Files
Amazing-Feature-Engineering/3.2_Demo_Discretisation.ipynb
DESKTOP-SAT83DL\yimeng.zhang 149c80cedf 2018.12.2 First commit.
2018-12-02 21:11:32 +08:00

866 lines
23 KiB
Plaintext

{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"import pandas as pd\n",
"import numpy as np\n",
"# import seaborn as sns\n",
"# import matplotlib.pyplot as plt\n",
"import os\n",
"from sklearn.model_selection import train_test_split\n",
"from feature_engineering import discretization as dc\n",
"\n",
"# plt.style.use('seaborn-colorblind')\n",
"# %matplotlib inline\n",
"#from feature_cleaning import rare_values as ra"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Load Dataset"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"use_cols = [\n",
" 'Pclass', 'Sex', 'Age', 'Fare', 'SibSp',\n",
" 'Survived'\n",
"]\n",
"\n",
"data = pd.read_csv('./data/titanic.csv', usecols=use_cols)\n"
]
},
{
"cell_type": "code",
"execution_count": 3,
"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>Survived</th>\n",
" <th>Pclass</th>\n",
" <th>Sex</th>\n",
" <th>Age</th>\n",
" <th>SibSp</th>\n",
" <th>Fare</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0</td>\n",
" <td>3</td>\n",
" <td>male</td>\n",
" <td>22.0</td>\n",
" <td>1</td>\n",
" <td>7.2500</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>1</td>\n",
" <td>1</td>\n",
" <td>female</td>\n",
" <td>38.0</td>\n",
" <td>1</td>\n",
" <td>71.2833</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>1</td>\n",
" <td>3</td>\n",
" <td>female</td>\n",
" <td>26.0</td>\n",
" <td>0</td>\n",
" <td>7.9250</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Survived Pclass Sex Age SibSp Fare\n",
"0 0 3 male 22.0 1 7.2500\n",
"1 1 1 female 38.0 1 71.2833\n",
"2 1 3 female 26.0 0 7.9250"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"data.head(3)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"((623, 6), (268, 6))"
]
},
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# Note that we include target variable in the X_train \n",
"# because we need it to supervise our discretization\n",
"# this is not the standard way of using train-test-split\n",
"X_train, X_test, y_train, y_test = train_test_split(data, data.Survived, test_size=0.3,\n",
" random_state=0)\n",
"X_train.shape, X_test.shape"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Equal width binning\n",
"divides the scope of possible values into N bins of the same width"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"from sklearn.preprocessing import KBinsDiscretizer\n",
"enc_equal_width = KBinsDiscretizer(n_bins=3,encode='ordinal',strategy='uniform').fit(X_train[['Fare']])"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array([array([ 0. , 170.7764, 341.5528, 512.3292])], dtype=object)"
]
},
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# equal width for every bins\n",
"enc_equal_width.bin_edges_"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"0.0 610\n",
"1.0 11\n",
"2.0 2\n",
"Name: 0, dtype: int64"
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"result = enc_equal_width.transform(X_train[['Fare']])\n",
"pd.DataFrame(result)[0].value_counts()"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" Survived Pclass Sex Age SibSp Fare Fare_equal_width\n",
"857 1 1 male 51.0 0 26.5500 0.0\n",
"52 1 1 female 49.0 1 76.7292 0.0\n",
"386 0 3 male 1.0 5 46.9000 0.0\n",
"124 0 1 male 54.0 0 77.2875 0.0\n",
"578 0 3 female NaN 1 14.4583 0.0\n",
"549 1 2 male 8.0 1 36.7500 0.0\n",
"118 0 1 male 24.0 0 247.5208 1.0\n",
"12 0 3 male 20.0 0 8.0500 0.0\n",
"157 0 3 male 30.0 0 8.0500 0.0\n",
"127 1 3 male 24.0 0 7.1417 0.0\n"
]
}
],
"source": [
"# add the new discretized variable\n",
"X_train_copy = X_train.copy(deep=True)\n",
"X_train_copy['Fare_equal_width'] = enc_equal_width.transform(X_train[['Fare']])\n",
"print(X_train_copy.head(10))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Equal frequency binning\n",
"divides the scope of possible values of the variable into N bins, \n",
"where each bin carries the same amount of observations"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"enc_equal_freq = KBinsDiscretizer(n_bins=3,encode='ordinal',strategy='quantile').fit(X_train[['Fare']])"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array([array([ 0. , 8.69303333, 26.2875 , 512.3292 ])],\n",
" dtype=object)"
]
},
"execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# check the bin edges\n",
"enc_equal_freq.bin_edges_"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"2.0 209\n",
"0.0 208\n",
"1.0 206\n",
"Name: 0, dtype: int64"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# equal number of case for every bins\n",
"result = enc_equal_freq.transform(X_train[['Fare']])\n",
"pd.DataFrame(result)[0].value_counts()"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" Survived Pclass Sex Age SibSp Fare Fare_equal_freq\n",
"857 1 1 male 51.0 0 26.5500 2.0\n",
"52 1 1 female 49.0 1 76.7292 2.0\n",
"386 0 3 male 1.0 5 46.9000 2.0\n",
"124 0 1 male 54.0 0 77.2875 2.0\n",
"578 0 3 female NaN 1 14.4583 1.0\n",
"549 1 2 male 8.0 1 36.7500 2.0\n",
"118 0 1 male 24.0 0 247.5208 2.0\n",
"12 0 3 male 20.0 0 8.0500 0.0\n",
"157 0 3 male 30.0 0 8.0500 0.0\n",
"127 1 3 male 24.0 0 7.1417 0.0\n"
]
}
],
"source": [
"# add the new discretized variable\n",
"X_train_copy = X_train.copy(deep=True)\n",
"X_train_copy['Fare_equal_freq'] = enc_equal_freq.transform(X_train[['Fare']])\n",
"print(X_train_copy.head(10))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## K-means binning\n",
"using k-means to partition values into clusters"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"enc_kmeans = KBinsDiscretizer(n_bins=3,encode='ordinal',strategy='kmeans').fit(X_train[['Fare']])"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array([array([ 0. , 93.5271531 , 338.08506324, 512.3292 ])],\n",
" dtype=object)"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# check the bin edges\n",
"enc_kmeans.bin_edges_"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"0.0 587\n",
"1.0 34\n",
"2.0 2\n",
"Name: 0, dtype: int64"
]
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"result = enc_kmeans.transform(X_train[['Fare']])\n",
"pd.DataFrame(result)[0].value_counts()"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" Survived Pclass Sex Age SibSp Fare Fare_kmeans\n",
"857 1 1 male 51.0 0 26.5500 0.0\n",
"52 1 1 female 49.0 1 76.7292 0.0\n",
"386 0 3 male 1.0 5 46.9000 0.0\n",
"124 0 1 male 54.0 0 77.2875 0.0\n",
"578 0 3 female NaN 1 14.4583 0.0\n",
"549 1 2 male 8.0 1 36.7500 0.0\n",
"118 0 1 male 24.0 0 247.5208 1.0\n",
"12 0 3 male 20.0 0 8.0500 0.0\n",
"157 0 3 male 30.0 0 8.0500 0.0\n",
"127 1 3 male 24.0 0 7.1417 0.0\n"
]
}
],
"source": [
"# add the new discretized variable\n",
"X_train_copy = X_train.copy(deep=True)\n",
"X_train_copy['Fare_kmeans'] = enc_kmeans.transform(X_train[['Fare']])\n",
"print(X_train_copy.head(10))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Discretisation with Decision Tree\n",
"using a decision tree to identify the optimal splitting points that would determine the bins"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"enc1 = dc.DiscretizeByDecisionTree(col='Fare',max_depth=2).fit(X=X_train,y=y_train)"
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=2,\n",
" max_features=None, max_leaf_nodes=None,\n",
" min_impurity_decrease=0.0, min_impurity_split=None,\n",
" min_samples_leaf=1, min_samples_split=2,\n",
" min_weight_fraction_leaf=0.0, presort=False, random_state=None,\n",
" splitter='best')"
]
},
"execution_count": 18,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"enc1.tree_model"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"data1 = enc1.transform(data)"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" Survived Pclass Sex Age SibSp Fare Fare_tree_discret\n",
"0 0 3 male 22.0 1 7.2500 0.107143\n",
"1 1 1 female 38.0 1 71.2833 0.442308\n",
"2 1 3 female 26.0 0 7.9250 0.255319\n",
"3 1 1 female 35.0 1 53.1000 0.442308\n",
"4 0 3 male 35.0 0 8.0500 0.255319\n",
"[0.10714286 0.44230769 0.25531915 0.74626866]\n"
]
}
],
"source": [
"# see how the new column Fare_tree_discret is distributed\n",
"# the values are corresponding to the proba of the prediction by the tree\n",
"print(data1.head(5))\n",
"\n",
"# the unique value of the discretisized column\n",
"print(data1.Fare_tree_discret.unique())"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" Fare Fare\n",
"Fare_tree_discret \n",
"0.107143 0.0000 7.5208\n",
"0.255319 7.5500 10.5167\n",
"0.442308 11.1333 73.5000\n",
"0.746269 75.2500 512.3292\n"
]
}
],
"source": [
"# see how the bins are cut\n",
"# because we use a tree with max-depth of 2, we have at most 2*2=4 bins generated by the tree\n",
"col='Fare'\n",
"bins = pd.concat([data1.groupby([col+'_tree_discret'])[col].min(),\n",
" data1.groupby([col+'_tree_discret'])[col].max()], axis=1)\n",
"print(bins)\n",
"\n",
"# all values between 0 to 7.5208 in the original variable 'Fare' \n",
"# are given new value 0.107143 in the new column 'Fare_tree_discret'\n",
"# and so on"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Discretisation with Decision Tree with optimal depth search"
]
},
{
"cell_type": "code",
"execution_count": 22,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"result ROC-AUC for each depth\n",
" depth roc_auc_mean roc_auc_std\n",
"0 2 0.662132 0.026253\n",
"1 3 0.647950 0.045010\n",
"2 4 0.650984 0.035127\n",
"3 5 0.651180 0.027663\n",
"4 6 0.653961 0.037421\n",
"5 7 0.643688 0.033513\n",
"optimal_depth: [2]\n"
]
}
],
"source": [
"# search for the best depth from range 2-7\n",
"# we see when depth=2 we get the best roc-auc mean\n",
"enc2 = dc.DiscretizeByDecisionTree(col='Fare',max_depth=[2,3,4,5,6,7]).fit(X=X_train,y=y_train)"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"DecisionTreeClassifier(class_weight=None, criterion='gini',\n",
" max_depth=array([2], dtype=int64), max_features=None,\n",
" max_leaf_nodes=None, min_impurity_decrease=0.0,\n",
" min_impurity_split=None, min_samples_leaf=1,\n",
" min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
" presort=False, random_state=None, splitter='best')"
]
},
"execution_count": 23,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# using optimal depth=2 we train the model, same result as last one\n",
"enc2.tree_model"
]
},
{
"cell_type": "code",
"execution_count": 24,
"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>Survived</th>\n",
" <th>Pclass</th>\n",
" <th>Sex</th>\n",
" <th>Age</th>\n",
" <th>SibSp</th>\n",
" <th>Fare</th>\n",
" <th>Fare_tree_discret</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0</td>\n",
" <td>3</td>\n",
" <td>male</td>\n",
" <td>22.0</td>\n",
" <td>1</td>\n",
" <td>7.2500</td>\n",
" <td>0.107143</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>1</td>\n",
" <td>1</td>\n",
" <td>female</td>\n",
" <td>38.0</td>\n",
" <td>1</td>\n",
" <td>71.2833</td>\n",
" <td>0.442308</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>1</td>\n",
" <td>3</td>\n",
" <td>female</td>\n",
" <td>26.0</td>\n",
" <td>0</td>\n",
" <td>7.9250</td>\n",
" <td>0.255319</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>1</td>\n",
" <td>1</td>\n",
" <td>female</td>\n",
" <td>35.0</td>\n",
" <td>1</td>\n",
" <td>53.1000</td>\n",
" <td>0.442308</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>0</td>\n",
" <td>3</td>\n",
" <td>male</td>\n",
" <td>35.0</td>\n",
" <td>0</td>\n",
" <td>8.0500</td>\n",
" <td>0.255319</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Survived Pclass Sex Age SibSp Fare Fare_tree_discret\n",
"0 0 3 male 22.0 1 7.2500 0.107143\n",
"1 1 1 female 38.0 1 71.2833 0.442308\n",
"2 1 3 female 26.0 0 7.9250 0.255319\n",
"3 1 1 female 35.0 1 53.1000 0.442308\n",
"4 0 3 male 35.0 0 8.0500 0.255319"
]
},
"execution_count": 24,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"data2 = enc2.transform(data)\n",
"data2.head(5)"
]
},
{
"cell_type": "markdown",
"metadata": {
"collapsed": true
},
"source": [
"## Discretisation with ChiMerge\n",
"supervised hierarchical bottom-up (merge) method that locally exploits the chi-square criterion to decide whether two adjacent intervals are similar enough to be merged"
]
},
{
"cell_type": "code",
"execution_count": 25,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Interval for variable Fare\n",
" variable interval flag_0 flag_1\n",
"0 Fare -inf,7.875 94.0 28.0\n",
"1 Fare 7.875,7.8792 0.0 3.0\n",
"2 Fare 7.8792,7.8958 25.0 1.0\n",
"3 Fare 7.8958,73.5 245.0 160.0\n",
"4 Fare 73.5+ 17.0 50.0\n"
]
}
],
"source": [
"enc3 = dc.ChiMerge(col='Fare',num_of_bins=5).fit(X=X_train,y='Survived')"
]
},
{
"cell_type": "code",
"execution_count": 26,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[-0.1, 7.875, 7.8792, 7.8958, 73.5, 512.3292]"
]
},
"execution_count": 26,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# the bins boundary created by ChiMerge\n",
"\n",
"enc3.bins"
]
},
{
"cell_type": "code",
"execution_count": 27,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"data3 = enc3.transform(data)"
]
},
{
"cell_type": "code",
"execution_count": 28,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" Survived Pclass Sex Age SibSp Fare Fare_chimerge\n",
"0 0 3 male 22.0 1 7.2500 (-0.101, 7.875]\n",
"1 1 1 female 38.0 1 71.2833 (7.896, 73.5]\n",
"2 1 3 female 26.0 0 7.9250 (7.896, 73.5]\n",
"3 1 1 female 35.0 1 53.1000 (7.896, 73.5]\n",
"4 0 3 male 35.0 0 8.0500 (7.896, 73.5]\n"
]
}
],
"source": [
"print(data3.head(5))"
]
},
{
"cell_type": "code",
"execution_count": 29,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[(-0.101, 7.875], (7.896, 73.5], (73.5, 512.329], (7.875, 7.879], (7.879, 7.896]]\n",
"Categories (5, interval[float64]): [(-0.101, 7.875] < (7.875, 7.879] < (7.879, 7.896] < (7.896, 73.5] < (73.5, 512.329]]"
]
},
"execution_count": 29,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# all values are grouped into 5 intervals\n",
"data3.Fare_chimerge.unique()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"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": "3.6.1"
}
},
"nbformat": 4,
"nbformat_minor": 2
}