本文整理汇总了Python中sklearn.decomposition.LatentDirichletAllocation.transform方法的典型用法代码示例。如果您正苦于以下问题:Python LatentDirichletAllocation.transform方法的具体用法?Python LatentDirichletAllocation.transform怎么用?Python LatentDirichletAllocation.transform使用的例子?那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类sklearn.decomposition.LatentDirichletAllocation
的用法示例。
在下文中一共展示了LatentDirichletAllocation.transform方法的15个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: score_lda
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def score_lda(src, dst):
##read sentence pairs to two lists
b1 = []
b2 = []
lines = 0
with open(src) as p:
for i, line in enumerate(p):
s = line.split('\t')
b1.append(s[0])
b2.append(s[1][:-1]) #remove \n
lines = i + 1
vectorizer = CountVectorizer()
vectors=vectorizer.fit_transform(b1 + b2)
lda = LatentDirichletAllocation(n_topics=n_topics, max_iter=5,
learning_method='online', learning_offset=50.,
random_state=0)
X = lda.fit_transform(vectors)
print X.shape
b1_v = vectorizer.transform(b1)
b2_v = vectorizer.transform(b2)
b1_vecs = lda.transform(b1_v)
b2_vecs = lda.transform(b2_v)
res = [round(5*(1 - spatial.distance.cosine(b1_vecs[i], b2_vecs[i])),2) for i in range(lines)]
with open(dst, 'w') as thefile:
thefile.write("\n".join(str(i) for i in res))
示例2: applyLDA2
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def applyLDA2(self, number_of_clusters, country_specific_tweets):
train, feature_names = self.extractFeatures(country_specific_tweets,False)
name = "lda"
if self.results:
print("Fitting LDA model with tfidf", end= " - ")
t0 = time()
lda = LatentDirichletAllocation(n_topics=number_of_clusters, max_iter=5,
learning_method='online', learning_offset=50.,
random_state=0)
lda.fit(train)
if self.results:
print("done in %0.3fs." % (time() - t0))
parameters = lda.get_params()
topics = lda.components_
doc_topic = lda.transform(train)
top10, labels = self.printTopicCluster(topics, doc_topic, feature_names)
labels = numpy.asarray(labels)
if self.results:
print("Silhouette Coefficient {0}: {1}".format(name, metrics.silhouette_score(train, labels)))
return name, parameters, top10, labels
示例3: get_features
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def get_features(vocab):
vectorizer_head = TfidfVectorizer(vocabulary=vocab, use_idf=False, norm='l2')
X_train_head = vectorizer_head.fit_transform(headlines)
vectorizer_body = TfidfVectorizer(vocabulary=vocab, use_idf=False, norm='l2')
X_train_body = vectorizer_body.fit_transform(bodies)
# calculates n most important topics of the bodies. Each topic contains all words but ordered by importance. The
# more important topic words a body contains of a certain topic, the higher its value for this topic
lda_body = LatentDirichletAllocation(n_topics=n_topics, learning_method='online', random_state=0, n_jobs=3)
print("latent_dirichlet_allocation_cos: fit and transform body")
t0 = time()
lda_body_matrix = lda_body.fit_transform(X_train_body)
print("done in %0.3fs." % (time() - t0))
print("latent_dirichlet_allocation_cos: transform head")
# use the lda trained for body topcis on the headlines => if the headlines and bodies share topics
# their vectors should be similar
lda_head_matrix = lda_body.transform(X_train_head)
#print_top_words(lda_body, vectorizer_body.get_feature_names(), 100)
print('latent_dirichlet_allocation_cos: calculating cosine distance between head and body')
# calculate cosine distance between the body and head
X = []
for i in range(len(lda_head_matrix)):
X_head_vector = np.array(lda_head_matrix[i]).reshape((1, -1)) #1d array is deprecated
X_body_vector = np.array(lda_body_matrix[i]).reshape((1, -1))
cos_dist = cosine_distances(X_head_vector, X_body_vector).flatten()
X.append(cos_dist.tolist())
return X
示例4: topicmodel
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def topicmodel( comments ):
_texts = []
texts = []
for c in comments:
c = c['text']
_texts.append( c )
texts.append( c )
tf_vectorizer = CountVectorizer(
max_df=.20,
min_df=10,
stop_words = stopwords )
texts = tf_vectorizer.fit_transform( texts )
## test between 2 and 20 topics
topics = {}
for k in range(2, 10):
print "Testing", k
model = LatentDirichletAllocation(
n_topics= k ,
max_iter=5,
learning_method='batch',
learning_offset=50.,
random_state=0
)
model.fit( texts )
ll = model.score( texts )
topics[ ll ] = model
topic = max( topics.keys() )
ret = collections.defaultdict( list )
## ugly, rewrite some day
model = topics[ topic ]
## for debug pront chosen models' names
feature_names = tf_vectorizer.get_feature_names()
for topic_idx, topic in enumerate(model.components_):
print "Topic #%d:" % topic_idx
print " ".join( [feature_names[i].encode('utf8') for i in topic.argsort()[:-5 - 1:-1]])
print
for i, topic in enumerate( model.transform( texts ) ):
topic = numpy.argmax( topic )
text = _texts[ i ].encode('utf8')
ret[ topic ].append( text )
return ret
示例5: LDA
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def LDA(matrix,preserve,n_topics=100):
lda = LatentDirichletAllocation(n_topics=n_topics, max_iter=10,
learning_method='online', learning_offset=50.,
random_state=randint(1,100))
lda.fit(matrix[preserve])
topic_model=lda.transform(matrix)
return topic_model
示例6: find_topics
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def find_topics(df_train, df_test, n_topics):
#http://scikit-learn.org/stable/auto_examples/applications/topics_extraction_with_nmf_lda.html
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.decomposition import LatentDirichletAllocation
# Use tf (raw term count) features for LDA.
print("Extracting character frequency features for topic modeling...")
#Need to create a dtm with combined (train/test) vocabulary in columns
n_train = df_train.shape[0]
df_combined = df_train.copy(deep = True).append(df_test.copy(deep = True))
vectorizer = CountVectorizer(decode_error = 'strict', analyzer = 'char')
corpus_combined = df_combined.loc[:,'text_read']
dtm_combined = vectorizer.fit_transform(corpus_combined)
#split the train and test data again to ensure we only use test set for
#supervised cross-validated learning
dtm_train = dtm_combined[:n_train,:]
dtm_test = dtm_combined[n_train:,:]
print("Fitting LDA models with character frequency features...")
#This requires sklearn.__version__ to be 0.17.X or greater
lda = LatentDirichletAllocation(n_topics=n_topics, learning_method='batch',
random_state=0)
#fit to the training document term matrix
lda.fit(dtm_train)
#create topic 'names' and columns in dataframe
topic_names = []
for i in range(0, n_topics):
name = 't' + str(i+1)
topic_names.append(name)
df_train.loc[:, name] = 0.0
df_test.loc[:, name] = 0.0
df_train.loc[:, topic_names] = lda.transform(dtm_train)
df_test.loc[:, topic_names] = lda.transform(dtm_test)
#normalize these topic features
df_train = normalize_features(df_train, topic_names)
df_test = normalize_features(df_test, topic_names)
return df_train
示例7: test_lda_fit_transform
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def test_lda_fit_transform(method):
# Test LDA fit_transform & transform
# fit_transform and transform result should be the same
rng = np.random.RandomState(0)
X = rng.randint(10, size=(50, 20))
lda = LatentDirichletAllocation(n_components=5, learning_method=method,
random_state=rng)
X_fit = lda.fit_transform(X)
X_trans = lda.transform(X)
assert_array_almost_equal(X_fit, X_trans, 4)
示例8: LDA
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
class LDA():
def __init__(self, args=None, from_file=None):
# Initialize LDA model from either arguments or a file. If both are
# provided, file will be used.
assert args or from_file, 'Improper initialization of LDA model'
if from_file is not None:
with open(from_file, 'rb') as f:
self.model, self.vectorizer = pickle.load(f, encoding='latin1')
else: # training for the first time
self.vectorizer = TfidfVectorizer(lowercase=False, token_pattern=u'[^;]+')
self.alpha = args.alpha
self.beta = args.beta
self.ntopics = args.ntopics
self.model = None
def top_words(self, n):
features = self.vectorizer.get_feature_names()
words = [OrderedDict([(features[i], topic[i]) for i in topic.argsort()[:-n - 1:-1]])
for topic in self.model.components_]
return words
def train(self, docs):
data = [';'.join(bow) for bow in docs]
vect = self.vectorizer.fit_transform(data)
self.alpha = self.alpha if self.alpha is not None else 50./self.ntopics
self.beta = self.beta if self.beta is not None else 200./len(self.vectorizer.vocabulary_)
print('{} words in vocabulary'.format(len(self.vectorizer.vocabulary_)))
print('Training LDA with {} topics, {} alpha, {} beta'.format(self.ntopics, self.alpha, self.beta))
self.model = LatentDirichletAllocation(self.ntopics,
doc_topic_prior=self.alpha, topic_word_prior=self.beta,
learning_method='batch', max_iter=100,
verbose=1, evaluate_every=1,
max_doc_update_iter=100, mean_change_tol=1e-5)
self.model.fit(vect)
# normalizing does not change subsequent inference, provided no further training is done
self.model.components_ /= self.model.components_.sum(axis=1)[:, np.newaxis]
def infer(self, docs):
data = [';'.join(bow) for bow in docs]
vect = self.vectorizer.transform(data)
dist = self.model.transform(vect)
assert vect.shape[0] == dist.shape[0]
# NOTE: if a document is empty, this method returns a zero topic-dist vector
samples = [list(doc_topic_dist) if m.nnz > 0 else ([0.] * self.model.n_components)
for m, doc_topic_dist in zip(vect, dist)]
return samples
示例9: get_topics
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def get_topics(n_topics):
t0 = time()
tf = np.genfromtxt('tf.txt', delimiter=',')
print "feature laoded in %0.3fs." % (time() - t0)
lda = LatentDirichletAllocation(n_components=n_topics, max_iter=10,
learning_method='online', learning_offset=50.,
random_state=0)
t0 = time()
lda.fit(tf.T)
doc_topic = lda.transform(tf.T)
print "lda done in %0.3fs." % (time() - t0)
#tfidf = np.genfromtxt('tfidf.txt', delimiter=',')
#nmf = NMF(n_components=n_topics, random_state=1, alpha=.1, l1_ratio=.5).fit(tfidf)
#doc_topic = nmf.transform(tfidf)
plt.imshow(doc_topic, cmap='hot', interpolation='nearest')
plt.show()
示例10: save_clusters
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
# Save the clusters
tf_feature_names = tf_vectorizer.get_feature_names()
for x in save_clusters(lda, tf_feature_names, n_top_words):
writer.writerow([str(x[0]),str(x[1])])
# # You can also pring out the clusters if you want to see them
# print_clusters(lda, tf_feature_names, n_top_words)
# Now match up the records with the best fit clusters & corresponding keywords
with open('records_to_ldaclusters_v2.csv', 'wb') as f2:
writer = csv.writer(f2)
writer.writerow(["record_index","record_text","five_best_clusters","suggested_keywords"])
# Restart the clock
t0 = time()
print("Finding the best keywords for each record and writing up results...")
results = lda.transform(tf)
for i in range(len(results)):
try:
best_results = (-results[i]).argsort()[:5]
keywords = []
for x in np.nditer(best_results):
keywords.extend(get_words(tf_feature_names, x))
flattened = " ".join(keywords)
writer.writerow([i, noaa_samples[i], best_results, flattened])
#TODO => need to figure out the Unicode Error
except UnicodeEncodeError: pass
print("done in %0.3fs." % (time() - t0))
示例11:
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
#topic matrix W: K x V
#components[i,j]: topic i, word j
topics = lda_vb.components_
f = plt.figure()
plt.matshow(topics, cmap = 'gray')
plt.gca().set_aspect('auto')
plt.title('learned topic matrix')
plt.ylabel('topics')
plt.xlabel('dictionary')
plt.show()
f.savefig('./figures/topic.png')
#topic proportions matrix: D x K
#note: np.sum(H, axis=1) is not 1
H = lda_vb.transform(A_tfidf_sp)
f = plt.figure()
plt.matshow(H, cmap = 'gray')
plt.gca().set_aspect('auto')
plt.show()
plt.title('topic proportions')
plt.xlabel('topics')
plt.ylabel('documents')
f.savefig('./figures/proportions.png')
#compute perplexity
print "perplexity: %.2f" % lda_vb.perplexity(A_tfidf_sp)
plot_perplexity_iter(A_tfidf_sp, num_topics)
plot_perplexity_topics(A_tfidf_sp)
plot_perplexity_batch(A_tfidf_sp, A_tfidf_sp.shape[0])
示例12: main
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
def main(trace_fpath, leaveout=0.3):
leaveout = float(leaveout)
df = pd.read_csv(trace_fpath, sep="\t", names=["dt", "u", "s", "d"])
num_lines = len(df)
to = int(num_lines - num_lines * leaveout)
df_train = df[:to]
df_test = df[to:]
documents_train_right = OrderedDict()
documents_train_left = OrderedDict()
tokens_train = set()
for _, u, s, d in df_train.values:
u = str(u)
s = str(s)
d = str(d)
if u not in documents_train_right:
documents_train_right[u] = []
documents_train_left[u] = []
documents_train_right[u].append(s)
documents_train_left[u].append(d)
tokens_train.add(s)
tokens_train.add(d)
for u in documents_train_right:
documents_train_right[u] = "\t".join(documents_train_right[u])
documents_train_left[u] = "\t".join(documents_train_left[u])
vectorizer = CountVectorizer(tokenizer=lambda x: x.split("\t"), vocabulary=tokens_train)
X_train_counts = vectorizer.fit_transform(documents_train_right.values())
Y_train_counts = vectorizer.transform(documents_train_left.values())
lda_model = LatentDirichletAllocation(n_topics=10, n_jobs=-1)
lda_model.fit(X_train_counts)
Theta_zh = lda_model.transform(X_train_counts).T
ph = X_train_counts.sum(axis=1)
pz = np.asarray(Theta_zh.dot(ph))[:, 0]
Psi_oz = lda_model.components_.T
pz = pz / pz.sum()
Psi_zo = (Psi_oz * pz).T
# Normalize matrices
Psi_oz = Psi_oz / Psi_oz.sum(axis=0)
Psi_zo = Psi_zo / Psi_zo.sum(axis=0)
X_train_probs = []
Y_train_probs = []
for _, u, s, d in df_train.values:
if str(s) in vectorizer.vocabulary_ and str(d) in vectorizer.vocabulary_:
id_s = vectorizer.vocabulary_.get(str(s))
id_d = vectorizer.vocabulary_.get(str(d))
X_train_probs.append(Psi_zo[:, id_s])
Y_train_probs.append(Psi_zo[:, id_d])
X_train_probs = np.array(X_train_probs)
Y_train_probs = np.array(Y_train_probs)
P_zz = lstsq(X_train_probs, Y_train_probs)[0].T
# numerical errors, expected as in paper.
P_zz[P_zz < 0] = 0
I = Psi_oz.dot(P_zz)
I = I / I.sum(axis=0)
probs_tmlda = {}
probs_lda = {}
ll_tmlda = 0.0
ll_lda = 0.0
n = 0
for _, u, s, d in df_test.values:
u = str(u)
s = str(s)
d = str(d)
if s in vectorizer.vocabulary_ and d in vectorizer.vocabulary_:
id_s = vectorizer.vocabulary_.get(s)
id_d = vectorizer.vocabulary_.get(d)
if (id_d, id_s) not in probs_tmlda:
probs_tmlda[id_d, id_s] = (Psi_zo[:, id_s] * I[id_s]).sum()
probs_lda[id_d, id_s] = (Psi_zo[:, id_s] * Psi_oz[id_s]).sum()
if probs_tmlda[id_d, id_s] != 0:
ll_tmlda += np.log(probs_tmlda[id_d, id_s])
if probs_lda[id_d, id_s] != 0:
ll_lda += np.log(probs_lda[id_d, id_s])
n += 1
print(ll_tmlda, ll_lda)
print(ll_tmlda / n, ll_lda / n)
print(n)
示例13: __init__
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
class BuildLda:
def __init__(self, print_list=True):
# Create dictionary
self.dictionary = Dictionary()
self.topics = ['Topic {}'.format(i) for i in range(1,31)]
self.print_list = print_list
def build_object(self):
self.build_model()
self.transform_set()
self.build_nearest_neighbours()
def build_model(self):
if self.print_list:
print('Building LDA')
strings = JobDescription.objects.values('url', 'body')
data_samples = []
seen_strings = set()
for string in strings:
if string['body'] not in seen_strings:
seen_strings.add(string['body'])
data_samples.append({'url': string['url'], 'string': self.dictionary.clean_string(string['body'])})
self.data_samples = DataFrame(data_samples)
n_features = 10000
n_topics = 15
n_top_words = 10
max_iter = 40
self.tf_vectorizer = CountVectorizer(max_features=n_features,
stop_words='english')
tf = self.tf_vectorizer.fit_transform(self.data_samples['string'])
self.lda = LatentDirichletAllocation(n_topics=n_topics, max_iter=max_iter,
learning_method='online')
self.lda.fit(tf)
if self.print_list:
print()
print("\nTopics in LDA model:")
tf_feature_names = self.tf_vectorizer.get_feature_names()
self.create_word_topics(self.lda, tf_feature_names)
if self.print_list:
self.print_top_words(self.lda, tf_feature_names, n_top_words)
def test_single_doc(self, string):
data_samples = DataFrame([{'string': self.dictionary.clean_string(string)}])
test = self.tf_vectorizer.transform(data_samples['string'])
lda_result = self.lda.transform(test)
top_tags = []
return_value = {'lda_result': lda_result, 'tags': []}
index_set = sorted(range(len(lda_result[0])), key=lambda i: lda_result[0][i], reverse=True)
position = 0
for index in index_set:
return_value['tags'].append({'tag': self.topics[index], 'position': position, 'score': lda_result[0][index]})
top_tags.append(self.topics[index])
position += 1
return return_value
def transform_set(self):
if self.print_list:
print('Getting LDA Transformation')
vectorizor_data = self.tf_vectorizer.transform(self.data_samples['string'])
self.results = self.lda.transform(vectorizor_data)
def build_nearest_neighbours(self):
if self.print_list:
print('Build Nearest Neighbours')
self.nbrs = NearestNeighbors(n_neighbors=10, algorithm='ball_tree').fit(self.results)
def get_neighbours(self, string, print=False):
return_result = self.test_single_doc(string)
return_result['distances'], return_result['indices'] = self.nbrs.kneighbors(return_result['lda_result'])
if print:
self.print_neighbours(return_result['indices'][0])
return_result['neighbours'] = self.return_neighbours(return_result['indices'][0], return_result['distances'][0])
return {'tags': return_result['tags'], 'neighbours': return_result['neighbours']}
def print_neighbours(self, indices):
print('Closest 10 jobs:')
for indice in indices:
url = self.data_samples.get_value(indice, 'url')
print('http://www.seek.com.au%s' % url)
def return_neighbours(self, indices, distances):
return_value = []
for index in range(len(indices)):
url = self.data_samples.get_value(indices[index], 'url')
return_value.append({'url': 'http://www.seek.com.au{}'.format(url), 'distance': distances[index]})
return return_value
def print_top_words(self, model, feature_names, n_top_words):
for topic_idx, topic in enumerate(model.components_):
print(self.topics[topic_idx]+": "+" ".join([feature_names[i]
#.........这里部分代码省略.........
示例14: enumerate
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
for i, each in enumerate(data): # 一级分类样本数5889 二级分类5887
if each[TAG_LEVEL].strip() == '':
continue
else:
rawdialogue.append(rawdata[i])
content.append(each[0])
tag.append(each[TAG_LEVEL])
total_acc = 0
for i in range(TIMES):
train_content, train_tag, train_raw, test_content, test_tag, test_raw = divideData(rawdialogue, content, tag, 0.2)
# 得到单词-文档共现矩阵
vectorizer = CountVectorizer(encoding='unicode', stop_words='english', max_features=N_FEATURES)
train_data = vectorizer.fit_transform(train_content)
test_data = vectorizer.fit_transform(test_content) # [n_samples, n_features]
model = LDA(n_topics=N_TOPICS, batch_size=64)
model.fit(train_data)
dt_matrix = model.transform(train_data)
test_dt_matrix = model.transform(test_data)
svc = SVC(C=0.99, kernel='linear')
svc = svc.fit(dt_matrix, train_tag)
pred = svc.predict(test_dt_matrix)
acc = np.round(np.mean(pred == test_tag), 4)
total_acc += acc
print 'LDA分类器的准确率: %.4f' % acc
print 'average accuary: ', total_acc / TIMES
示例15: range
# 需要导入模块: from sklearn.decomposition import LatentDirichletAllocation [as 别名]
# 或者: from sklearn.decomposition.LatentDirichletAllocation import transform [as 别名]
for i in range(TIMES):
train_content, train_tag, train_raw, test_content, test_tag, test_raw = divideData(rawdialogue, content, tag, 0.2)
# 得到单词-文档共现矩阵
vectorizer = CountVectorizer(encoding='unicode', stop_words='english', max_features=N_FEATURES)
train_data = vectorizer.fit_transform(train_content)
train_tag = np.array(train_tag)
test_data = vectorizer.fit_transform(test_content) # [n_samples, n_features]
model = LDA(n_topics=N_TOPICS, max_iter=5, batch_size=128)
model.fit(train_data)
train_data_distr = model.transform(train_data)
pred_tag = train_data_distr.argmax(axis=1)
# 投票
id2class = dict()
for idx in range(N_TOPICS):
idxs = np.where(pred_tag == idx)[0]
# print Counter(train_tag[idxs])
id2class[idx] = Counter(train_tag[idxs]).most_common(1)[0][0]
print id2class
doc_topic_distr = model.transform(test_data) # [n_samples, n_topics]
class_id = doc_topic_distr.argmax(axis=1)
pred = [id2class[each] for each in class_id]
pred=np.array(pred)
test_tag=np.array(test_tag)
acc=np.mean(pred==test_tag)