{ "cells": [ { "cell_type": "markdown", "metadata": { "_cell_guid": "b80ef5b5-3590-f283-cced-6964a306ef59" }, "source": [ "Just a little look at how prices impact interest_level" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "fae6b375-b481-3c14-85c6-db6653472751" }, "outputs": [], "source": [ "# This Python 3 environment comes with many helpful analytics libraries installed\n", "# It is defined by the kaggle/python docker image: https://github.com/kaggle/docker-python\n", "# For example, here's several helpful packages to load in \n", "\n", "import numpy as np # linear algebra\n", "import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n", "\n", "# Input data files are available in the \"../input/\" directory.\n", "# For example, running this (by clicking run or pressing Shift+Enter) will list the files in the input directory\n", "\n", "from subprocess import check_output\n", "print(check_output([\"ls\", \"../input\"]).decode(\"utf8\"))\n", "\n", "# Any results you write to the current directory are saved as output." ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "d4d2ca3e-17eb-5be5-7f96-08f3c9c3d023" }, "source": [ "## Some useful imports" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "f174d72c-e8c3-eb05-f3ca-75025364dc1b" }, "outputs": [], "source": [ "import seaborn as sns\n", "import matplotlib.pyplot as plt\n", "%matplotlib inline" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "18dfccae-6749-96e3-d0a3-2a8978ce08dc" }, "source": [ "## Read data" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "cf1f67a8-5975-b72e-dab6-3f0669ff1541" }, "outputs": [], "source": [ "train = pd.read_json('../input/train.json').set_index('listing_id')\n", "test = pd.read_json('../input/test.json').set_index('listing_id')" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "662c19ea-9d94-d0a9-2dc1-295d833a53f9" }, "source": [ "## Scatter plot: price vs bedrooms colored by interest_level " ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "4c6e02c0-c2a0-fb22-d1cb-bc4ee64faf65" }, "outputs": [], "source": [ "# limit number of bedrooms and prices\n", "usable_train = train[(train.price<10000) & (train.bedrooms<=4)]\n", "palette = {\"high\": \"r\", \"low\":\"g\", \"medium\":\"orange\"}\n", "plt.figure(figsize=(11,10))\n", "for interest in ['low', 'medium', 'high']:\n", " plt.scatter(usable_train[usable_train.interest_level==interest].bedrooms, \n", " usable_train[usable_train.interest_level==interest].price, \n", " c=palette[interest])" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "2b064012-2658-0d88-7c1a-f201a05478b6" }, "source": [ "High interest goes for low prices" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "28e70612-c7c6-4186-bbcd-d7e58ad372d0" }, "source": [ "## Boxplot gives even better insights " ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "eb7ca231-ac9f-96b4-8b80-896ea0bfe13f" }, "outputs": [], "source": [ "plt.figure(figsize=(11,10))\n", "sns.boxplot(x=\"bedrooms\", y=\"price\", hue=\"interest_level\", data=usable_train, palette=palette)" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "2efea442-9278-7139-86ca-3dde83ef7eb0" }, "source": [ "## How can we use this ?" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "bf71c80b-242b-bdae-95b3-d34c6ba94bac" }, "outputs": [], "source": [ "def add_median_price(key=None, suffix=\"\", trn_df=None, tst_df=None):\n", " \"\"\"\n", " Compute median prices for renthop dataset.\n", " The function adds 2 columns to the pandas DataFrames : the median prices and a ratio\n", " between nthe actual price of the rent and the median\n", " \n", " :param key: list of columns on which to groupby and compute median prices\n", " :param suffix: string used to suffix the newly created columns/features\n", " :param trn_df: training dataset as a pandas DataFrame\n", " :param tst_df: test dataset as a pandas DataFrame\n", " :return: updated train and test DataFrames\n", "\n", " :Example\n", " \n", " train, test = add_median_price(key=['bedrooms', 'bathrooms'], \n", " suffix='rooms', \n", " trn_df=train, \n", " tst_df=test)\n", "\n", " \"\"\"\n", " # Set features to be used\n", " median_features = key.copy()\n", " median_features.append('price')\n", " # Concat train and test to find median prices over whole dataset\n", " median_prices = pd.concat([trn_df[median_features], tst_df[median_features]], axis=0)\n", " # Group data by key to compute median prices\n", " medians_by_key = median_prices.groupby(by=key)['price'].median().reset_index()\n", " # Rename median column with provided suffix\n", " medians_by_key.rename(columns={'price': 'median_price_' + suffix}, inplace=True)\n", " # Update data frames, note that merge seems to reset the index\n", " # that's why I reset first and set again the index\n", " trn_df = trn_df.reset_index().merge(medians_by_key, on=key, how='left').set_index('listing_id')\n", " tst_df = tst_df.reset_index().merge(medians_by_key, on=key, how='left').set_index('listing_id')\n", " trn_df['price_to_median_ratio_' + suffix] = trn_df['price'] /trn_df['median_price_' + suffix]\n", " tst_df['price_to_median_ratio_' + suffix] = tst_df['price'] / tst_df['median_price_' + suffix]\n", "\n", " return trn_df, tst_df" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "fd415d05-aae4-7426-9f66-b1058ba772b2" }, "source": [ "## Define classifier over 10 folds" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "32bc8f36-d3ec-ae9d-eb9c-b48a8c215c30" }, "outputs": [], "source": [ "from sklearn.model_selection import StratifiedKFold\n", "from sklearn.metrics import log_loss\n", "from xgboost import XGBClassifier\n", "import time\n", "\n", "def run_classifier():\n", " n_folds = 3\n", " skf = StratifiedKFold(n_splits=n_folds, shuffle=True, random_state=67594235)\n", " train_features = [ f for f in train.columns \n", " if (train[f].dtype != 'object') & (f != 'interest_level') ]\n", "\n", " target_num_map = {'high': 0, 'medium': 1, 'low': 2}\n", " target = np.array(train['interest_level'].apply(lambda x: target_num_map[x]))\n", "\n", " tst_scores = []\n", " tst_rounds = []\n", "\n", " data_X = train[train_features].values\n", " start = time.time()\n", " features_imp = np.zeros(len(train_features))\n", " for fold, (trn_idx, tst_idx) in enumerate(skf.split(data_X, target)):\n", " # Create a classifier\n", " clf = XGBClassifier(n_estimators=10000,\n", " objective='multi:softprob',\n", " learning_rate=0.3,\n", " max_depth=3,\n", " min_child_weight=1,\n", " subsample=.8,\n", " colsample_bytree=.9,\n", " colsample_bylevel=.5,\n", " gamma=0.0005,\n", " scale_pos_weight=1,\n", " base_score=.5,\n", " reg_lambda=0,\n", " reg_alpha=0,\n", " missing=0,\n", " seed=0)\n", "\n", " # Split the data\n", " trn_X = data_X[trn_idx]\n", " trn_Y = target[trn_idx]\n", " tst_X = data_X[tst_idx]\n", " tst_Y = target[tst_idx]\n", "\n", " # Train the model\n", " clf.fit(trn_X, trn_Y,\n", " eval_set=[(trn_X, trn_Y), (tst_X, tst_Y)],\n", " verbose=False,\n", " eval_metric='mlogloss',\n", " early_stopping_rounds=50)\n", "\n", " # Get features importance\n", " features_imp += clf.feature_importances_\n", "\n", " # Predict the data\n", " preds = clf.predict_proba(tst_X, ntree_limit=clf.best_ntree_limit)\n", "\n", " tst_scores.append(log_loss(tst_Y, preds))\n", " tst_rounds.append(clf.best_ntree_limit)\n", " \n", " print(\"LogLoss for fold %2d : %.5f\" % (fold+1, log_loss(tst_Y, preds)))\n", "\n", " print(\"Average LogLoss : %.5f / %.6f in %4d rounds [%5.1f mn]\"\n", " % (np.mean(tst_scores),\n", " np.std(tst_scores),\n", " np.mean(tst_rounds), (time.time() - start)/60))\n", " \n", " return train_features, features_imp" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "0d75374c-1f5c-0dab-86ca-bec060514883" }, "source": [ "## Train with basic features" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "eb9251dc-0e88-5657-855d-77f6bb76bc47" }, "outputs": [], "source": [ "for df in (train, test):\n", " df['nb_images'] = df['photos'].apply(len)\n", " df['nb_features'] = df['features'].apply(len)\n", " df['nb_words'] = df['description'].apply(lambda x: len(x.split()))\n", "\n", "train_features, features_imp = run_classifier()\n", " " ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "a8f37cc0-f21f-c17e-8806-d0891037c252" }, "source": [ "## Train with median price over bedrooms" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "8e59b44d-be1c-a1ba-9106-7a061a12f4e2" }, "outputs": [], "source": [ "train, test = add_median_price(key=['bedrooms'],\n", " suffix=\"bed\",\n", " trn_df=train, tst_df=test)\n", "\n", "train_features, features_imp = run_classifier()" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "9a3cc9ea-c632-bc49-13f9-8f802335bd02" }, "source": [ "## Show feature_importances" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "52777661-ba36-c260-e385-5a4608174d3f" }, "outputs": [], "source": [ "imp_df = pd.DataFrame(data={'feature': train_features, \n", " 'importance': features_imp})\n", "imp_df.sort_values(by='importance', ascending=False, inplace=True)\n", "sns.barplot(x=\"importance\", y=\"feature\", data=imp_df)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "63db44f5-924d-dbf2-2eec-11ec5083ab54" }, "outputs": [], "source": [ "" ] } ], "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 }