{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "competent-surrey",
   "metadata": {},
   "source": [
    "## Neural Population Control\n",
    "#### Pipeline for roi and testing the encoding models, performing in-silico ephys and generating the most-exciting images (MEIs)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "adolescent-parish",
   "metadata": {},
   "source": [
    "Change below the constants that pertain to your experiment, such as names/dates/conditions, and setup info (pixels per degree):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "short-postage",
   "metadata": {},
   "outputs": [],
   "source": [
    "import npc\n",
    "import numpy as np\n",
    "import math\n",
    "\n",
    "# data-constants:\n",
    "monkey_names = ['monkeyF','monkeyF','monkeyF','monkeyF','monkeyF','monkeyF','monkeyF','monkeyF',\n",
    "              'monkeyF','monkeyF','monkeyF','monkeyF','monkeyF','monkeyF','monkeyF','monkeyF']\n",
    "roi_names = ['1','2','3','4','5','6','7','8','9','10','11','12','13','14','15','16']\n",
    "gen_path = '/media/stijn/2bb74e85-3681-4561-88b7-abd98482de61/paolo/Data/'\n",
    "\n",
    "# layers for traning and testing:\n",
    "# 'False' on top will run the grid-search first and then stop\n",
    "# 'True' on top will run the grid-search first and then will train the best layer\n",
    "layers = ['False',\n",
    "          'conv2d1',\n",
    "          'conv2d2',\n",
    "          'mixed3a',\n",
    "          'mixed3b',\n",
    "          'mixed4a',\n",
    "          'mixed4b',\n",
    "          'mixed4c',\n",
    "          'mixed4d',\n",
    "          'mixed4e',\n",
    "          'mixed5a',\n",
    "          'mixed5b']\n",
    "\n",
    "\n",
    "# constants for mei generation:\n",
    "seed = 0\n",
    "pixperdeg = 25.92\n",
    "stim_size = 500\n",
    "mask_size = 128\n",
    "shift_x = 100\n",
    "shift_y = 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ceramic-white",
   "metadata": {},
   "outputs": [],
   "source": [
    "import npc\n",
    "import numpy as np\n",
    "import math\n",
    "\n",
    "# data-constants:\n",
    "monkey_names = ['monkeyN']\n",
    "roi_names = ['transient_IT']\n",
    "gen_path = '/media/stijn/2bb74e85-3681-4561-88b7-abd98482de61/paolo/Data/'\n",
    "\n",
    "# layers for traning and testing:\n",
    "# 'False' on top will run the grid-search first and then stop\n",
    "# 'True' on top will run the grid-search first and then will train the best layer\n",
    "layers = ['False',\n",
    "          'conv2d1',\n",
    "          'conv2d2',\n",
    "          'mixed4e']\n",
    "\n",
    "\n",
    "# constants for mei generation:\n",
    "seed = 0\n",
    "pixperdeg = 25.92\n",
    "stim_size = 500\n",
    "mask_size = 128\n",
    "shift_x = 100\n",
    "shift_y = 100"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "wrong-radio",
   "metadata": {},
   "source": [
    "Perform the data formatting and train the models:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "metropolitan-zimbabwe",
   "metadata": {},
   "outputs": [],
   "source": [
    "### extract the data and train all the models\n",
    "# it'll take a lot of time!\n",
    "for monkey_name, roi in zip(monkey_names, roi_names):\n",
    "    !python3 extract_data.py $gen_path $monkey_name $roi\n",
    "    for layer in layers:\n",
    "        !CUDA_VISIBLE_DEVICES=0 python3 train_neural_model.py $gen_path $monkey_name $roi $layer $layers"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "downtown-restoration",
   "metadata": {},
   "source": [
    "Finally, compute the MEIs and the properties of each neuron:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "turned-laser",
   "metadata": {},
   "outputs": [],
   "source": [
    "# plot and save the modeled RFs and MEIs:\n",
    "for monkey_name, roi in zip(monkey_names, roi_names):\n",
    "    print('======================')\n",
    "    print('Processing: ' + monkey_name + ', ' + roi)\n",
    "    if roi != '':\n",
    "        roi = '_' + roi\n",
    "    all_corrs,layer = npc.plot_layer_corrs(gen_path,monkey_name,layers[1:],roi)\n",
    "    monkey_model,n_neurons = npc.load_trained_model(gen_path,monkey_name,roi,layer)\n",
    "    corrs = npc.compute_val_corrs(gen_path,monkey_name,monkey_model,roi)\n",
    "    goods,all_good_rfs = npc.good_neurons(monkey_model,n_neurons,corrs,make_plots=False)\n",
    "    #all_corrs = []\n",
    "    _ = npc.gaussian_RFs(gen_path,all_good_rfs,goods,corrs,all_corrs,pixperdeg,stim_size,mask_size,monkey_model,monkey_name,roi,shift_x,shift_y)\n",
    "    npc.generate_MEIS(gen_path,monkey_model,monkey_name,goods,roi)"
   ]
  }
 ],
 "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": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}