LLR

import pandas as pd
import numpy as np
from sklearn.preprocessing import LabelBinarizer
from math import *
from decimal import Decimal
from sklearn import metrics


def create_dummy_cooc():
    d = {'A': [0, 2, 1], 'B':[2, 0, 3], 'C': [1, 3, 0]}
    data = pd.DataFrame(d, index=['A', 'B', 'C'])
    labels = data.index
    data = data.as_matrix()
    return data, labels


def get_log_likelihood_df(data, labels):
    items = []
    for brief1 in labels:
        for brief2 in labels:
            if brief1 == brief2:
                items.append(dict(Brief1=brief1, Brief2=brief2, LLR=0))
            else:
                llr = get_log_likelihood(data, labels, brief1, brief2)
                items.append(dict(Brief1=brief1, Brief2=brief2, LLR=llr))
                print brief1, brief2, llr
    constructed = pd.DataFrame(items)
    return pd.pivot_table(data=constructed, columns='Brief1', index='Brief2', aggfunc=sum)


def get_log_likelihood(data, labels, brief1, brief2):
    X = data[:, labels == brief1]
    X = np.append(X, np.sum(data[:, labels != brief1], axis=1, keepdims=True), axis=1)
    y = LabelBinarizer().fit_transform(labels != brief2)
    y = np.append(1 - y, y, axis=1)
    x_table = np.dot(y.T, X)
    print x_table
    feature_count = np.sum(X, axis=0, keepdims=True)
    class_prob = np.mean(y, axis=0, keepdims=True)
    expected_table = np.dot(class_prob.T, feature_count)
    llr = np.sum(x_table * np.log(x_table / expected_table))
    return llr


def main():
    data, labels = create_dummy_cooc()
    llr_df = get_log_likelihood_df(data, labels)
    print llr_df    # yay
    return llr_df


if __name__ == '__main__':
    main()