analyze_nn_weights.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 183,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The autoreload extension is already loaded. To reload it, use:\n",
      "  %reload_ext autoreload\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import matplotlib\n",
    "matplotlib.use('Agg')\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.backends.backend_pdf import PdfPages\n",
    "import seaborn as sns\n",
    "%matplotlib inline\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "from scripts.parsing_utils import *\n",
    "from utils.utils import *\n",
    "from scripts.parse_results import *\n",
    "import torch\n",
    "from utils.net import *\n",
    "import klepto"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 187,
   "metadata": {},
   "outputs": [],
   "source": [
    "#RES_DIR = \"./debug/nested_loop_index7-mscn-df:10-nn:1:128-loss:mse-0.0--D0.1-831/\"\n",
    "RES_DIR = \"/home/pari/learned-cardinalities/all_results/inl_fixed_scan_ops/nested_loop_index7/test_diff/subq_imp2/diff_partitions-nested_loop_index7-mscn-df:10-nn:1:256-loss:mse-0.0--D0.1-926/\"\n",
    "RES_DIR = \"/home/pari/learned-cardinalities/all_results/inl_fixed_scan_ops/\"\n",
    "#RES_DIR = \"./debug/\"\n",
    "WEIGHT_FN = \"model_weights.pt\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 216,
   "metadata": {},
   "outputs": [],
   "source": [
    "def viz_weight_means(weights, kind):\n",
    "    flow_weights = weights[kind].cpu().numpy()\n",
    "    #flow_weights = np.maximum(flow_weights, 0.0)\n",
    "    flow_weights = np.mean(flow_weights, axis=0)\n",
    "    sns.barplot(y=flow_weights)\n",
    "    plt.title(kind)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 189,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Namespace(adaptive_lr=0, adaptive_lr_patience=20, algs='nn', avg_factor=1, avg_jl_num_last=5, avg_jl_priority=1, baseline_join_alg='EXHAUSTIVE', cache_dir='./caches/', cl_recompute=0, clip_gradient=20.0, compute_runtime=0, cost_model='nested_loop_index7', cost_model_plan_err=1, data_dir='/data/pari/cards/', db_file_name=None, db_host='localhost', db_name='imdb', debug_set=0, diff_templates_seed=2, diff_templates_type=3, dropout=0.0, eval_epoch=10, eval_epoch_flow_err=1, eval_epoch_jerr=1, eval_epoch_plan_err=1, eval_flow_loss=1, eval_num_tables=0, eval_on_job=1, eval_parallel=0, eval_test_while_training=1, execution_cache_threshold=20, exp_name='card_exp', exp_prefix='diff_partitions', feat_pg_costs=1, feat_pg_est_one_hot=0, feat_pg_path=1, feat_rel_pg_ests=1, feat_rel_pg_ests_onehot=0, feat_tolerance=1, flow_features=0, gen_bn_dist=0, gen_synth_data=0, group_models=0, heuristic_features=1, hidden_layer_multiple=None, hidden_layer_size=256, jl_indexes=1, jl_use_postgres=1, jl_variant=0, job_query_dir='./job_queries/', join_features=1, join_loss_data_file=None, join_loss_pool_num=60, load_query_together=0, loss_func='mse', losses='qerr,join-loss,flow-loss,plan-loss', lr=0.0001, max_discrete_featurizing_buckets=10, max_epochs=20, max_hid=128, min_corr=0.0, net_name='FCNN', nn_type='mscn', nn_weights_init_pg=0, no7a=0, normalization_type='mscn', normalize_flow_loss=1, num_bins=100, num_groups=1, num_hidden_layers=1, num_samples_per_template=-1, num_singular_vals=5, num_tables_feature=1, only_nonzero_samples=1, optimizer_name='adamw', pgm_alg_name='chow-liu', pgm_backend='ourpgm', pgm_merge_aliases=0, pgm_sampling_percentage=0.001, port=5432, pred_features=1, preload_features=1, prioritize_epoch=1, priority_err_type='jerr', priority_normalize_type='', priority_query_len_scale=0, pwd='', query_batch_size=1, query_directory='./our_dataset/queries', query_templates='all', random_seed=2112, rel_jloss=0, rel_qerr_loss=0, reprioritize_epoch=1, result_dir='all_results/inl_fixed_scan_ops/nested_loop_index7/test_diff/subq_imp2', rf_trees=128, sampling_key=None, sampling_percentage=10, sampling_priority_alpha=0.0, sampling_priority_type='query', sampling_type='ss', save_cur_cache_dir=None, save_gradients=1, single_threaded_nt=0, start_validation=5, synth_num_columns=2, synth_num_vals=100000, synth_period_len=10, synth_table='test', table_features=1, template_dir=None, test=1, test_diff_templates=1, test_size=0.5, tfboard=1, train_card_key='actual', use_best_val_model=1, use_subqueries=1, use_svd=0, use_val_set=0, user='', viz_fn='./test', viz_join_plans=0, weight_decay=0.1, weighted_mse=0.0, weighted_qloss=0)\n"
     ]
    }
   ],
   "source": [
    "args = load_object(RES_DIR + \"/args.pkl\")\n",
    "print(args)\n",
    "weights = torch.load(RES_DIR + WEIGHT_FN)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 190,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "collections.OrderedDict"
      ]
     },
     "execution_count": 190,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 191,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "odict_keys(['sample_mlp1.0.weight', 'sample_mlp1.0.bias', 'sample_mlp2.0.weight', 'sample_mlp2.0.bias', 'predicate_mlp1.0.weight', 'predicate_mlp1.0.bias', 'predicate_mlp2.0.weight', 'predicate_mlp2.0.bias', 'join_mlp1.0.weight', 'join_mlp1.0.bias', 'join_mlp2.0.weight', 'join_mlp2.0.bias', 'out_mlp1.0.weight', 'out_mlp1.0.bias', 'out_mlp2.0.weight', 'out_mlp2.0.bias'])"
      ]
     },
     "execution_count": 191,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "weights.keys()\n",
    "# TODO: remove negative values in analysis as ReLU will remove them too"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 192,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "836\n"
     ]
    }
   ],
   "source": [
    "print(weights[\"predicate_mlp1.0.weight\"].shape[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 193,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "256"
      ]
     },
     "execution_count": 193,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "args.hidden_layer_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 197,
   "metadata": {},
   "outputs": [],
   "source": [
    "if args.flow_features:\n",
    "    net = SetConv(weights[\"sample_mlp1.0.weight\"].shape[1], weights[\"predicate_mlp1.0.weight\"].shape[1], \n",
    "             weights[\"join_mlp1.0.weight\"].shape[1], weights[\"flow_mlp1.0.weight\"].shape[1], args.hidden_layer_size)\n",
    "else:\n",
    "    net = SetConv(weights[\"sample_mlp1.0.weight\"].shape[1], weights[\"predicate_mlp1.0.weight\"].shape[1], \n",
    "        weights[\"join_mlp1.0.weight\"].shape[1], 0, args.hidden_layer_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SetConv(\n",
       "  (sample_mlp1): Sequential(\n",
       "    (0): Linear(in_features=21, out_features=256, bias=True)\n",
       "    (1): ReLU()\n",
       "  )\n",
       "  (sample_mlp2): Sequential(\n",
       "    (0): Linear(in_features=256, out_features=256, bias=True)\n",
       "    (1): ReLU()\n",
       "  )\n",
       "  (predicate_mlp1): Sequential(\n",
       "    (0): Linear(in_features=836, out_features=256, bias=True)\n",
       "    (1): ReLU()\n",
       "  )\n",
       "  (predicate_mlp2): Sequential(\n",
       "    (0): Linear(in_features=256, out_features=256, bias=True)\n",
       "    (1): ReLU()\n",
       "  )\n",
       "  (join_mlp1): Sequential(\n",
       "    (0): Linear(in_features=129, out_features=256, bias=True)\n",
       "    (1): ReLU()\n",
       "  )\n",
       "  (join_mlp2): Sequential(\n",
       "    (0): Linear(in_features=256, out_features=256, bias=True)\n",
       "    (1): ReLU()\n",
       "  )\n",
       "  (out_mlp1): Sequential(\n",
       "    (0): Linear(in_features=768, out_features=256, bias=True)\n",
       "    (1): ReLU()\n",
       "  )\n",
       "  (out_mlp2): Sequential(\n",
       "    (0): Linear(in_features=256, out_features=1, bias=True)\n",
       "    (1): Sigmoid()\n",
       "  )\n",
       "  (drop_layer): Dropout(p=0.0, inplace=False)\n",
       ")"
      ]
     },
     "execution_count": 198,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 199,
   "metadata": {},
   "outputs": [],
   "source": [
    "misc_cache = klepto.archives.dir_archive(\"./misc_cache\",\n",
    "            cached=True, serialized=True)\n",
    "db_key = deterministic_hash(\"db-\" + args.query_directory + \\\n",
    "                args.query_templates + str(args.eval_on_job))\n",
    "assert db_key in misc_cache.archive\n",
    "db = misc_cache.archive[db_key]\n",
    "db.init_featurizer(num_tables_feature = args.num_tables_feature,\n",
    "            max_discrete_featurizing_buckets =\n",
    "            args.max_discrete_featurizing_buckets,\n",
    "            heuristic_features = args.heuristic_features,\n",
    "            flow_features = args.flow_features,\n",
    "            feat_pg_costs = args.feat_pg_costs,\n",
    "            feat_pg_path = args.feat_pg_path,\n",
    "            feat_rel_pg_ests = args.feat_rel_pg_ests,\n",
    "            feat_rel_pg_ests_onehot = args.feat_rel_pg_ests_onehot,\n",
    "            feat_pg_est_one_hot = args.feat_pg_est_one_hot,\n",
    "            feat_tolerance = args.feat_tolerance,\n",
    "            cost_model = args.cost_model)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 200,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[833, 834, 835]\n"
     ]
    }
   ],
   "source": [
    "pred_maps = db.get_pred_features_map()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 201,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "835\n",
      "(256, 836)\n"
     ]
    }
   ],
   "source": [
    "print(max(pred_maps[\"idx\"]))\n",
    "pred_weights = weights[\"predicate_mlp1.0.weight\"].cpu().numpy()\n",
    "print(pred_weights.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 202,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>idx</th>\n",
       "      <th>descr</th>\n",
       "      <th>continuous</th>\n",
       "      <th>pg</th>\n",
       "      <th>feature_type</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>t.production_year-lt</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>t.production_year-like</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>t.production_year-nlike</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>t.production_year-in</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>t.production_year-eq</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>7</td>\n",
       "      <td>t.production_year-pg</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>5</td>\n",
       "      <td>t.production_year-range</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>6</td>\n",
       "      <td>t.production_year-range</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>8</td>\n",
       "      <td>t.title-lt</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>9</td>\n",
       "      <td>t.title-like</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>pred</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   idx                    descr  continuous     pg feature_type\n",
       "0    0     t.production_year-lt        True  False         pred\n",
       "1    1   t.production_year-like        True  False         pred\n",
       "2    2  t.production_year-nlike        True  False         pred\n",
       "3    3     t.production_year-in        True  False         pred\n",
       "4    4     t.production_year-eq        True  False         pred\n",
       "5    7     t.production_year-pg        True   True         pred\n",
       "6    5  t.production_year-range        True  False         pred\n",
       "7    6  t.production_year-range        True  False         pred\n",
       "8    8               t.title-lt       False  False         pred\n",
       "9    9             t.title-like       False  False         pred"
      ]
     },
     "execution_count": 202,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_maps.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 208,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred_weights = weights[\"predicate_mlp1.0.weight\"].cpu().numpy()\n",
    "#pred_weights = np.maximum(pred_weights, 0.0)\n",
    "mean_weights = np.mean(pred_weights, axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 219,
   "metadata": {},
   "outputs": [],
   "source": [
    "# FIXME: this should map exactly?\n",
    "pred_maps[\"weight\"] = pred_maps.apply(lambda row: mean_weights[row[\"idx\"]], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 221,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.barplot(x=\"pg\", y=\"weight\", data=pred_maps)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 222,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "if args.flow_features:\n",
    "    viz_weight_means(weights, kind=\"flow_mlp1.0.weight\")\n",
    "viz_weight_means(weights, kind=\"sample_mlp1.0.weight\")\n",
    "viz_weight_means(weights, kind=\"join_mlp1.0.weight\")\n",
    "viz_weight_means(weights, kind=\"predicate_mlp1.0.weight\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 223,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "if args.flow_features:\n",
    "    viz_weight_means(weights, kind=\"flow_mlp1.0.bias\")\n",
    "viz_weight_means(weights, kind=\"sample_mlp1.0.bias\")\n",
    "viz_weight_means(weights, kind=\"join_mlp1.0.bias\")\n",
    "viz_weight_means(weights, kind=\"predicate_mlp1.0.bias\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 224,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n"
     ]
    }
   ],
   "source": [
    "print(args.flow_features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "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": "3.8.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}