{ "cells": [ { "cell_type": "markdown", "metadata": { "_cell_guid": "878625c3-c4b9-044d-dba1-63712f7555e1" }, "source": [ "This notebook just uses the four label columns and uses Owen Zhangs leave one out encoding to get an LB of ~0.66. He explains it better than me so follow the youtube [link][1] and look around the 59 minute mark! \n", "\n", "\n", " [1]: https://www.youtube.com/watch?v=LgLcfZjNF44" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "a5660463-c512-97ea-40c3-48c4c48d748a" }, "outputs": [], "source": [ "import numpy as np\n", "import pandas as pd\n", "import xgboost as xgb\n", "from sklearn.preprocessing import LabelEncoder" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "40000e68-c2e2-c422-25f5-38afce19c42f" }, "outputs": [], "source": [ "def LeaveOneOut(data1, data2, columnName, useLOO=False, addNoise=False):\n", " grpOutcomes = data1.groupby(columnName)['interest_level'].mean().reset_index()\n", " grpCount = data1.groupby(columnName)['interest_level'].count().reset_index()\n", " grpOutcomes['cnt'] = grpCount.interest_level\n", " if(useLOO):\n", " grpOutcomes = grpOutcomes[grpOutcomes.cnt > 4]\n", " grpOutcomes.drop('cnt', inplace=True, axis=1)\n", " outcomes = data2['interest_level'].values\n", " x = pd.merge(data2[[columnName, 'interest_level']], grpOutcomes,\n", " suffixes=('x_', ''),\n", " how='left',\n", " on=columnName,\n", " left_index=True)['interest_level']\n", " if(useLOO):\n", " x = ((x*x.shape[0])-outcomes)/(x.shape[0]-1)\n", " if(addNoise):\n", " x = x + np.random.normal(0, .25, x.shape[0])\n", " return x.fillna(x.mean())\n", "\n", "def runXGB(train_X, train_y, test_X, test_y=None, feature_names=None, seed_val=0, num_rounds=1000):\n", " param = {}\n", " param['objective'] = 'multi:softprob'\n", " param['eta'] = 0.03\n", " param['max_depth'] = 4\n", " param['silent'] = 1\n", " param['num_class'] = 3\n", " param['eval_metric'] = \"mlogloss\"\n", " param['min_child_weight'] = 1\n", " param['subsample'] = 0.7\n", " param['colsample_bytree'] = 0.6\n", " param['seed'] = seed_val\n", " param['seed'] = seed_val\n", " num_rounds = num_rounds\n", "\n", " plst = list(param.items())\n", " xgtrain = xgb.DMatrix(train_X, label=train_y)\n", "\n", " if test_y is not None:\n", " xgtest = xgb.DMatrix(test_X, label=test_y)\n", " watchlist = [ (xgtrain,'train'), (xgtest, 'test') ]\n", " model = xgb.train(plst, xgtrain, num_rounds, watchlist, early_stopping_rounds=20)\n", " else:\n", " xgtest = xgb.DMatrix(test_X)\n", " model = xgb.train(plst, xgtrain, num_rounds)\n", "\n", " pred_test_y = model.predict(xgtest)\n", " return pred_test_y, model" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "0c9d98a4-17a5-7f48-ca38-cdf8b0310c91" }, "outputs": [], "source": [ "labels = ['building_id','display_address','street_address','manager_id']\n", "train = pd.read_json('../input/train.json')\n", "test = pd.read_json('../input/test.json')\n", "train.loc[train.interest_level=='low','interest_level'] = 0\n", "train.loc[train.interest_level=='medium','interest_level'] = 1\n", "train.loc[train.interest_level=='high','interest_level'] = 2\n", "train.interest_level = train.interest_level.astype(float)\n", "test['interest_level'] = -1" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "52ab21e6-b1f0-45b3-46f7-ecd09906ac27" }, "outputs": [], "source": [ "for f in labels:\n", " lbl = LabelEncoder()\n", " lbl.fit(list(train[f].values) + list(test[f].values))\n", " train[f] = lbl.transform(train[f].values).astype(int)\n", " test[f] = lbl.transform(test[f].values).astype(int)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "46635c78-16a5-2b41-a26c-b4fbcf39b8ae" }, "outputs": [], "source": [ "actualcolumns = []\n", "for col in labels: \n", " for lvl in range(3):\n", " kftrain = train.copy()\n", " kftrain.interest_level = (kftrain.interest_level==lvl).astype(int)\n", " train['loo_'+col+'_'+str(lvl)] = LeaveOneOut(kftrain.copy(),\n", " train.copy(),\n", " col, True, True).values\n", " test['loo_'+col+'_'+str(lvl)] = LeaveOneOut(kftrain.copy(),\n", " test.copy(),\n", " col, True, False).values\n", " actualcolumns.extend(['loo_'+col+'_'+str(lvl)])" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "8c18f0c1-7490-a55f-a84d-c11771ad6827" }, "outputs": [], "source": [ "preds, model = runXGB(train[actualcolumns],\n", " train.interest_level,\n", " test[actualcolumns],\n", " None,\n", " feature_names=None,\n", " seed_val=0,\n", " num_rounds=500)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "bec481d9-cb1c-5a7d-d945-80a5fa14b953" }, "outputs": [], "source": [ "out_df = pd.DataFrame(preds)\n", "out_df.columns = [\"low\", \"medium\", \"high\" ]\n", "out_df[\"listing_id\"] = test.listing_id.values\n", "out_df = out_df[['high', 'medium', 'low','listing_id']]\n", "out_df.to_csv(\"loo_xgb_starter.csv\", index=False)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "8c064201-5ec5-4b0d-5a8a-548b59642ecc" }, "outputs": [], "source": [ "print(out_df[['low','medium','high']].mean())" ] } ], "metadata": { "_change_revision": 0, "_is_fork": false, "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": "3.6.0" } }, "nbformat": 4, "nbformat_minor": 0 }