examples/model_examples.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-06-27T18:38:10.349669Z",
     "start_time": "2024-06-27T18:38:10.343595Z"
    }
   },
   "outputs": [],
   "source": [
    "# set up an absolute path to the project \n",
    "# not needed in case of `pip install`\n",
    "%run -i tools/setup_env.py"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Usage examples of `torchcnnbuilder.models`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This submodule contains CNN-model templates that were assembled using `torchcnnbuilder.builder`. So far, there is only a model template for predicting n-dimensional time series *(two-dimensional convolutional layers are used)*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Model `ForecasterBase`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Initialization params:\n",
    "* **input_size** (Tuple[int, int]): input size of the input tensor of the one time point\n",
    "* **n_layers** (int): number of the convolution layers in the encoder part\n",
    "* **in_time_points** (int): number of time points (channels) in the first input tensor (prehistory size)\n",
    "* **out_time_points** number of time points (channels) in the last output tensor (forecasting size)\n",
    "* **conv_dim**: the dimension of the convolutional operation 2 or 3. Default: 2\n",
    "* **n_transpose_layers** (Optional[int]): number of the transpose convolution layers in the encoder part. Default: None (same as n_layers)\n",
    "* **convolve_params** (Dict[str, int]): parameters of convolutional layers (by default same as in torch). Default: None\n",
    "* **transpose_convolve_params** (Dict[str, int]): parameters of transpose convolutional layers (by default same as in torch). Default: None\n",
    "* **activation_function** (nn.Module): activation function. Default: nn.ReLU(inplace=True)\n",
    "* **finish_activation_function** (Union[str, Optional[nn.Module]]): last activation function, can be same as activation_function (str `'same'`). Default: None\n",
    "* **normalization** (str): choice of normalization between str `'dropout'` and `'batchnorm'`. Default: None\n",
    "\n",
    "Other attributes:\n",
    "* **convolve** (nn.Sequential): convolutional sequence - encoder part\n",
    "* **transpose** (nn.Sequential): transpose convolutional sequence - decoder part\n",
    "* **conv_channels** (List[int]): history list of output channels after each convolutional layer\n",
    "* **transpose_conv_channels** (List[int]): history list of output channels after each transposed convolutional layer\n",
    "* **conv_layers** (List[tuple]): history list of output tensor sizes after each convolutional layer\n",
    "* **transpose_conv_layers** (List[tuple]): history list of output tensor sizes after each transposed convolutional layer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is the template class of the time series prediction CNN-architecture. The source of the original [article code](https://github.com/ITMO-NSS-team/ice-concentration-prediction-paper?ysclid=lrhxbvsk8s328492826). Below is a diagram of the original architecture of the model (2D version)\n",
    "\n",
    "<img src=\"./tools/media/ForecasterBase.png\" alt=\"forecaster model\" style=\"width:70%; display: block; margin-left: auto; margin-right: auto;\">"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:39.235640Z",
     "iopub.status.busy": "2024-01-25T18:18:39.234694Z",
     "iopub.status.idle": "2024-01-25T18:18:42.600887Z",
     "shell.execute_reply": "2024-01-25T18:18:42.600101Z",
     "shell.execute_reply.started": "2024-01-25T18:18:39.235602Z"
    },
    "ExecuteTime": {
     "end_time": "2024-07-13T06:52:24.694379Z",
     "start_time": "2024-07-13T06:52:23.888551Z"
    }
   },
   "outputs": [],
   "source": [
    "from torchcnnbuilder.models import ForecasterBase"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:42.602812Z",
     "iopub.status.busy": "2024-01-25T18:18:42.602398Z",
     "iopub.status.idle": "2024-01-25T18:18:42.667243Z",
     "shell.execute_reply": "2024-01-25T18:18:42.666551Z",
     "shell.execute_reply.started": "2024-01-25T18:18:42.602785Z"
    },
    "ExecuteTime": {
     "end_time": "2024-07-13T06:52:33.333390Z",
     "start_time": "2024-07-13T06:52:33.318811Z"
    }
   },
   "outputs": [],
   "source": [
    "import torch.nn as nn\n",
    "\n",
    "# example of initialization\n",
    "model = ForecasterBase(input_size=[65, 65],\n",
    "                       in_time_points=120,\n",
    "                       out_time_points=40,\n",
    "                       n_layers=5,\n",
    "                       normalization='batchnorm',\n",
    "                       finish_activation_function=nn.ReLU(inplace=True))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's train the model on the synthetic data that was demonstrated in the `prerocess_examples.ipynb` file. First, we will generate the necessary data and create datasets for the train and test parts. We will predict 40 frames of the square movement based on the previous 120 frames and will use batch normalization. A frame-by-frame animation of the train part is attached below:\n",
    "\n",
    "<img src=\"./tools/media/train.gif\" alt=\"animation\" style=\"width:40%; display: block; margin-left: auto; margin-right: auto;\">"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:42.685262Z",
     "iopub.status.busy": "2024-01-25T18:18:42.684978Z",
     "iopub.status.idle": "2024-01-25T18:18:42.776584Z",
     "shell.execute_reply": "2024-01-25T18:18:42.775749Z",
     "shell.execute_reply.started": "2024-01-25T18:18:42.685238Z"
    },
    "ExecuteTime": {
     "end_time": "2024-07-13T06:35:55.750781Z",
     "start_time": "2024-07-13T06:35:55.161216Z"
    }
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.optim as optim\n",
    "from torch.utils.data import DataLoader\n",
    "\n",
    "# using ssim as loss and score metric (firstly you need to pip install it)\n",
    "from pytorch_msssim import ssim\n",
    "\n",
    "# time-series preprocessing\n",
    "from torchcnnbuilder.preprocess.time_series import multi_output_tensor, single_output_tensor\n",
    "\n",
    "from tqdm.notebook import tqdm\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:42.778004Z",
     "iopub.status.busy": "2024-01-25T18:18:42.777664Z",
     "iopub.status.idle": "2024-01-25T18:18:43.018338Z",
     "shell.execute_reply": "2024-01-25T18:18:43.017463Z",
     "shell.execute_reply.started": "2024-01-25T18:18:42.777973Z"
    },
    "ExecuteTime": {
     "end_time": "2024-07-13T06:35:56.651775Z",
     "start_time": "2024-07-13T06:35:56.582592Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Calculating on device: cpu\n"
     ]
    }
   ],
   "source": [
    "device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
    "model = model.to(device)\n",
    "\n",
    "print(f'Calculating on device: {device}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "%%capture\n",
    "from tools.generating_time_series import synthetic_time_series"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:43.077882Z",
     "iopub.status.busy": "2024-01-25T18:18:43.077309Z",
     "iopub.status.idle": "2024-01-25T18:18:47.630327Z",
     "shell.execute_reply": "2024-01-25T18:18:47.629371Z",
     "shell.execute_reply.started": "2024-01-25T18:18:43.077852Z"
    }
   },
   "outputs": [],
   "source": [
    "%%capture\n",
    "# generating data\n",
    "_, data = synthetic_time_series(num_frames=360,\n",
    "                                matrix_size=65,\n",
    "                                square_size=13)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:47.632216Z",
     "iopub.status.busy": "2024-01-25T18:18:47.631955Z",
     "iopub.status.idle": "2024-01-25T18:18:47.636452Z",
     "shell.execute_reply": "2024-01-25T18:18:47.635588Z",
     "shell.execute_reply.started": "2024-01-25T18:18:47.632194Z"
    }
   },
   "outputs": [],
   "source": [
    "# train consists of three cycles of a full circle of the square so that the model can see the trend\n",
    "train = data + data + data\n",
    "test = data[10:170]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:47.637856Z",
     "iopub.status.busy": "2024-01-25T18:18:47.637563Z",
     "iopub.status.idle": "2024-01-25T18:18:47.648489Z",
     "shell.execute_reply": "2024-01-25T18:18:47.647746Z",
     "shell.execute_reply.started": "2024-01-25T18:18:47.637832Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train dataset len: 1080, One matrix shape: (65, 65)\n",
      "Test dataset len: 160, One matrix shape: (65, 65)\n"
     ]
    }
   ],
   "source": [
    "print(f'Train dataset len: {len(train)}, One matrix shape: {train[0].shape}')\n",
    "print(f'Test dataset len: {len(test)}, One matrix shape: {test[0].shape}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:47.651083Z",
     "iopub.status.busy": "2024-01-25T18:18:47.650649Z",
     "iopub.status.idle": "2024-01-25T18:18:49.789509Z",
     "shell.execute_reply": "2024-01-25T18:18:49.788528Z",
     "shell.execute_reply.started": "2024-01-25T18:18:47.651052Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.10/site-packages/torchcnnbuilder/preprocess/time_series.py:87: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at /usr/local/src/pytorch/torch/csrc/utils/tensor_new.cpp:245.)\n",
      "  tensors = torch.Tensor(data)\n"
     ]
    }
   ],
   "source": [
    "train_dataset = multi_output_tensor(data=train,\n",
    "                                    pre_history_len=120,\n",
    "                                    forecast_len=40)\n",
    "\n",
    "test_dataset = single_output_tensor(data=test, \n",
    "                                   forecast_len=40)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:49.790939Z",
     "iopub.status.busy": "2024-01-25T18:18:49.790630Z",
     "iopub.status.idle": "2024-01-25T18:18:49.799103Z",
     "shell.execute_reply": "2024-01-25T18:18:49.798206Z",
     "shell.execute_reply.started": "2024-01-25T18:18:49.790914Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "batch number: 0\n",
      "X shape: torch.Size([120, 65, 65])\n",
      "Y shape: torch.Size([40, 65, 65])\n",
      "\n",
      "Dataset len (number of batches/X-windows): 920\n"
     ]
    }
   ],
   "source": [
    "# checking train_dataset\n",
    "for i, batch in enumerate(train_dataset):\n",
    "    print(f'batch number: {i}',\n",
    "          f'X shape: {batch[0].shape}\\nY shape: {batch[1].shape}',\n",
    "          sep='\\n',\n",
    "          end='\\n\\n')\n",
    "    break\n",
    "print(f'Dataset len (number of batches/X-windows): {len(train_dataset)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:49.800594Z",
     "iopub.status.busy": "2024-01-25T18:18:49.800274Z",
     "iopub.status.idle": "2024-01-25T18:18:49.821674Z",
     "shell.execute_reply": "2024-01-25T18:18:49.820798Z",
     "shell.execute_reply.started": "2024-01-25T18:18:49.800564Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X shape: torch.Size([120, 65, 65])\n",
      "Y shape: torch.Size([40, 65, 65])\n",
      "\n",
      "Dataset len (number of batches/X-windows): 1\n"
     ]
    }
   ],
   "source": [
    "# checking test_dataset\n",
    "for i, batch in enumerate(test_dataset):\n",
    "    print(f'X shape: {batch[0].shape}\\nY shape: {batch[1].shape}',\n",
    "          end='\\n\\n')\n",
    "\n",
    "print(f'Dataset len (number of batches/X-windows): {len(test_dataset)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:49.822940Z",
     "iopub.status.busy": "2024-01-25T18:18:49.822646Z",
     "iopub.status.idle": "2024-01-25T18:18:49.832032Z",
     "shell.execute_reply": "2024-01-25T18:18:49.831180Z",
     "shell.execute_reply.started": "2024-01-25T18:18:49.822918Z"
    }
   },
   "outputs": [],
   "source": [
    "epochs = 1000\n",
    "batch_size = 50\n",
    "\n",
    "dataloader = DataLoader(train_dataset, batch_size=batch_size, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:18:49.833885Z",
     "iopub.status.busy": "2024-01-25T18:18:49.833209Z",
     "iopub.status.idle": "2024-01-25T18:18:49.843440Z",
     "shell.execute_reply": "2024-01-25T18:18:49.842692Z",
     "shell.execute_reply.started": "2024-01-25T18:18:49.833860Z"
    }
   },
   "outputs": [],
   "source": [
    "optimizer = optim.AdamW(model.parameters(), lr=1e-4)\n",
    "criterion = nn.L1Loss()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T18:19:00.382605Z",
     "iopub.status.busy": "2024-01-25T18:19:00.381868Z",
     "iopub.status.idle": "2024-01-25T20:25:36.410329Z",
     "shell.execute_reply": "2024-01-25T20:25:36.409349Z",
     "shell.execute_reply.started": "2024-01-25T18:19:00.382574Z"
    }
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d203b880e0774f95a231e003b68bcdbd",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/1000 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch : 1/1000, recon loss = 0.13874564\n",
      "epoch : 11/1000, recon loss = 0.02502373\n",
      "epoch : 21/1000, recon loss = 0.02142571\n",
      "epoch : 31/1000, recon loss = 0.01604175\n",
      "epoch : 41/1000, recon loss = 0.01732834\n",
      "epoch : 51/1000, recon loss = 0.01441913\n",
      "epoch : 61/1000, recon loss = 0.01253408\n",
      "epoch : 71/1000, recon loss = 0.01293192\n",
      "epoch : 81/1000, recon loss = 0.01109324\n",
      "epoch : 91/1000, recon loss = 0.01111370\n",
      "epoch : 101/1000, recon loss = 0.01173207\n",
      "epoch : 111/1000, recon loss = 0.01020600\n",
      "epoch : 121/1000, recon loss = 0.00907689\n",
      "epoch : 131/1000, recon loss = 0.00993255\n",
      "epoch : 141/1000, recon loss = 0.00914604\n",
      "epoch : 151/1000, recon loss = 0.01321855\n",
      "epoch : 161/1000, recon loss = 0.01007256\n",
      "epoch : 171/1000, recon loss = 0.00887348\n",
      "epoch : 181/1000, recon loss = 0.00775756\n",
      "epoch : 191/1000, recon loss = 0.00820381\n",
      "epoch : 201/1000, recon loss = 0.00821034\n",
      "epoch : 211/1000, recon loss = 0.00732546\n",
      "epoch : 221/1000, recon loss = 0.00733761\n",
      "epoch : 231/1000, recon loss = 0.00715669\n",
      "epoch : 241/1000, recon loss = 0.00705734\n",
      "epoch : 251/1000, recon loss = 0.00732890\n",
      "epoch : 261/1000, recon loss = 0.00732492\n",
      "epoch : 271/1000, recon loss = 0.00671205\n",
      "epoch : 281/1000, recon loss = 0.00653366\n",
      "epoch : 291/1000, recon loss = 0.00691777\n",
      "epoch : 301/1000, recon loss = 0.00625921\n",
      "epoch : 311/1000, recon loss = 0.00606658\n",
      "epoch : 321/1000, recon loss = 0.00588256\n",
      "epoch : 331/1000, recon loss = 0.00669962\n",
      "epoch : 341/1000, recon loss = 0.00587592\n",
      "epoch : 351/1000, recon loss = 0.00800121\n",
      "epoch : 361/1000, recon loss = 0.00774155\n",
      "epoch : 371/1000, recon loss = 0.00567457\n",
      "epoch : 381/1000, recon loss = 0.00578017\n",
      "epoch : 391/1000, recon loss = 0.00560832\n",
      "epoch : 401/1000, recon loss = 0.00532252\n",
      "epoch : 411/1000, recon loss = 0.00854660\n",
      "epoch : 421/1000, recon loss = 0.00586663\n",
      "epoch : 431/1000, recon loss = 0.00550154\n",
      "epoch : 441/1000, recon loss = 0.00521067\n",
      "epoch : 451/1000, recon loss = 0.00531547\n",
      "epoch : 461/1000, recon loss = 0.00512418\n",
      "epoch : 471/1000, recon loss = 0.00506014\n",
      "epoch : 481/1000, recon loss = 0.00498361\n",
      "epoch : 491/1000, recon loss = 0.00536798\n",
      "epoch : 501/1000, recon loss = 0.00480852\n",
      "epoch : 511/1000, recon loss = 0.00474047\n",
      "epoch : 521/1000, recon loss = 0.00464868\n",
      "epoch : 531/1000, recon loss = 0.00508922\n",
      "epoch : 541/1000, recon loss = 0.00515903\n",
      "epoch : 551/1000, recon loss = 0.00473539\n",
      "epoch : 561/1000, recon loss = 0.00456650\n",
      "epoch : 571/1000, recon loss = 0.00444237\n",
      "epoch : 581/1000, recon loss = 0.00491305\n",
      "epoch : 591/1000, recon loss = 0.00444375\n",
      "epoch : 601/1000, recon loss = 0.00450915\n",
      "epoch : 611/1000, recon loss = 0.00483104\n",
      "epoch : 621/1000, recon loss = 0.00420071\n",
      "epoch : 631/1000, recon loss = 0.00447442\n",
      "epoch : 641/1000, recon loss = 0.00445188\n",
      "epoch : 651/1000, recon loss = 0.00444460\n",
      "epoch : 661/1000, recon loss = 0.00418368\n",
      "epoch : 671/1000, recon loss = 0.00416128\n",
      "epoch : 681/1000, recon loss = 0.00435332\n",
      "epoch : 691/1000, recon loss = 0.00430444\n",
      "epoch : 701/1000, recon loss = 0.00407504\n",
      "epoch : 711/1000, recon loss = 0.00381193\n",
      "epoch : 721/1000, recon loss = 0.00399310\n",
      "epoch : 731/1000, recon loss = 0.00392363\n",
      "epoch : 741/1000, recon loss = 0.00393251\n",
      "epoch : 751/1000, recon loss = 0.00387976\n",
      "epoch : 761/1000, recon loss = 0.00371953\n",
      "epoch : 771/1000, recon loss = 0.00439701\n",
      "epoch : 781/1000, recon loss = 0.00589116\n",
      "epoch : 791/1000, recon loss = 0.00398315\n",
      "epoch : 801/1000, recon loss = 0.00389089\n",
      "epoch : 811/1000, recon loss = 0.00373581\n",
      "epoch : 821/1000, recon loss = 0.00416677\n",
      "epoch : 831/1000, recon loss = 0.00355723\n",
      "epoch : 841/1000, recon loss = 0.00399345\n",
      "epoch : 851/1000, recon loss = 0.00371136\n",
      "epoch : 861/1000, recon loss = 0.00366443\n",
      "epoch : 871/1000, recon loss = 0.00351296\n",
      "epoch : 881/1000, recon loss = 0.00383161\n",
      "epoch : 891/1000, recon loss = 0.00366378\n",
      "epoch : 901/1000, recon loss = 0.00350117\n",
      "epoch : 911/1000, recon loss = 0.00362062\n",
      "epoch : 921/1000, recon loss = 0.00349954\n",
      "epoch : 931/1000, recon loss = 0.00348236\n",
      "epoch : 941/1000, recon loss = 0.00381448\n",
      "epoch : 951/1000, recon loss = 0.00433518\n",
      "epoch : 961/1000, recon loss = 0.00333077\n",
      "epoch : 971/1000, recon loss = 0.00338416\n",
      "epoch : 981/1000, recon loss = 0.00329954\n",
      "epoch : 991/1000, recon loss = 0.00331641\n"
     ]
    }
   ],
   "source": [
    "losses = []\n",
    "\n",
    "for epoch in tqdm(range(epochs)):\n",
    "    loss = 0\n",
    "    \n",
    "    for train_features, test_features in dataloader:\n",
    "        train_features = train_features.to(device)\n",
    "        test_features = test_features.to(device)\n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        outputs = model(train_features)\n",
    "    \n",
    "        train_loss = criterion(outputs, test_features)\n",
    "        \n",
    "        train_loss.backward()\n",
    "        optimizer.step()\n",
    "        \n",
    "        loss += train_loss.item()\n",
    "    \n",
    "    loss = loss / len(dataloader)\n",
    "    \n",
    "    losses.append(loss)\n",
    "    \n",
    "    if epoch % 10 == 0 or epoch - 1 == epochs:\n",
    "        tqdm.write(\"epoch : {}/{}, recon loss = {:.8f}\".format(epoch+1, epochs, loss))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T20:25:44.853196Z",
     "iopub.status.busy": "2024-01-25T20:25:44.852841Z",
     "iopub.status.idle": "2024-01-25T20:25:45.081999Z",
     "shell.execute_reply": "2024-01-25T20:25:45.081117Z",
     "shell.execute_reply.started": "2024-01-25T20:25:44.853169Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(list(range(epochs)), losses)\n",
    "plt.grid()\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('L1 Loss')\n",
    "plt.title('Loss history')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T20:40:46.275642Z",
     "iopub.status.busy": "2024-01-25T20:40:46.274946Z",
     "iopub.status.idle": "2024-01-25T20:40:46.289144Z",
     "shell.execute_reply": "2024-01-25T20:40:46.288335Z",
     "shell.execute_reply.started": "2024-01-25T20:40:46.275610Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "L1 score: 0.005647140555083752\n",
      "SSIM score: 0.9594206213951111\n"
     ]
    }
   ],
   "source": [
    "# checking the result score\n",
    "with torch.no_grad():\n",
    "    model.eval()\n",
    "    for X, Y in test_dataset:\n",
    "        X = X[None, :].to(device)\n",
    "        Y = Y[None, :].to(device)\n",
    "            \n",
    "        outputs = model(X)\n",
    "        \n",
    "        l1_val = criterion(outputs, Y)\n",
    "        ssim_val = ssim(outputs, Y, data_range=1, size_average=False)\n",
    "        \n",
    "        print(f'L1 score: {l1_val}')\n",
    "        print(f'SSIM score: {ssim_val[0]}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The L1 score tends to 0, and the SSIM score should tend to 1, which means complete similarity with the original image *(in our case, the forecast of the time series)*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T20:36:40.577637Z",
     "iopub.status.busy": "2024-01-25T20:36:40.577183Z",
     "iopub.status.idle": "2024-01-25T20:36:40.854543Z",
     "shell.execute_reply": "2024-01-25T20:36:40.853721Z",
     "shell.execute_reply.started": "2024-01-25T20:36:40.577600Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7d04bf5bc400>"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow((1-outputs[0].detach().cpu().numpy()[39]), cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T20:36:41.766028Z",
     "iopub.status.busy": "2024-01-25T20:36:41.765654Z",
     "iopub.status.idle": "2024-01-25T20:36:41.976917Z",
     "shell.execute_reply": "2024-01-25T20:36:41.976026Z",
     "shell.execute_reply.started": "2024-01-25T20:36:41.766002Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7d04bf409540>"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(1 - Y[0].cpu()[39], cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# saving results as a gif-files\n",
    "from tools.generating_time_series import save_gif"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-25T20:31:46.112045Z",
     "iopub.status.busy": "2024-01-25T20:31:46.111663Z",
     "iopub.status.idle": "2024-01-25T20:31:49.292278Z",
     "shell.execute_reply": "2024-01-25T20:31:49.291529Z",
     "shell.execute_reply.started": "2024-01-25T20:31:46.112020Z"
    }
   },
   "outputs": [],
   "source": [
    "%%capture\n",
    "save_gif(matrices=outputs[0].detach().cpu().numpy(),\n",
    "         name='predict')\n",
    "\n",
    "save_gif(matrices=test[-40:],\n",
    "         name='test')\n",
    "\n",
    "save_gif(matrices=train,\n",
    "         name='train')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# saving models weights\n",
    "torch.save(model.state_dict(), \"model.pth\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": false
   },
   "source": [
    "The final result is below (test data and predict for next 40 frames by 120 previous frames)\n",
    "\n",
    "<table>\n",
    "  <tr>\n",
    "    <td align=\"center\">\n",
    "        <b>Test<b>\n",
    "      <img src=\"./tools/media/test.gif\">\n",
    "    </td>\n",
    "    <td align=\"center\">\n",
    "        <b>Predict<b>\n",
    "      <img src=\"./tools/media/predict.gif\">\n",
    "    </td>\n",
    "  </tr>\n",
    "</table>"
   ]
  },
  {
   "cell_type": "markdown",
   "source": [
    "Also, you can do the same with 3d or 1d convolution version of the Forecaster model:\n",
    "\n",
    "> Note: if you use 3d convolutions, channels number will be count with using param `channel_growth_rate='power'` in order to make model less heavy. Also, your input_size param should be [H, W]. ForecasterBase3d works only with 2d tensors, which create a 3d tensor with a time period of `in_time_points`.\n",
    "\n",
    "> Note: ForecasterBase1D needs code refactoring wth `in_time_points` and `out_time_point`. Be careful with dimensions to use it for 1d predictions. It will be fixed in future minor versions"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [
    {
     "data": {
      "text/plain": "ForecasterBase(\n  (convolve): Sequential(\n    (conv 1): Sequential(\n      (0): Conv3d(1, 2, kernel_size=(3, 3, 3), stride=(1, 1, 1))\n      (1): Dropout3d(p=0.5, inplace=False)\n      (2): ReLU(inplace=True)\n    )\n    (conv 2): Sequential(\n      (0): Conv3d(2, 3, kernel_size=(3, 3, 3), stride=(1, 1, 1))\n      (1): Dropout3d(p=0.5, inplace=False)\n      (2): ReLU(inplace=True)\n    )\n    (conv 3): Sequential(\n      (0): Conv3d(3, 4, kernel_size=(3, 3, 3), stride=(1, 1, 1))\n      (1): Dropout3d(p=0.5, inplace=False)\n      (2): ReLU(inplace=True)\n    )\n  )\n  (transpose): Sequential(\n    (deconv 1): Sequential(\n      (0): ConvTranspose3d(4, 3, kernel_size=(3, 3, 3), stride=(1, 1, 1))\n      (1): Dropout3d(p=0.5, inplace=False)\n      (2): ReLU(inplace=True)\n    )\n    (deconv 2): Sequential(\n      (0): ConvTranspose3d(3, 2, kernel_size=(3, 3, 3), stride=(1, 1, 1))\n      (1): Dropout3d(p=0.5, inplace=False)\n      (2): ReLU(inplace=True)\n    )\n    (deconv 3): Sequential(\n      (0): ConvTranspose3d(2, 1, kernel_size=(3, 3, 3), stride=(1, 1, 1))\n      (1): Dropout3d(p=0.5, inplace=False)\n      (2): ReLU(inplace=True)\n    )\n    (resize): AdaptiveAvgPool3d(output_size=(5, 65, 65))\n  )\n)"
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "forecaster3d = ForecasterBase(input_size=[65, 65],\n",
    "                              in_time_points=10,\n",
    "                              out_time_points=5,\n",
    "                              n_layers=3,\n",
    "                              normalization='dropout',\n",
    "                              conv_dim=3,\n",
    "                              finish_activation_function=nn.ReLU(inplace=True))\n",
    "forecaster3d"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-07-13T06:52:37.649792Z",
     "start_time": "2024-07-13T06:52:37.641744Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [],
   "metadata": {
    "collapsed": false
   }
  }
 ],
 "metadata": {
  "kaggle": {
   "accelerator": "nvidiaTeslaT4",
   "dataSources": [
    {
     "datasetId": 4298338,
     "sourceId": 7393613,
     "sourceType": "datasetVersion"
    }
   ],
   "dockerImageVersionId": 30636,
   "isGpuEnabled": true,
   "isInternetEnabled": true,
   "language": "python",
   "sourceType": "notebook"
  },
  "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": "3.10.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}