def cross_validate(j_lines,e_lines,index,ngram_order=1,verbose=False):
print "[Start] Cross Validating Data"
rightfile = open("out_right.txt" % index, "w")
wrongfile = open("out_wrong.txt" % index, "w")
get_right = 0
total_score = 0.0
buckets = [0,0,0,0,0,0,0,0,0,0]
print "[.....] Creating training and testing data sets"
train_j,train_e,test_j,test_e = split_data(j_lines,e_lines,index)
total_test = len(test_e)
print "[Start] Converting training set to %i-Grams" % ngram_order
n_gramed_japanese = [ngramer.ngram(x, ngram_order) for x in train_j]
n_gramed_english = [ngramer.ngram(x, ngram_order) for x in train_e]
print "[ End ] Converting training set to %i-Grams" % ngram_order
model = construct_hmm(n_gramed_japanese,n_gramed_english)
# NOW STAT THE TESTING PHASE
# need to first also build the truth into dictionary
print "[.....] Preparing test set"
ground_truth = {}
for i in range(len(test_e)):
e = ' '.join(test_e[i])
j = ' '.join(test_j[i])
ground_truth[e] = j
print "[.....] Testing"
for eng in test_e:
e = ' '.join(eng)
score,correct,guess = test_one(model,e,ground_truth[e],ngram_order=ngram_order,verbose=verbose)
if correct:
get_right +=1
rightfile.write("%s,%s\n" % (e, ground_truth[e]))
else:
wrongfile.write("%s,%s,%s\n" % (e, ground_truth[e],guess))
total_score += score
bucket = int(score*10)
if bucket > 9:
bucket = 9
buckets[bucket] = buckets[bucket] + 1
avg_score = total_score/total_test
precision_word_wise = float(get_right)/float(total_test)
print "[.....] Average Score: %f" % avg_score
print "[.....] Wordwise Precision: %f" % precision_word_wise
print "[.....] Buckets:"
for i in range(10):
print "[.....] %i to %i Percent: %i" % ((i*10), ((i+1)*10), buckets[i])
print "[ End ] Cross Validating Data"
rightfile.close()
wrongfile.close()
return avg_score,precision_word_wise
def guess(model,teststring,ngram_order=1,verbose=False): # lable a single test string and return it's score # change "AA R B AX T ER" into ['AA','R','B', 'AX', 'T','ER'] unlabeled = teststring.split() unlabeled_ngram = ngramer.ngram(unlabeled, ngram_order) if verbose: print "Converted to %i-Gram: %s " % (ngram_order, unlabeled_ngram) result = model.tag(unlabeled_ngram) # Fetch the (possibly N-Gramed) result result_ngram = [] for align in result: result_ngram.append(align[1]) if verbose: print "Recovered: %s " % (result_ngram) # Un-NGram the result result_recovered = ngramer.unngram(result_ngram) if verbose: print "Fixed to: %s " % (result_recovered) # Construct the guess from this array kata_guess = "".join(result_recovered) return kata_guess
def smalltest(dictionary_file, ngram_order=1, verbose=False): import ngramer japanese, english = read_dictionary(dictionary_file) aligned_japanese, aligned_english = align_phoneme_sets(japanese, english) print "[Start] Converting dictionary to %i-Grams" % ngram_order n_gramed_japanese = [ngramer.ngram(x, ngram_order) for x in aligned_japanese] n_gramed_english = [ngramer.ngram(x, ngram_order) for x in aligned_english] print "[ End ] Converting dictionary to %i-Grams" % ngram_order model = construct_hmm(n_gramed_japanese,n_gramed_english) true_japanese = "A - BA n" english = "ER B AX N" test_one(model,"AA M AX N D AX","A - MO n DO",ngram_order,verbose)
def newwordtest(dictionary_file, wordfile, ngram_order=1, verbose=False):
import ngramer
japanese, english = read_dictionary(dictionary_file)
aligned_japanese, aligned_english = align_phoneme_sets(japanese, english)
print "[Start] Converting dictionary to %i-Grams" % ngram_order
n_gramed_japanese = [ngramer.ngram(x, ngram_order) for x in aligned_japanese]
n_gramed_english = [ngramer.ngram(x, ngram_order) for x in aligned_english]
print "[ End ] Converting dictionary to %i-Grams" % ngram_order
model = construct_hmm(n_gramed_japanese,n_gramed_english)
testfile = open(wordfile)
testlines = testfile.readlines()
for i in range(len(testlines)):
line = testlines[i]
tup = line.split(",")
english = tup[0]
sound = tup[1]
kguess = guess(model,sound,ngram_order,verbose)
print "%s -> %s -> %s" % (english, sound, kguess)