activation_plots.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import os\n",
    "import torch\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "cmap = plt.get_cmap(\"viridis\")\n",
    "from tqdm import tqdm\n",
    "from sklearn.metrics import confusion_matrix\n",
    "import seaborn as sns\n",
    "\n",
    "import config\n",
    "from graph_models import FullGraphModel\n",
    "from complete_training_data_processing import CompleteModelsDataProcessor\n",
    "from model_inspection_funcs import neuron_data_from_image, propagate_neuron_data, sample_images\n",
    "\n",
    "num_test_pairs = 200\n",
    "device = torch.device(\"cpu\")\n",
    "dtype = torch.float32"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_processor = CompleteModelsDataProcessor(\n",
    "    neurons=config.neurons,\n",
    "    voronoi_criteria=config.voronoi_criteria,\n",
    "    random_synapses=config.random_synapses,\n",
    "    log_transform_weights=config.log_transform_weights,\n",
    ")\n",
    "\n",
    "model = FullGraphModel(\n",
    "    input_shape=data_processor.number_of_synapses,\n",
    "    num_connectome_passes=config.NUM_CONNECTOME_PASSES,\n",
    "    decision_making_vector=data_processor.decision_making_vector,\n",
    "    batch_size=config.batch_size,\n",
    "    dtype=config.dtype,\n",
    "    edge_weights=data_processor.synaptic_matrix.data,\n",
    "    device=config.DEVICE,\n",
    "    num_classes=len(config.CLASSES),\n",
    ").to(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# horrible data stuff\n",
    "checkpoint = torch.load(\"models/model_2024-05-20 03:41:43.pth\", map_location=\"cpu\")\n",
    "model.load_state_dict(checkpoint[\"model\"])\n",
    "model.eval()\n",
    "connections = (\n",
    "    pd.read_csv(\n",
    "        \"adult_data/connections.csv\",\n",
    "        dtype={\n",
    "            \"pre_root_id\": \"string\",\n",
    "            \"post_root_id\": \"string\",\n",
    "            \"syn_count\": np.int32,\n",
    "        },\n",
    "    )\n",
    "    .groupby([\"pre_root_id\", \"post_root_id\"])\n",
    "    .sum(\"syn_count\")\n",
    "    .reset_index()\n",
    ")\n",
    "\n",
    "connections[\"weight\"] = model.connectome.edge_weight_multiplier.detach()\n",
    "right_root_ids = data_processor.right_root_ids\n",
    "all_neurons = (\n",
    "    pd.read_csv(\"adult_data/classification_clean.csv\")\n",
    "    .merge(right_root_ids, on=\"root_id\")\n",
    "    .fillna(\"Unknown\")\n",
    ")\n",
    "neuron_data = pd.read_csv(\n",
    "    \"adult_data/right_visual_positions_selected_neurons.csv\",\n",
    "    dtype={\"root_id\": \"string\"},\n",
    ").drop(columns=[\"x\", \"y\", \"z\", \"PC1\", \"PC2\"])\n",
    "data_cols = [\"x_axis\", \"y_axis\"]\n",
    "all_coords = pd.read_csv(\"adult_data/all_coords_clean.csv\", dtype={\"root_id\": \"string\"})\n",
    "rational_cell_types = pd.read_csv(\"adult_data/rational_cell_types.csv\")\n",
    "all_neurons[\"decision_making\"] = np.where(\n",
    "    all_neurons[\"cell_type\"].isin(rational_cell_types[\"cell_type\"].values.tolist()),\n",
    "    1,\n",
    "    0,\n",
    ")\n",
    "all_neurons[\"root_id\"] = all_neurons[\"root_id\"].astype(\"string\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 400/400 [17:16<00:00,  2.59s/it]\n"
     ]
    }
   ],
   "source": [
    "num_passes = 4\n",
    "base_dir = \"images/five_to_fifteen/train\"\n",
    "sub_dirs = [\"yellow\", \"blue\"]\n",
    "\n",
    "sampled_images = sample_images(base_dir, sub_dirs, num_test_pairs)\n",
    "\n",
    "dms = {}\n",
    "for img in tqdm(sampled_images):\n",
    "    activated_data = neuron_data_from_image(img, neuron_data)\n",
    "    propagation = propagate_neuron_data(\n",
    "        activated_data, connections, all_coords, all_neurons, num_passes\n",
    "    )\n",
    "    dms[os.path.basename(img)] = propagation[\"decision_making\"][all_neurons[\"decision_making\"] == 1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = pd.DataFrame(dms)\n",
    "means = pd.DataFrame(data.mean(axis=0))\n",
    "means = means.rename(columns={0: \"mean\"})\n",
    "means[\"yellow\"] = [int(a.split(\"_\")[1]) for a in means.index]\n",
    "means[\"blue\"] = [int(a.split(\"_\")[2]) for a in means.index]\n",
    "means[\"color\"] = means[[\"yellow\", \"blue\"]].idxmax(axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: xlabel='color', ylabel='mean'>"
      ]
     },
     "execution_count": 29,
     "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": [
    "sns.boxplot(x=\"color\", y=\"mean\", data=means)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.36  , 0.14  ],\n",
       "       [0.3775, 0.1225]])"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "means[\"pred\"] = np.where(means[\"mean\"] > means[\"mean\"].mean(), \"yellow\", \"blue\")\n",
    "# confusion matrix between color and pred\n",
    "\n",
    "confusion_matrix(means[\"color\"], means[\"pred\"]) / len(means)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy = 0.4825\n"
     ]
    }
   ],
   "source": [
    "print(f\"accuracy = {np.mean(means['color'] == means['pred'])}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# With reshuffled weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# reshuffle column post_rood_id of the dataframe connections\n",
    "shuffled_connections = connections.copy()\n",
    "shuffled_connections[\"post_root_id\"] = np.random.permutation(connections[\"post_root_id\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 400/400 [23:47<00:00,  3.57s/it]\n"
     ]
    }
   ],
   "source": [
    "num_passes = 4\n",
    "base_dir = \"images/five_to_fifteen/train\"\n",
    "sub_dirs = [\"yellow\", \"blue\"]\n",
    "\n",
    "sampled_images = sample_images(base_dir, sub_dirs, num_test_pairs)\n",
    "\n",
    "dms = {}\n",
    "for img in tqdm(sampled_images):\n",
    "    activated_data = neuron_data_from_image(img, neuron_data)\n",
    "    propagation = propagate_neuron_data(\n",
    "        activated_data, shuffled_connections, all_coords, all_neurons, num_passes\n",
    "    )\n",
    "    dms[os.path.basename(img)] = propagation[\"decision_making\"][\n",
    "        all_neurons[\"decision_making\"] == 1\n",
    "    ]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = pd.DataFrame(dms)\n",
    "means = pd.DataFrame(data.mean(axis=0))\n",
    "means = means.rename(columns={0: \"mean\"})\n",
    "means[\"yellow\"] = [int(a.split(\"_\")[1]) for a in means.index]\n",
    "means[\"blue\"] = [int(a.split(\"_\")[2]) for a in means.index]\n",
    "means[\"color\"] = means[[\"yellow\", \"blue\"]].idxmax(axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "sns.boxplot(x=\"color\", y=\"mean\", data=means)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.2275, 0.2725],\n",
       "       [0.2075, 0.2925]])"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "means[\"pred\"] = np.where(means[\"mean\"] > means[\"mean\"].mean(), \"yellow\", \"blue\")\n",
    "# confusion matrix between color and pred\n",
    "\n",
    "confusion_matrix(means[\"color\"], means[\"pred\"]) / len(means)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy = 0.52\n"
     ]
    }
   ],
   "source": [
    "print(f\"accuracy = {np.mean(means['color'] == means['pred'])}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "venv",
   "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}