def RunEKD(dirname, model, train_test='train',TopK=5):
if train_test.lower() == 'train':
Y = Util.LoadClassFile(os.path.join(dirname, 'train_classes.txt'))
else:
Y = Util.LoadClassFile(os.path.join(dirname, 'test_classes.txt'))
# # Manditory settings for the Features Module. Others are up to you
import Features as FeaturesModule
FeaturesModule.FUNIT = FeaturesModule.FeatureUnits.WORD
FeaturesModule.FTYPE = FeaturesModule.FeatureType.BINARY
FeaturesModule.FREP = FeaturesModule.FeatureRepresentation.STRING
FeaturesModule.REMOVE_FEATURES_APPEARING_IN_ONLY_ONE_DOCUMENT=False
FeaturesModule.REMOVE_FEATURES_ONLY_APPEARING_ONE_TIME=False
FeaturesModule.DisplayConfiguration()
(feature_def, X) = FeaturesModule.Features(os.path.join(dirname, train_test), frep=FeaturesModule.FREP, ftype=FeaturesModule.FTYPE, funit=FeaturesModule.FUNIT)
(word_idx, emb) = Word2VecExecuter.Word2VecLoadWordsHashTable(model, list(feature_def.T[0]))
X = FeaturesModule.Features(os.path.join(dirname, train_test), frep=FeaturesModule.FREP, ftype=FeaturesModule.FTYPE, funit=FeaturesModule.FUNIT, feature=np.array(word_idx.keys()).reshape((len(word_idx), 1)))
count1 = 0
N = Util.get_num_samples(os.path.join(dirname, train_test))
dists = np.zeros((N,N))
for i in xrange(0, N):
Doc1 = EKD.Document(embeddings=emb[X[i, :], :])
for j in xrange(0, N):
if not i == j:
Doc2 = EKD.Document(embeddings=emb[X[j, :], :])
dists[i,j] = Doc1.distance(Doc2)
precision = 0
recall = 0
print("\n**Results**")
for i in xrange(0, N):
R = dists[i,:].argsort()
R = R[1:TopK]
precision += Metrics.RetrievalPrecision(Y[i], Y[R])
recall += Metrics.RetrievalRecall(Y[i], Y[R], Y, TopK)
print("Document Number: " + str(i) + " Class: " + str(Y[i]))
print("Top Matching Documents: " + str(R))
print("Top Matching Documents (Associated Class Labels: " + str(Y[R]))
print(" Precision: " + str(Metrics.RetrievalPrecision(Y[i], Y[R])) + " Recall: " + str(Metrics.RetrievalRecall(Y[i], Y[R], Y, TopK)))
precision = precision / N
recall = recall / N
print("Average Precision: " + str(precision) + " Average Recall: " + str(recall))
return (precision, recall)
# # This will be for pairwise
# # Compare document to every other document and say if they are the same or different
# def Run(FeaturesModule, dirname, train_test='train', Score Cut off):
# def RunEKD(dirname, train_test='train'):
def RunPAD(dirname, model, train_test, TopK=5):
if train_test.lower() == 'train':
Y = Util.LoadClassFile(os.path.join(dirname, 'train_classes.txt'))
else:
Y = Util.LoadClassFile(os.path.join(dirname, 'test_classes.txt'))
count1 = 0
N = Util.get_num_samples(os.path.join(dirname, train_test))
dists = np.zeros((N,N))
for filename1 in os.listdir(os.path.join(dirname, train_test)):
count2 = 0
if '.srl' in filename1:
Doc1 = PAD.Document(doc_file_name=os.path.join(dirname, train_test, filename1), model=model, use_lemma=False)
for filename2 in os.listdir(os.path.join(dirname, train_test)):
if filename1 == filename2:
dists[count1, count2] = 0
count2 += 1
elif '.srl' in filename2:
Doc2 = PAD.Document(doc_file_name=os.path.join(dirname, train_test, filename2), model=model, use_lemma=False)
dists[count1, count2] = Doc1.distance(Doc2)
count2 += 1
count1 += 1
precision = 0
recall = 0
print("\n**Results**")
for i in xrange(0, N):
R = dists[i,:].argsort()
R = R[1:TopK]
precision += Metrics.RetrievalPrecision(Y[i], Y[R])
recall += Metrics.RetrievalRecall(Y[i], Y[R], Y, TopK)
print("Document Number: " + str(i) + " Class: " + str(Y[i]))
print("Top Matching Documents: " + str(R))
print("Top Matching Documents (Associated Class Labels: " + str(Y[R]))
print(" Precision: " + str(Metrics.RetrievalPrecision(Y[i], Y[R])) + " Recall: " + str(Metrics.RetrievalRecall(Y[i], Y[R], Y, TopK)))
precision = precision / N
recall = recall / N
print("Average Precision: " + str(precision) + " Average Recall: " + str(recall))
return (precision, recall)