[Fig07.a]-approximation-vary-dimension.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Plots for the paper"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from data_import import *\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.decomposition import PCA\n",
    "from plotting import *\n",
    "\n",
    "device = 'cuda:0'\n",
    "\n",
    "df_orig = load_result_table('feeds/phase_diagram_T64').reset_index()\n",
    "df_orig = df_orig[df_orig['epoch']==499]\n",
    "figure_dir = FIGURE_DIR \n",
    "\n",
    "df = df_orig.groupby(CONFIG_COLS).mean(numeric_only=True).reset_index()\n",
    "df_std = df_orig.groupby(CONFIG_COLS).std(numeric_only=True).reset_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                           name  _step    val_acc     _runtime  val_loss  \\\n",
      "40              rich-sound-1432  500.0  94.219702  1254.684850  0.153975   \n",
      "44           polished-surf-1428  500.0  98.059462  1223.191868  0.072469   \n",
      "48            hopeful-surf-1424  500.0  98.557657  1234.753695  0.076387   \n",
      "52             sleek-field-1420  500.0  96.463717  1239.446829  0.146970   \n",
      "55    resilient-wildflower-1417  500.0  88.762824  1222.627018  0.442263   \n",
      "59         usual-armadillo-1413  500.0  69.845176  1234.132040  0.916680   \n",
      "63           charmed-disco-1409  500.0  43.116927  1225.352758  1.476472   \n",
      "309          stellar-disco-1163  500.0  96.003515  1288.571615  0.121839   \n",
      "316         lucky-capybara-1156  500.0  97.509500  1295.309000  0.080772   \n",
      "322             absurd-dew-1150  500.0  97.943105  1294.767126  0.090064   \n",
      "327            desert-bush-1145  500.0  95.138204  1290.020306  0.165580   \n",
      "334           spring-totem-1138  500.0  87.317631  1294.435409  0.495906   \n",
      "340           eager-galaxy-1132  500.0  67.482429  1295.440318  0.893828   \n",
      "346          scarlet-brook-1126  500.0  39.582067  1288.954044  1.489249   \n",
      "610           quiet-glitter-862  500.0  96.167365  1303.564093  0.122082   \n",
      "616            magic-cherry-856  500.0  98.260830  1275.576500  0.063789   \n",
      "622             bright-bird-850  500.0  97.219321  1299.143044  0.097872   \n",
      "628              deft-cloud-844  500.0  96.687406  1293.775566  0.157305   \n",
      "634               soft-frog-838  500.0  90.448329  1299.176755  0.434249   \n",
      "640             fresh-dream-832  500.0  69.745917  1297.658052  0.902004   \n",
      "646             easy-puddle-826  500.0  39.997151  1290.584022  1.533567   \n",
      "911            fiery-shadow-561  500.0  95.073140  1295.442864  0.135849   \n",
      "917            robust-salad-555  500.0  98.513014  1287.671207  0.066158   \n",
      "923             rich-galaxy-549  500.0  98.445100  1288.856296  0.078608   \n",
      "930               sweet-bee-542  500.0  96.681707  1288.983872  0.157392   \n",
      "936         easy-wildflower-536  500.0  86.671734  1291.450492  0.465438   \n",
      "943          gallant-cosmos-529  500.0  65.840141  1281.286534  0.933058   \n",
      "949            elated-water-523  500.0  43.040939  1314.221984  1.486891   \n",
      "1214            hardy-paper-258  500.0  95.171924  1319.881178  0.139565   \n",
      "1220            young-music-252  500.0  98.310222  1308.063475  0.068069   \n",
      "1226        vivid-resonance-246  500.0  98.817915  1295.742202  0.062830   \n",
      "1233         lively-morning-239  500.0  93.114078  1312.145781  0.191254   \n",
      "1239             dark-dream-233  500.0  81.533056  1305.574404  0.465472   \n",
      "1245            worthy-lake-227  500.0  67.500476  1300.309975  0.928058   \n",
      "1251          silvery-disco-221  500.0  46.675532  1299.757892  1.440746   \n",
      "\n",
      "        _timestamp  train_loss  epoch             _wandb  params_count  ...  \\\n",
      "40    1.707092e+09    0.113537  499.0  {'runtime': 1254}       49931.0  ...   \n",
      "44    1.707090e+09    0.094780  499.0  {'runtime': 1222}       17163.0  ...   \n",
      "48    1.707089e+09    0.098125  499.0  {'runtime': 1234}        6923.0  ...   \n",
      "52    1.707088e+09    0.131703  499.0  {'runtime': 1239}        3339.0  ...   \n",
      "55    1.707087e+09    0.432904  499.0  {'runtime': 1221}        1931.0  ...   \n",
      "59    1.707086e+09    0.915238  499.0  {'runtime': 1233}        1323.0  ...   \n",
      "63    1.707084e+09    1.480421  499.0  {'runtime': 1225}        1043.0  ...   \n",
      "309   1.707029e+09    0.118336  499.0  {'runtime': 1288}       49931.0  ...   \n",
      "316   1.707027e+09    0.087130  499.0  {'runtime': 1295}       17163.0  ...   \n",
      "322   1.707026e+09    0.090736  499.0  {'runtime': 1294}        6923.0  ...   \n",
      "327   1.707025e+09    0.155344  499.0  {'runtime': 1289}        3339.0  ...   \n",
      "334   1.707024e+09    0.470607  499.0  {'runtime': 1294}        1931.0  ...   \n",
      "340   1.707022e+09    0.888931  499.0  {'runtime': 1295}        1323.0  ...   \n",
      "346   1.707021e+09    1.480387  499.0  {'runtime': 1288}        1043.0  ...   \n",
      "610   1.706965e+09    0.117284  499.0  {'runtime': 1303}       49931.0  ...   \n",
      "616   1.706963e+09    0.081949  499.0  {'runtime': 1275}       17163.0  ...   \n",
      "622   1.706962e+09    0.083208  499.0  {'runtime': 1298}        6923.0  ...   \n",
      "628   1.706961e+09    0.162690  499.0  {'runtime': 1293}        3339.0  ...   \n",
      "634   1.706960e+09    0.440267  499.0  {'runtime': 1298}        1931.0  ...   \n",
      "640   1.706958e+09    0.891797  499.0  {'runtime': 1297}        1323.0  ...   \n",
      "646   1.706957e+09    1.522254  499.0  {'runtime': 1290}        1043.0  ...   \n",
      "911   1.706901e+09    0.144350  499.0  {'runtime': 1291}       49931.0  ...   \n",
      "917   1.706899e+09    0.088007  499.0  {'runtime': 1287}       17163.0  ...   \n",
      "923   1.706898e+09    0.092967  499.0  {'runtime': 1288}        6923.0  ...   \n",
      "930   1.706897e+09    0.160632  499.0  {'runtime': 1288}        3339.0  ...   \n",
      "936   1.706896e+09    0.450332  499.0  {'runtime': 1291}        1931.0  ...   \n",
      "943   1.706894e+09    0.924093  499.0  {'runtime': 1282}        1323.0  ...   \n",
      "949   1.706893e+09    1.475886  499.0  {'runtime': 1313}        1043.0  ...   \n",
      "1214  1.706836e+09    0.131710  499.0  {'runtime': 1319}       49931.0  ...   \n",
      "1220  1.706835e+09    0.078565  499.0  {'runtime': 1307}       17163.0  ...   \n",
      "1226  1.706834e+09    0.079247  499.0  {'runtime': 1295}        6923.0  ...   \n",
      "1233  1.706832e+09    0.139767  499.0  {'runtime': 1311}        3339.0  ...   \n",
      "1239  1.706831e+09    0.427418  499.0  {'runtime': 1305}        1931.0  ...   \n",
      "1245  1.706830e+09    0.923856  499.0  {'runtime': 1300}        1323.0  ...   \n",
      "1251  1.706828e+09    1.443641  499.0  {'runtime': 1299}        1043.0  ...   \n",
      "\n",
      "      batch_size  no_softmax  num_samples  dataset_type       project_name  \\\n",
      "40            32       False        10000      backward  phase_diagram_T64   \n",
      "44            32       False        10000      backward  phase_diagram_T64   \n",
      "48            32       False        10000      backward  phase_diagram_T64   \n",
      "52            32       False        10000      backward  phase_diagram_T64   \n",
      "55            32       False        10000      backward  phase_diagram_T64   \n",
      "59            32       False        10000      backward  phase_diagram_T64   \n",
      "63            32       False        10000      backward  phase_diagram_T64   \n",
      "309           32       False        10000      backward  phase_diagram_T64   \n",
      "316           32       False        10000      backward  phase_diagram_T64   \n",
      "322           32       False        10000      backward  phase_diagram_T64   \n",
      "327           32       False        10000      backward  phase_diagram_T64   \n",
      "334           32       False        10000      backward  phase_diagram_T64   \n",
      "340           32       False        10000      backward  phase_diagram_T64   \n",
      "346           32       False        10000      backward  phase_diagram_T64   \n",
      "610           32       False        10000      backward  phase_diagram_T64   \n",
      "616           32       False        10000      backward  phase_diagram_T64   \n",
      "622           32       False        10000      backward  phase_diagram_T64   \n",
      "628           32       False        10000      backward  phase_diagram_T64   \n",
      "634           32       False        10000      backward  phase_diagram_T64   \n",
      "640           32       False        10000      backward  phase_diagram_T64   \n",
      "646           32       False        10000      backward  phase_diagram_T64   \n",
      "911           32       False        10000      backward  phase_diagram_T64   \n",
      "917           32       False        10000      backward  phase_diagram_T64   \n",
      "923           32       False        10000      backward  phase_diagram_T64   \n",
      "930           32       False        10000      backward  phase_diagram_T64   \n",
      "936           32       False        10000      backward  phase_diagram_T64   \n",
      "943           32       False        10000      backward  phase_diagram_T64   \n",
      "949           32       False        10000      backward  phase_diagram_T64   \n",
      "1214          32       False        10000      backward  phase_diagram_T64   \n",
      "1220          32       False        10000      backward  phase_diagram_T64   \n",
      "1226          32       False        10000      backward  phase_diagram_T64   \n",
      "1233          32       False        10000      backward  phase_diagram_T64   \n",
      "1239          32       False        10000      backward  phase_diagram_T64   \n",
      "1245          32       False        10000      backward  phase_diagram_T64   \n",
      "1251          32       False        10000      backward  phase_diagram_T64   \n",
      "\n",
      "     attention_input  offline_learning include_validation  entity  \\\n",
      "40          only_sem             False               True   feeds   \n",
      "44          only_sem             False               True   feeds   \n",
      "48          only_sem             False               True   feeds   \n",
      "52          only_sem             False               True   feeds   \n",
      "55          only_sem             False               True   feeds   \n",
      "59          only_sem             False               True   feeds   \n",
      "63          only_sem             False               True   feeds   \n",
      "309         only_sem             False               True   feeds   \n",
      "316         only_sem             False               True   feeds   \n",
      "322         only_sem             False               True   feeds   \n",
      "327         only_sem             False               True   feeds   \n",
      "334         only_sem             False               True   feeds   \n",
      "340         only_sem             False               True   feeds   \n",
      "346         only_sem             False               True   feeds   \n",
      "610         only_sem             False               True   feeds   \n",
      "616         only_sem             False               True   feeds   \n",
      "622         only_sem             False               True   feeds   \n",
      "628         only_sem             False               True   feeds   \n",
      "634         only_sem             False               True   feeds   \n",
      "640         only_sem             False               True   feeds   \n",
      "646         only_sem             False               True   feeds   \n",
      "911         only_sem             False               True   feeds   \n",
      "917         only_sem             False               True   feeds   \n",
      "923         only_sem             False               True   feeds   \n",
      "930         only_sem             False               True   feeds   \n",
      "936         only_sem             False               True   feeds   \n",
      "943         only_sem             False               True   feeds   \n",
      "949         only_sem             False               True   feeds   \n",
      "1214        only_sem             False               True   feeds   \n",
      "1220        only_sem             False               True   feeds   \n",
      "1226        only_sem             False               True   feeds   \n",
      "1233        only_sem             False               True   feeds   \n",
      "1239        only_sem             False               True   feeds   \n",
      "1245        only_sem             False               True   feeds   \n",
      "1251        only_sem             False               True   feeds   \n",
      "\n",
      "                project  \n",
      "40    phase_diagram_T64  \n",
      "44    phase_diagram_T64  \n",
      "48    phase_diagram_T64  \n",
      "52    phase_diagram_T64  \n",
      "55    phase_diagram_T64  \n",
      "59    phase_diagram_T64  \n",
      "63    phase_diagram_T64  \n",
      "309   phase_diagram_T64  \n",
      "316   phase_diagram_T64  \n",
      "322   phase_diagram_T64  \n",
      "327   phase_diagram_T64  \n",
      "334   phase_diagram_T64  \n",
      "340   phase_diagram_T64  \n",
      "346   phase_diagram_T64  \n",
      "610   phase_diagram_T64  \n",
      "616   phase_diagram_T64  \n",
      "622   phase_diagram_T64  \n",
      "628   phase_diagram_T64  \n",
      "634   phase_diagram_T64  \n",
      "640   phase_diagram_T64  \n",
      "646   phase_diagram_T64  \n",
      "911   phase_diagram_T64  \n",
      "917   phase_diagram_T64  \n",
      "923   phase_diagram_T64  \n",
      "930   phase_diagram_T64  \n",
      "936   phase_diagram_T64  \n",
      "943   phase_diagram_T64  \n",
      "949   phase_diagram_T64  \n",
      "1214  phase_diagram_T64  \n",
      "1220  phase_diagram_T64  \n",
      "1226  phase_diagram_T64  \n",
      "1233  phase_diagram_T64  \n",
      "1239  phase_diagram_T64  \n",
      "1245  phase_diagram_T64  \n",
      "1251  phase_diagram_T64  \n",
      "\n",
      "[35 rows x 28 columns]\n",
      "2 [4, 8, 32]\n",
      "4 [4, 8, 32]\n",
      "feeds/phase_diagram_T64/gallant-cosmos-529-model:v0 | val_acc: 65.840\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m:   1 of 1 files downloaded.  \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "8 [4, 8, 32]\n",
      "feeds/phase_diagram_T64/dark-dream-233-model:v0 | val_acc: 81.533\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m:   1 of 1 files downloaded.  \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "16 [4, 8, 32]\n",
      "32 [4, 8, 32]\n",
      "feeds/phase_diagram_T64/bright-bird-850-model:v0 | val_acc: 97.219\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m:   1 of 1 files downloaded.  \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "64 [4, 8, 32]\n",
      "128 [4, 8, 32]\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 900x600 with 6 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from memory_exp import HistogramDatasetSampleBackward, HistogramDatasetSampleForward, HistogramDatasetSampleBackwardWithBOS\n",
    "from data_import import *\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.decomposition import PCA\n",
    "from memory_exp import *\n",
    "import matplotlib as mpl\n",
    "\n",
    "bright_distinguishable_colors = plt.cm.tab20.colors\n",
    "bright_distinguishable_colors = {d: c for c, d in zip(bright_distinguishable_colors,df_orig.model_dim.unique())}\n",
    "\n",
    "fig, axes = plt.subplots(2,3,figsize=(9,6))\n",
    "model_config = MODELS['dot+sftm']\n",
    "k = df_orig\n",
    "p = 64\n",
    "T = 64\n",
    "a = k[(k.attention_input == model_config['attention_input'])\n",
    "            & (k.no_softmax == model_config['no_softmax']) \n",
    "            & (k.dataset_type == model_config['dataset'])\n",
    "            & (k.p == p)]\n",
    "print(a)\n",
    "ds = [4,8,32]\n",
    "ax_idx_d = {d: i for i, d in enumerate(ds)}\n",
    "for d, g in a.reset_index().groupby('model_dim'):\n",
    "    print(d,ds)\n",
    "    if d not in ds:\n",
    "        continue\n",
    "    idx = ax_idx_d[d]\n",
    "    # find the row with the highest validation accuracy\n",
    "    g = g.set_index('name').sort_values('val_acc',ascending=True)\n",
    "    config = g.iloc[0]\n",
    "    model, word_embeddings = load_model(config)\n",
    "    dataset_cls = {'forward': HistogramDatasetSampleForward,\n",
    "                    'backward': HistogramDatasetSampleBackward,\n",
    "                    'backward_BOS': HistogramDatasetSampleBackwardWithBOS}[config['dataset_type']]\n",
    "    val_dataset = dataset_cls(config['seq_len'], config['T'], int(0.3 * config['num_samples']))\n",
    "    train_dataloader = DataLoader(val_dataset, batch_size=32, shuffle=False)\n",
    "    criterion = nn.CrossEntropyLoss()\n",
    "    n_classes = config['seq_len']+1\n",
    "    activ = []\n",
    "    ys = []\n",
    "    outputs = []\n",
    "    attn_probs = []\n",
    "    mixeds = []\n",
    "    Xs = []\n",
    "    preds = []\n",
    "    acc = 0\n",
    "    val_loss = 0\n",
    "    counter = word_embeddings.sum(axis=0)/T\n",
    "    for X, y in train_dataloader:\n",
    "        X = X.to(device)\n",
    "        y = y.to(device)\n",
    "        output = model(X)\n",
    "        \n",
    "        # compute accuracy\n",
    "        pred = output.argmax(axis=-1)\n",
    "        preds.append(pred.detach().cpu().numpy())\n",
    "        ys.append(y.detach().cpu().numpy())\n",
    "        Xs.append(X.detach().cpu().numpy())\n",
    "\n",
    "    preds = np.concatenate(preds).flatten()\n",
    "    ys = np.concatenate(ys).flatten()\n",
    "    \n",
    "    confusion = np.zeros((config['seq_len']+1,config['seq_len']+1))\n",
    "    for i in range(len(ys)):\n",
    "        confusion[ys[i],preds[i]] += 1\n",
    "\n",
    "    im = axes[0,idx].imshow(confusion/confusion.sum(), origin='lower', cmap='binary',norm=mpl.colors.LogNorm())\n",
    "    \n",
    "    \n",
    "    axes[0,idx].set_xlim(0.5,config['seq_len']-0.5)\n",
    "    axes[0,idx].set_ylim(0.5,config['seq_len']-0.5)\n",
    "    axes[0,idx].set_xticks(range(1, config['seq_len']))\n",
    "    \n",
    "    \n",
    "    \n",
    "    word_embeddings = word_embeddings #/ torch.norm(word_embeddings,dim=-1).unsqueeze(-1)\n",
    "    outer = word_embeddings @ word_embeddings.T\n",
    "\n",
    "    outer = outer.flatten()\n",
    "    mask = (np.eye(word_embeddings.shape[0]) == 0).flatten()\n",
    "    axes[1,idx].hist(outer[mask].detach().numpy().flatten(),bins=100,histtype='step',color='tab:orange',density=True)\n",
    "    axes[1,idx].hist(outer[~mask].detach().numpy().flatten(),bins=30,histtype='step',color='tab:blue',density=True)\n",
    "    axes[1,idx].axvline(outer[mask].median().item(),color='tab:orange')\n",
    "    axes[1,idx].axvline(outer[~mask].median().item(),color='tab:blue')\n",
    "    axes[1,idx].set_xlabel('overlap')\n",
    "    \n",
    "    \n",
    "    axes[0,idx].text(1, 8, f'${config[\"val_acc\"]:.2f}\\%$ acc.')\n",
    "    axes[0,idx].set_title(f'$d={d}$')\n",
    "    axes[0,0].set_xlabel('prediction')\n",
    "\n",
    "axes[0,0].set_ylabel('ground truth')\n",
    "axes[1,0].set_ylabel('density')\n",
    "    \n",
    "\n",
    "    \n",
    "from mpl_toolkits.axes_grid1.inset_locator import inset_axes\n",
    "cax = axes[0,-1].inset_axes([0.63,0.05,0.04,0.4])\n",
    "\n",
    "# adjust the margin between the subplots to be smaller\n",
    "fig.subplots_adjust(wspace=0.3,hspace=0.3)\n",
    "# Add a colorbar to the inset_axes\n",
    "cbar = plt.colorbar(im, cax=cax, label='density')\n",
    "axes[1,0].plot([],[],label=r'same - $\\langle e_t, e_t \\rangle$',color='tab:blue')\n",
    "axes[1,0].plot([],[],label=r'different - $\\langle e_t, e_s \\rangle$',color='tab:orange')\n",
    "axes[1,0].legend()\n",
    "plt.savefig(figure_dir / 'changing-d-for-fixed-p-dotsftm_final.pdf',bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}