track the gradients\n",
" # Define the lists to store the outputs for each batch\n",
" conc_out = []\n",
" conc_label = []\n",
" for image_batch, _ in dataloader:\n",
" # Move tensor to the proper device\n",
" image_batch = image_batch.to(device)\n",
" # Encode data\n",
" encoded_data = encoder(image_batch)\n",
" # Decode data\n",
" decoded_data = decoder(encoded_data)\n",
" # Append the network output and the original image to the lists\n",
" conc_out.append(decoded_data.cpu())\n",
" conc_label.append(image_batch.cpu())\n",
" # Create a single tensor with all the values in the lists\n",
" conc_out = torch.cat(conc_out)\n",
" conc_label = torch.cat(conc_label) \n",
" # Evaluate global loss\n",
" val_loss = loss_fn(conc_out, conc_label)\n",
" return val_loss.data"
]
},
{
"cell_type": "code",
"execution_count": 29,
"id": "592ab5f1",
"metadata": {},
"outputs": [],
"source": [
"def plot_ae_outputs(encoder,decoder,n=10):\n",
" plt.figure(figsize=(16,4.5))\n",
" targets = test_dataset.targets.numpy()\n",
" t_idx = {i:np.where(targets==i)[0][0] for i in range(n)}\n",
" for i in range(n):\n",
" ax = plt.subplot(2,n,i+1)\n",
" img = test_dataset[t_idx[i]][0].unsqueeze(0).to(device)\n",
" encoder.eval()\n",
" decoder.eval()\n",
" with torch.no_grad():\n",
" rec_img = decoder(encoder(img))\n",
" plt.imshow(img.cpu().squeeze().numpy(), cmap='gist_gray')\n",
" ax.get_xaxis().set_visible(False)\n",
" ax.get_yaxis().set_visible(False) \n",
" if i == n//2:\n",
" ax.set_title('Original images')\n",
" ax = plt.subplot(2, n, i + 1 + n)\n",
" plt.imshow(rec_img.cpu().squeeze().numpy(), cmap='gist_gray') \n",
" ax.get_xaxis().set_visible(False)\n",
" ax.get_yaxis().set_visible(False) \n",
" if i == n//2:\n",
" ax.set_title('Reconstructed images')\n",
" plt.show() "
]
},
{
"cell_type": "code",
"execution_count": 34,
"id": "5f8b646b",
"metadata": {},
"outputs": [
{
"ename": "ValueError",
"evalue": "too many values to unpack (expected 2)",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mValueError\u001b[0m Traceback (most recent call last)",
"Input \u001b[0;32mIn [34]\u001b[0m, in \u001b[0;36m<cell line: 3>\u001b[0;34m()\u001b[0m\n\u001b[1;32m 2\u001b[0m diz_loss \u001b[38;5;241m=\u001b[39m {\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mtrain_loss\u001b[39m\u001b[38;5;124m'\u001b[39m:[],\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mval_loss\u001b[39m\u001b[38;5;124m'\u001b[39m:[]}\n\u001b[1;32m 3\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m epoch \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mrange\u001b[39m(num_epochs):\n\u001b[0;32m----> 4\u001b[0m train_loss \u001b[38;5;241m=\u001b[39m\u001b[43mtrain_epoch\u001b[49m\u001b[43m(\u001b[49m\u001b[43mencoder\u001b[49m\u001b[43m,\u001b[49m\u001b[43mdecoder\u001b[49m\u001b[43m,\u001b[49m\u001b[43mdevice\u001b[49m\u001b[43m,\u001b[49m\u001b[43mtrain_loader\u001b[49m\u001b[43m,\u001b[49m\u001b[43mloss_fn\u001b[49m\u001b[43m,\u001b[49m\u001b[43moptim\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 5\u001b[0m val_loss \u001b[38;5;241m=\u001b[39m test_epoch(encoder,decoder,device,test_loader,loss_fn)\n\u001b[1;32m 6\u001b[0m \u001b[38;5;28mprint\u001b[39m(\u001b[38;5;124m'\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;124m EPOCH \u001b[39m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[38;5;124m/\u001b[39m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[38;5;124m \u001b[39m\u001b[38;5;130;01m\\t\u001b[39;00m\u001b[38;5;124m train loss \u001b[39m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[38;5;124m \u001b[39m\u001b[38;5;130;01m\\t\u001b[39;00m\u001b[38;5;124m val loss \u001b[39m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[38;5;124m'\u001b[39m\u001b[38;5;241m.\u001b[39mformat(epoch \u001b[38;5;241m+\u001b[39m \u001b[38;5;241m1\u001b[39m, num_epochs,train_loss,val_loss))\n",
"Input \u001b[0;32mIn [33]\u001b[0m, in \u001b[0;36mtrain_epoch\u001b[0;34m(encoder, decoder, device, dataloader, loss_fn, optimizer)\u001b[0m\n\u001b[1;32m 6\u001b[0m train_loss \u001b[38;5;241m=\u001b[39m []\n\u001b[1;32m 7\u001b[0m \u001b[38;5;66;03m# Iterate the dataloader (we do not need the label values, this is unsupervised learning)\u001b[39;00m\n\u001b[0;32m----> 8\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m image_batch, _ \u001b[38;5;129;01min\u001b[39;00m dataloader: \u001b[38;5;66;03m# with \"_\" we just ignore the labels (the second element of the dataloader tuple)\u001b[39;00m\n\u001b[1;32m 9\u001b[0m \u001b[38;5;66;03m# Move tensor to the proper device\u001b[39;00m\n\u001b[1;32m 10\u001b[0m image_batch \u001b[38;5;241m=\u001b[39m image_batch\u001b[38;5;241m.\u001b[39mto(device)\n\u001b[1;32m 11\u001b[0m \u001b[38;5;66;03m# Encode data\u001b[39;00m\n",
"\u001b[0;31mValueError\u001b[0m: too many values to unpack (expected 2)"
]
}
],
"source": [
"num_epochs = 30\n",
"diz_loss = {'train_loss':[],'val_loss':[]}\n",
"for epoch in range(num_epochs):\n",
" train_loss =train_epoch(encoder,decoder,device,train_loader,loss_fn,optim)\n",
" val_loss = test_epoch(encoder,decoder,device,test_loader,loss_fn)\n",
" print('\\n EPOCH {}/{} \\t train loss {} \\t val loss {}'.format(epoch + 1, num_epochs,train_loss,val_loss))\n",
" diz_loss['train_loss'].append(train_loss)\n",
" diz_loss['val_loss'].append(val_loss)\n",
" plot_ae_outputs(encoder,decoder,n=10)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.12"
}
},
"nbformat": 4,
"nbformat_minor": 5
}