Generate consensus array from pandas DataFrame (NaN values are ignored)

consensus_array.py

import pandas as pd

# load data
mat = pd.read_table('class_matrix.txt', index_col=0)

# initialize consensus array
consensus_a = pd.Series(index=mat.index)

# define columns subset on which compute consensus
# in this case all columns are used
columns = mat.columns

# compute consensus array
for ind in mat.index:
    most_class = mat.ix[ind, columns].value_counts()[0]
    n_elements = mat.ix[ind, columns].value_counts().sum()
    consensus_a.ix[ind] = float(most_class) / n_elements

# save to csv
mat.to_csv('consensus_array.txt', sep='\t', index_label='Samples')