protein fold

gistfile1.txt

import numpy as np
import pandas as pd
import numpy.random as nr

# Read In
DF = pd.read_csv('train.csv',sep=',',header=0)
for i in xrange(28):
    DF[str(i)] = DF['Target'].map(lambda x: int(str(i) in x.split(' ')))

value_counts =  DF.ix[:,2:].apply(np.sum, axis=0)
class_order = list(value_counts.sort_values().index)
DF['inpool'] = 0

# Config Fold Num

Nfold = 5

# Arrange Fold From Least Labelled Class

for item in class_order:
    #print item,
    value_counts[item]
    origin_pool = [value_counts[item]/Nfold for i in range(Nfold)]
    randadd = nr.choice(Nfold, value_counts[item]%Nfold)
    while len(randadd)!=len(set(randadd)):
        randadd = nr.choice(Nfold, value_counts[item]%Nfold)
    for index in randadd:
        origin_pool[index] = origin_pool[index]+1
    existing_pool = DF[DF[item]==1].groupby('inpool').count().ix[:,0].to_dict()
    for i in range(Nfold):
        if i+1 not in existing_pool:
            existing_pool[i+1] = 0

    waiting_pool = [each for each in origin_pool]
    for i in range(Nfold):
        if waiting_pool[i]<existing_pool[i+1]:
            print "Error Allocating Label: Could Not Balance"
        waiting_pool[i]  = waiting_pool[i] - existing_pool[i+1]

    unperturb_label = [pool_id+1 for pool_id in range(Nfold) for each in range(waiting_pool[pool_id]) ]
    perturb_label = nr.permutation(unperturb_label)
    #print perturb_label,
    #print len(perturb_label),
    #print len(DF[(DF['inpool']==0) & (DF[item]==1)]['inpool'])
    for index,dfindex in enumerate(DF[(DF['inpool']==0) & (DF[item]==1)]['inpool'].index):
        DF.ix[dfindex,'inpool'] = perturb_label[index]
        
# Check Balancing

DF.groupby('inpool').sum()