[WIP] Add GLM example with the Negative Binomial distribution.

examples/plot_negative_binomial.py

@@ -0,0 +1,94 @@
+# -*- coding: utf-8 -*-
+"""
+=======================================
+GLM with Negative Binomial Distribution
+=======================================
+
+This is an example of GLM with negative binomial distribution.
+We wrote this example taking inspiration from the R community
+below
+https://stats.idre.ucla.edu/r/dae/negative-binomial-regression/
+
+Here, we would like to predict the number of days absence of high school
+juniors at two schools from there type of program they are enrolled,
+and their math score.
+
+The nature of the empirical data suggests that we need to model
+count data (the number of days absent). In such scenarios, a common model
+we could use is the Poisson regression.
+
+However, if we inspect the dataset more closely, we will notice that
+the dataset is over-dispersed since the conditional mean exceeds the
+conditional variance. We would need to apply another model which is the
+Negative Binomial regression.
+
+The Negative Binomial regression can be seen as a mixture of Poisson
+regression in which the mean of the Poisson distribution can be seen
+as a random variable drawn from a Gamma distribution.
+
+This gives us an extra parameter which can be used to account for the over
+dispersion.
+"""
+
+########################################################
+
+# Author: Titipat Achakulvisut <my.titipat@gmail.com>
+#         Giovanni De Toni <giovanni.det@gmail.com>
+# License: MIT
+
+########################################################
+
+
+########################################################
+# Import relevance libraries
+
+import pandas as pd
+from pyglmnet import GLM
+
+from sklearn.model_selection import train_test_split
+
+import matplotlib.pyplot as plt
+
+########################################################
+# Read and preprocess data
+df = pd.read_stata("https://stats.idre.ucla.edu/stat/stata/dae/nb_data.dta")
+
+# Histogram of type of program they are enrolled
+df.hist(column='daysabs', by=['prog'])
+plt.show()
+
+# Print mean and standard deviation for each program enrolled.
+# We can see from here that the variance is higher that then mean for all
+# the levels, therefore hinting for over-dispersion.
+prog_mean = df.groupby('prog').agg({'daysabs': ['mean', 'std']})
+print(prog_mean)
+
+########################################################
+# Feature
+X = df.drop('daysabs', axis=1)
+y = df['daysabs'].values
+
+# design matrix
+program_df = pd.get_dummies(df.prog)
+Xdsgn = pd.concat((df['math'], program_df.drop(3.0, axis=1)), axis=1).values
+
+# Split the dataset into training and test
+Xtrain, Xtest, ytrain, ytest = train_test_split(Xdsgn, y, test_size=0.2)
+
+########################################################
+# Fit the model using the GLM
+glm_neg_bino = GLM(distr='neg-binomial',
+                   alpha=0.0,
+                   reg_lambda=0.0,
+                   score_metric='pseudo_R2',
+                   verbose=True)
+glm_neg_bino.fit(Xtrain, ytrain)
+
+########################################################
+# Predict
+y_hat = glm_neg_bino.predict(Xtest)
+
+########################################################
+# Return the learned betas and the score
+print(glm_neg_bino.beta0_, glm_neg_bino.beta_)
+print(glm_neg_bino.score(Xtest, y_hat))