notebooks/model_tutorial/1b_model_training_sensorium+.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "# How to train the Baseline Models for the SENSORIUM+ track"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "### This notebook will show how to\n",
    "- instantiate dataloader for the Sensorium+ track\n",
    "- instantiate pytorch model\n",
    "- instantiate a trainer function\n",
    "- train two baselines for this competition track\n",
    "- save the model weights (the model weights can already be found in './model_checkpoints/pretrained/')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "### Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n",
    "\n",
    "from nnfabrik.builder import get_data, get_model, get_trainer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "### Instantiate DataLoader for Sensorium+"
   ]
  },
  {
   "cell_type": "markdown",
   "source": [
    "The only difference to the Sensorium track is that here, we include the behavioral variables and the eye position,\n",
    "by setting include_behavior=True, and include_eye_position=True.\n",
    "this will append the behavioral variables to the input images, and the eye position will be passed to\n",
    "the shifter network of the model.\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [
    "# loading the SENSORIUM+ dataset\n",
    "filenames = ['../data/static27204-5-13-GrayImageNet-94c6ff995dac583098847cfecd43e7b6.zip', ]\n",
    "\n",
    "\n",
    "dataset_fn = 'sensorium.datasets.static_loaders'\n",
    "dataset_config = {'paths': filenames,\n",
    "                 'normalize': True,\n",
    "                 'include_behavior': True,\n",
    "                 'include_eye_position': True,\n",
    "                 'batch_size': 128,\n",
    "                 'scale':.25,\n",
    "                 }\n",
    "\n",
    "dataloaders = get_data(dataset_fn, dataset_config)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Instantiate State of the Art Model (SOTA)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Because the behavioral variables are available, we instantiate the Shifter network\n",
    "by setting Shifter=True in the model configuration."
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [
    "model_fn = 'sensorium.models.stacked_core_full_gauss_readout'\n",
    "model_config = {'pad_input': False,\n",
    "  'stack': -1,\n",
    "  'layers': 4,\n",
    "  'input_kern': 9,\n",
    "  'gamma_input': 6.3831,\n",
    "  'gamma_readout': 0.0076,\n",
    "  'hidden_kern': 7,\n",
    "  'hidden_channels': 64,\n",
    "  'depth_separable': True,\n",
    "  'grid_mean_predictor': {'type': 'cortex',\n",
    "   'input_dimensions': 2,\n",
    "   'hidden_layers': 1,\n",
    "   'hidden_features': 30,\n",
    "   'final_tanh': True},\n",
    "  'init_sigma': 0.1,\n",
    "  'init_mu_range': 0.3,\n",
    "  'gauss_type': 'full',\n",
    "  'shifter': True,\n",
    "}\n",
    "\n",
    "model = get_model(model_fn=model_fn,\n",
    "                  model_config=model_config,\n",
    "                  dataloaders=dataloaders,\n",
    "                  seed=42,)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Because the behavioral variables are available, we instantiate the Shifter network\n",
    "by setting Shifter=True in the model configuration."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "model_fn = 'sensorium.models.stacked_core_full_gauss_readout'\n",
    "model_config = {'pad_input': False,\n",
    "  'stack': -1,\n",
    "  'layers': 4,\n",
    "  'input_kern': 9,\n",
    "  'gamma_input': 6.3831,\n",
    "  'gamma_readout': 0.0076,\n",
    "  'hidden_kern': 7,\n",
    "  'hidden_channels': 64,\n",
    "  'depth_separable': True,\n",
    "  'grid_mean_predictor': {'type': 'cortex',\n",
    "   'input_dimensions': 2,\n",
    "   'hidden_layers': 1,\n",
    "   'hidden_features': 30,\n",
    "   'final_tanh': True},\n",
    "  'init_sigma': 0.1,\n",
    "  'init_mu_range': 0.3,\n",
    "  'gauss_type': 'full',\n",
    "  'shifter': True,\n",
    "}\n",
    "\n",
    "model = get_model(model_fn=model_fn,\n",
    "                  model_config=model_config,\n",
    "                  dataloaders=dataloaders,\n",
    "                  seed=42,)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "## Configure Trainer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "trainer_fn = \"sensorium.training.standard_trainer\"\n",
    "\n",
    "trainer_config = {'max_iter': 200,\n",
    "                 'verbose': False,\n",
    "                 'lr_decay_steps': 4,\n",
    "                 'avg_loss': False,\n",
    "                 'lr_init': 0.009,\n",
    "                 }\n",
    "\n",
    "trainer = get_trainer(trainer_fn=trainer_fn, \n",
    "                     trainer_config=trainer_config)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "# Run model training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "validation_score, trainer_output, state_dict = trainer(model, dataloaders, seed=42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "## Save model checkpoints"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "torch.save(model.state_dict(), './model_checkpoints/sensorium_p_sota_model.pth')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "## Load Model Checkpoints"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "model.load_state_dict(torch.load(\"./model_checkpoints/pretrained/sensorium_p_sota_model.pth\"));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "# Train a simple LN model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "# this will remove all nonlinearities from the CNN, and creates essentially a ln model: linear core + readout, with a subsequent non-linearity\n",
    "\n",
    "model_fn = 'sensorium.models.stacked_core_full_gauss_readout'\n",
    "model_config = {'pad_input': False,\n",
    "              'stack': -1,\n",
    "              'layers': 3,\n",
    "              'input_kern': 9,\n",
    "              'gamma_input': 6.3831,\n",
    "              'gamma_readout': 0.0076,\n",
    "              'hidden_kern': 7,\n",
    "              'hidden_channels': 64,\n",
    "              'grid_mean_predictor': {'type': 'cortex',\n",
    "              'input_dimensions': 2,\n",
    "              'hidden_layers': 1,\n",
    "              'hidden_features': 30,\n",
    "              'final_tanh': True},\n",
    "              'depth_separable': True,\n",
    "              'init_sigma': 0.1,\n",
    "              'init_mu_range': 0.3,\n",
    "              'gauss_type': 'full',\n",
    "              'linear': True,\n",
    "              'shifter': True,\n",
    "               }\n",
    "model = get_model(model_fn=model_fn,\n",
    "                  model_config=model_config,\n",
    "                  dataloaders=dataloaders,\n",
    "                  seed=42,)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "pycharm": {
     "is_executing": true,
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "validation_score, trainer_output, state_dict = trainer(model, dataloaders, seed=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "torch.save(model.state_dict(), './model_checkpoints/sensorium_p_ln_model.pth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "model.load_state_dict(torch.load(\"./model_checkpoints/pretrained/sensorium_p_ln_model.pth\"));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "---"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.8.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}