Naive bayes with pandas, not working yet!... see https://higgsml.lal.in2p3.fr/software/starting-kit/

pandasNaiveBayes.py

import pandas as pd

TRAINING_SET = 'data/training.csv'
TEST_SET = 'data/test.csv'

training = pd.read_csv(TRAINING_SET)


#background and signal total weights
total_weights = traininig.groupby(training.Label).sum()

training.Weight[training.Label == 's'] *= 0.5 / total_weights.s
training.Weight[training.Label == 'b'] *= 0.5 / total_weights.b

numFeatures = 30
numBins = 10

logPs = np.empty((numFeatures, numBins))

training_feats = training.drop(['EventId', 'Weight', 'Label'], axis=1)


for idx, fI in enumerate(training_feats.keys()):
    groups = training.groupby(pd.cut(training_feats[fI], bins=10))
    for bI, (group, weights) in enumerate(groups.Weight):
        wS = weights[training.Label == 's'].sum()
        wB = weights[training.Label == 'b'].sum()
        logPs[idx, bI] = np.log(wS / (wS + wB))


def score(x):
    logP = 0
    for fI in range(numFeatures):
        logP += logPs[fI, bI] - np.log(0.5)

    return logP


def AMS(s, b):
    assert s >= 0
    assert b >= 0
    bReg = 10.
    return np.sqrt(2 * ((s + b + bReg) * 
        np.log(1 + s / (b + bReg)) - s))