notebooks/plot_SGD_gradient_analysis.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "392d0571",
   "metadata": {},
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "Grouped-barplot comparing gradient cosine-similarity of current batch with previous points k steps in history of the learning trajectory.\n",
    "\n",
    "This is obtained by running experiments and setting in the config:\n",
    "    cfg.METHOD.ANALYZE_GRADS_WINDOW = True  # Compare current grad with grad in prev\n",
    "    cfg.METHOD.MAX_ANALYZE_GRADS_WINDOW_SIZE = 10 # Window size to look back at\n",
    "\n",
    "Download the results in a CSV from WandB of these experiments, and make sure the '*_grad_cos_sim*' metrics are included.\n",
    "These results are then visualized in this notebook.\n",
    "\"\"\"\n",
    "import datetime\n",
    "import os\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "from IPython.core.interactiveshell import InteractiveShell\n",
    "\n",
    "InteractiveShell.ast_node_interactivity = \"all\"\n",
    "pd.set_option('display.max_rows', 1000)\n",
    "\n",
    "plt.rcParams['font.family'] = 'DeJavu Serif'\n",
    "plt.rcParams['font.serif'] = ['Times New Roman']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "102f55d5",
   "metadata": {},
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "CONFIG: Add your config params here\n",
    "\"\"\"\n",
    "\n",
    "########### READ RESULTS CONFIG ###############\n",
    "# Open wandb downloaded csv (e.g. from GUI)\n",
    "csv_path = '/your/path/to/wandb_GUI_downloaded_csv/wandb_export_2022-10-23T15_13_53.325-07_00.csv'\n",
    "\n",
    "# For which model parts we check the gradients from\n",
    "model_parts = [\"full\", \"slow\", \"fast\", \"head\", \"feat\", ]\n",
    "\n",
    "########### OUTPUT CONFIG ###############\n",
    "title = \"SGD_GRAD_ANALYSIS\"\n",
    "main_outdir = f\"../imgs/{title}\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [
    "\"\"\" Get dataframe. \"\"\"\n",
    "orig_df = pd.read_csv(csv_path)\n",
    "\n",
    "# Reformat cols to have per metric, the SE and mean in separate columns, and 1->10 in rows, take 'steps' as another column\n",
    "df_dict_list = []\n",
    "for nb_steps_lookback in range(1, 11):  # 1 row per step\n",
    "    single_df_row_dict = {'step': nb_steps_lookback, }\n",
    "\n",
    "    for model_part in model_parts:\n",
    "        mean = orig_df[\n",
    "            f\"adhoc_users_aggregate/analyze_action_batch/LOOKBACK_STEP_{nb_steps_lookback}/{model_part}_grad_cos_sim/mean\"].to_list()[0]\n",
    "        SE = orig_df[\n",
    "            f\"adhoc_users_aggregate/analyze_action_batch/LOOKBACK_STEP_{nb_steps_lookback}/{model_part}_grad_cos_sim/SE\"].to_list()[0]\n",
    "\n",
    "        single_df_row_dict[f\"adhoc_users_aggregate/analyze_action_batch/{model_part}_grad_cos_sim/mean\"] = mean\n",
    "        single_df_row_dict[f\"adhoc_users_aggregate/analyze_action_batch/{model_part}_grad_cos_sim/SE\"] = SE\n",
    "\n",
    "    # Update\n",
    "    df_dict_list.append(single_df_row_dict)\n",
    "\n",
    "transformed_df = pd.DataFrame(df_dict_list)\n",
    "transformed_df.set_index('step',inplace=True) # Set step as index (for plotting later)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7cf9b514",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "\"\"\" Plot. \"\"\"\n",
    "plot_df = transformed_df\n",
    "GROUPS_TO_PLOT={ # Map to y-label\n",
    "    'adhoc_users_aggregate/analyze_action_batch/full_grad_cos_sim':r'$F \\circ H$',\n",
    "    'adhoc_users_aggregate/analyze_action_batch/feat_grad_cos_sim':r'$F$',\n",
    "    'adhoc_users_aggregate/analyze_action_batch/head_grad_cos_sim':r'$H$',\n",
    "}\n",
    "\n",
    "sns.set(font_scale=1.6)\n",
    "plt.rcParams['font.family'] = 'DeJavu Serif'\n",
    "plt.rcParams['font.serif'] = ['Times New Roman']\n",
    "plt.rcParams['font.size'] = 18\n",
    "\n",
    "# Use latex in mpl\n",
    "plt.rcParams['text.usetex'] = True\n",
    "plt.rcParams['text.latex.preamble'] = r'\\usepackage{amsmath}' #for \\text command\n",
    "\n",
    "# Save config\n",
    "parent_dirname = datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S') + \"_\" + title\n",
    "parent_dirpath = os.path.join(main_outdir, parent_dirname)\n",
    "\n",
    "# Plot configs\n",
    "plot_config = {\n",
    "    \"color\": 'royalblue',\n",
    "    \"dpi\": 600,\n",
    "    \"figsize\": (8,4),\n",
    "    \"xlabel\": \"re-exposure iterations\",\n",
    "    \"ylabel\": \"\",\n",
    "    \"title\": None\n",
    "}\n",
    "\n",
    "colors = [\n",
    "    sns.color_palette(\"rocket\")[1],\n",
    "    sns.color_palette(\"rocket\")[3],\n",
    "    sns.color_palette(\"rocket\")[5],\n",
    "]\n",
    "\n",
    "yerr_names = []\n",
    "y_names = []\n",
    "legend_names = []\n",
    "for name, legend_name in GROUPS_TO_PLOT.items():\n",
    "    yerr_names.append(f\"{name}/SE\")\n",
    "    y_names.append(f\"{name}/mean\")\n",
    "    legend_names.append(legend_name)\n",
    "\n",
    "yerr = plot_df[yerr_names].to_numpy().T\n",
    "\n",
    "fig, ax = plt.subplots(figsize=plot_config['figsize'])\n",
    "plot_df[y_names].plot(kind='bar', yerr=yerr, alpha=1, error_kw=dict(ecolor='k'),ax=ax,capsize=2,rot=0, color=colors)\n",
    "\n",
    "ax.legend([r\"$F \\circ H$\",r\"$F$\",r\"$H$\"])\n",
    "ax.set(xlabel=r'$k$', ylabel=r'$\\cos_\\angle (g_t, \\  g_{t-k})$')\n",
    "ax.legend(legend_names,loc='lower right',ncol=1)\n",
    "\n",
    "plt.tight_layout()\n",
    "filename = f'{title}.pdf'\n",
    "filepath = os.path.join(parent_dirpath,filename)\n",
    "print(f\"Saving at: {filepath}\")\n",
    "os.makedirs(parent_dirpath, exist_ok=True)\n",
    "fig.savefig(filepath)\n",
    "\n",
    "plt.show()\n",
    "plt.close('all')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a93346e7",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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
}