logistic regression

Logistic Regression with StatsModels/logistic.py

@@ -0,0 +1,152 @@
+"""
+Created on Wed Sep 09 12:38:16 2015
+@author: ujjwal.karn
+"""
+
+import pandas as pd                #for handling datasets
+import statsmodels.api as sm       #for statistical modeling
+import pylab as pl                 #for plotting
+import numpy as np                 #for numerical computation
+
+# read the data in
+dfTrain = pd.read_csv("C:\\Users\\ujjwal.karn\\Desktop\\Python\\train.csv")
+dfTest = pd.read_csv("C:\\Users\\ujjwal.karn\\Desktop\\Python\\test.csv")
+
+# take a look at the dataset
+print dfTrain.head()
+#   admit  gre   gpa prestige
+#0      0  380  3.61     good
+#1      1  660  3.67     good
+#2      1  800  4.00     best
+#3      1  640  3.19       ok
+#4      0  520  2.93       ok
+
+print dfTest.head()
+#   gre   gpa  prestige
+#0  640  3.30  veryGood
+#1  660  3.60      good
+#2  400  3.15  veryGood
+#3  680  3.98  veryGood
+#4  220  2.83      good
+
+
+# summarize the data
+print dfTrain.describe()
+#            admit         gre         gpa
+#count  300.000000  300.000000  300.000000
+#mean     0.306667  590.866667    3.386233
+#std      0.461880  117.717630    0.374880
+#min      0.000000  300.000000    2.260000
+#25%      0.000000  515.000000    3.130000
+#50%      0.000000  600.000000    3.390000
+#75%      1.000000  680.000000    3.642500
+#max      1.000000  800.000000    4.000000
+
+# take a look at the standard deviation of each column
+print dfTrain.std()
+#admit      0.46188
+#gre      117.71763
+#gpa        0.37488
+
+# frequency table cutting presitge and whether or not someone was admitted
+print pd.crosstab(dfTrain['admit'], dfTrain['prestige'], rownames=['dmit'])
+#prestige  best  good  ok  veryGood
+#admit                             
+#0           20    73  47        68
+#1           25    19   9        39
+
+#explore data
+dfTrain.groupby('admit').mean()
+#              gre       gpa
+#admit                      
+#0      573.461538  3.336587
+#1      630.217391  3.498478
+
+# plot one column
+dfTrain['gpa'].hist()
+pl.title('Histogram of GPA')
+pl.xlabel('GPA')
+pl.ylabel('Frequency')
+pl.show()
+
+# barplot of gre score grouped by admission status (True or False)
+pd.crosstab(dfTrain.gre, dfTrain.admit.astype(bool)).plot(kind='bar')
+pl.title('GRE score by Admission Status')
+pl.xlabel('GRE score')
+pl.ylabel('Frequency')
+pl.show()
+
+# dummify prestige
+dummy_ranks = pd.get_dummies(dfTrain['prestige'], prefix='prestige')
+print dummy_ranks.head()
+#      prestige_best  prestige_good  prestige_ok  prestige_veryGood
+#0              0              1            0                  0
+#1              0              1            0                  0
+#2              1              0            0                  0
+#3              0              0            1                  0
+#4              0              0            1                  0
+
+# create a clean data frame for the regression
+cols_to_keep = ['admit', 'gre', 'gpa']
+data = dfTrain[cols_to_keep].join(dummy_ranks.ix[:, 'prestige_good':])
+print data.head()
+#     admit  gre   gpa  prestige_good  prestige_ok  prestige_veryGood
+#0      0  380  3.61              1            0                  0
+#1      1  660  3.67              1            0                  0
+#2      1  800  4.00              0            0                  0
+#3      1  640  3.19              0            1                  0
+#4      0  520  2.93              0            1                  0
+
+# manually add the intercept
+data['intercept'] = 1.0
+
+print data.head()
+
+train_cols = data.columns[1:]
+print data.columns[1:]
+# Index([u'gre', u'gpa', u'prestige_good', u'prestige_ok', u'prestige_veryGood', u'intercept'], dtype='object')
+
+#Logistic Regression
+logit = sm.Logit(data['admit'], data[train_cols])
+
+# fit the model
+result = logit.fit()
+print result.summary()
+
+# recreate the dummy variables
+dummy_ranks_test = pd.get_dummies(dfTest['prestige'], prefix='prestige')
+print dummy_ranks_test
+
+#create intercept column
+dfTest['intercept'] = 1.0
+
+# keep only what we need for making predictions
+cols_to_keep = ['gre', 'gpa', 'prestige', 'intercept']
+dfTest = dfTest[cols_to_keep].join(dummy_ranks_test.ix[:, 'prestige_good':])
+
+dfTest.head()
+# make predictions on the enumerated dataset
+dfTest['admit_pred'] = result.predict(dfTest[train_cols])
+
+#see probabilities
+print dfTest.head()
+
+#convert probabilities to 'yes' 'no'
+dfTest['admit_yn']= np.where(dfTest['admit_pred'] > 0.5,'yes','no')
+print dfTest.head()
+
+cols= ['gre', 'gpa', 'admit_yn']
+dfTest[cols].groupby('admit_yn').mean()
+#                 gre       gpa
+#admit_yn                      
+#no        556.585366  3.324268
+#yes       676.666667  3.750000
+
+cols= ['gre', 'gpa', 'admit_yn']
+dfTest[cols].groupby('admit_yn').mean()
+#                 gre       gpa
+#admit_yn                      
+#no        556.585366  3.324268
+#yes       676.666667  3.750000
+
+dfTest.to_csv('C:\\Users\\ujjwal.karn\\Desktop\\Python\\output.csv', sep=',')

Logistic Regression with StatsModels/test.csv

@@ -0,0 +1,101 @@
+gre,gpa,prestige
+640,3.3,veryGood
+660,3.6,good
+400,3.15,veryGood
+680,3.98,veryGood
+220,2.83,good
+580,3.46,ok
+540,3.17,best
+580,3.51,veryGood
+540,3.13,veryGood
+440,2.98,good
+560,4,good
+660,3.67,veryGood
+660,3.77,good
+520,3.65,ok
+540,3.46,ok
+300,2.84,veryGood
+340,3,veryGood
+780,3.63,ok
+480,3.71,ok
+540,3.28,best
+460,3.14,good
+460,3.58,veryGood
+500,3.01,ok
+420,2.69,veryGood
+520,2.7,good
+680,3.9,best
+680,3.31,veryGood
+560,3.48,veryGood
+580,3.34,veryGood
+500,2.93,ok
+740,4,good
+660,3.59,good
+420,2.96,best
+560,3.43,good
+460,3.64,good
+620,3.71,best
+520,3.15,good
+620,3.09,ok
+540,3.2,best
+660,3.47,good
+500,3.23,ok
+560,2.65,good
+500,3.95,ok
+580,3.06,veryGood
+520,3.35,good
+500,3.03,good
+600,3.35,veryGood
+580,3.8,veryGood
+400,3.36,veryGood
+620,2.85,veryGood
+780,4,veryGood
+620,3.43,good
+580,3.12,good
+700,3.52,veryGood
+540,3.78,veryGood
+760,2.81,best
+700,3.27,veryGood
+720,3.31,best
+560,3.69,good
+720,3.94,good
+520,4,best
+540,3.49,best
+680,3.14,veryGood
+460,3.44,veryGood
+560,3.36,best
+480,2.78,good
+460,2.93,good
+620,3.63,good
+580,4,best
+800,3.89,veryGood
+540,3.77,veryGood
+680,3.76,good
+680,2.42,best
+620,3.37,best
+560,3.78,veryGood
+560,3.49,ok
+620,3.63,veryGood
+800,4,veryGood
+640,3.12,good
+540,2.7,veryGood
+700,3.65,veryGood
+540,3.49,veryGood
+540,3.51,veryGood
+660,4,best
+480,2.62,veryGood
+420,3.02,best
+740,3.86,veryGood
+580,3.36,veryGood
+640,3.17,veryGood
+640,3.51,veryGood
+800,3.05,veryGood
+660,3.88,veryGood
+600,3.38,good
+620,3.75,veryGood
+460,3.99,good
+620,4,veryGood
+560,3.04,good
+460,2.63,veryGood
+700,3.65,veryGood
+600,3.89,good