calc cr & ar, for ocr benchmark.

cr_ar.py

def custom_min(data):
    assert len(data) >= 1 
    if len(data) <= 0:
        return data,0
    min_idx = 0
    min_data = data[min_idx]
    for i,item in enumerate(data):
        if item < min_data:
            min_data = item
            min_idx = i 
    return min_data,min_idx 
    
def wer(r, h):
    # initialisation
    import numpy
    d = numpy.zeros((len(r)+1)*(len(h)+1), dtype=numpy.uint8)
    d = d.reshape((len(r)+1, len(h)+1))
    for i in range(len(r)+1):
        for j in range(len(h)+1):
            if i == 0:
                d[0][j] = j
            elif j == 0:
                d[i][0] = i

    # computation
    for i in range(1, len(r)+1):
        for j in range(1, len(h)+1):
            if r[i-1] == h[j-1]:
                d[i][j] = d[i-1][j-1]                
            else:
                substitution = d[i-1][j-1] + 1
                insertion    = d[i][j-1] + 1
                deletion     = d[i-1][j] + 1
                d[i][j],min_idx = custom_min([substitution, insertion, deletion])
    dist = d[len(r)][len(h)]
    
    #reverse track
    #subsitution, insertion, deletion, total
    details_count = [0,0,0,0]
    min_y = len(r)
    min_x = len(h)
    min_value = d[min_y][min_x]
    while True:        
        substitution = d[min_y-1][min_x-1]
        insertion = d[min_y][min_x-1]
        deletion = d[min_y-1][min_x]        
        cur_min_value,min_idx = custom_min([substitution, insertion, deletion])        
        details_count[-1] += 1
        if min_idx == 0:
            min_y -= 1
            min_x -= 1
            if cur_min_value < min_value:
                details_count[0] += 1
        elif min_idx == 1:            
            min_x -= 1
            if cur_min_value < min_value:
                details_count[2] += 1
        elif min_idx == 2:            
            min_y -= 1
            if cur_min_value < min_value:
                details_count[1] += 1
                
        min_value = cur_min_value
        if min_x < 0 or min_y < 0:
            break    
    return dist,details_count[0],details_count[1],details_count[2]
    
def cr_ar(subsitution, insertion, deletion, golden_len):
    cr = float(golden_len - deletion - subsitution) / float(golden_len)
    ar = float(golden_len - deletion - subsitution - insertion) / float(golden_len)
    #can be negative
    return cr,ar

def score(golden, predicts):
    golden_len = len(golden)
    _, subsitution, insertion, deletion = wer(golden, predicts)
    return cr_ar(subsitution, insertion, deletion, golden_len)