produceInfoGainMatrix.py

Go to the documentation of this file.

import sys
import caffe
import matplotlib
import numpy as np
import lmdb
import argparse
import leveldb
import ROOT
import h5py
from collections import defaultdict

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument('--leveldb', type=str, required=True)

    args = parser.parse_args()

    count = 0

    matrix= np.zeros(392, dtype = 'f4' )
    
    caffe.set_mode_gpu()
    db = leveldb.LevelDB(args.leveldb)
    
    nEvents=200000
    for key, value in db.RangeIter():
        if count > nEvents: 
            break
            
        datum = caffe.proto.caffe_pb2.Datum()
        datum.ParseFromString(value)
        label = int(datum.label)
        matrix[label] += 1
        count +=1
        
    print matrix

    total=0
    for i in range(0, 392):
        if matrix[i] < (0.001*count):
            matrix[i]= (0.001*count)

        print matrix[i]
        total=total+matrix[i]*((1./(matrix[i]/count))/(3920.*count))
        matrix[i]=(1./(matrix[i]/count))/(3920.)    
        print matrix[i]
        matrix[i]=0.7+(matrix[i])
        print matrix[i]

    print total
    print count

    print matrix
    
    summatrix=np.zeros((392, 392), dtype = 'f4')
    for i in range(0, 392):
        summatrix[i][i]=matrix[i]

    readyForH=summatrix

    print readyForH
    
    print readyForH.shape

    blob=caffe.io.array_to_blobproto(readyForH.reshape(1,1,392,392))

    with open('infoGain.binaryproto','wb') as f:
        f.write(blob.SerializeToString())