def feature_pred(features, chik, ldak):
global users
wn.ensure_loaded()
facts = gt.get_fact_topics(DIR)
if NEW_DATA:
users = gt.get_users(DIR)
transactions = gt.get_transactions(DIR)
print(transactions.describe())
tr_hsh = transactions['fact'].values
# if castillo: comment cond2 out
cond = facts['hash'].isin(tr_hsh)
cond2 = facts['true'] == 1 | facts['true'] == 0
facts = facts[cond & cond2]
facts = Parallel(n_jobs=num_jobs)(delayed(get_features)(
fact, transactions[transactions['fact'] == fact['hash']], [
u for u in users if int(u.user_id) in list(transactions[
transactions['fact'] == fact['hash']]['user_id'].values)
]) for idx, fact in facts.iterrows())
facts = pd.DataFrame(facts)
with open('model_data/feature_data', 'wb') as tmpfile:
pickle.dump(facts, tmpfile)
else:
with open('model_data/feature_data', 'rb') as tmpfile:
facts = pickle.load(tmpfile)
print(facts[list(features)].describe())
X = facts[list(features)].values
y = facts['y'].values
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20)
std_clf = make_pipeline(StandardScaler(), PCA(n_components=ldak),
SVC(C=1, gamma=1))
std_clf.fit(X_train, y_train)
pred_test_std = std_clf.predict(X_test)
precision, recall, fscore, sup = precision_recall_fscore_support(
y_test, pred_test_std, average='macro')
score = metrics.accuracy_score(y_test, pred_test_std)
print("Accuracy: %0.3f, Precision: %0.3f, Recall: %0.3f, F1 score: %0.3f" %
(score, precision, recall, fscore))
acc_scores = cross_val_score(std_clf, X, y, cv=3)
pr_scores = cross_val_score(std_clf, X, y, scoring='precision', cv=3)
re_scores = cross_val_score(std_clf, X, y, scoring='recall', cv=3)
f1_scores = cross_val_score(std_clf, X, y, scoring='f1', cv=3)
print("\t Cross validated Accuracy: %0.3f (+/- %0.3f)" %
(acc_scores.mean(), acc_scores.std() * 2))
print("\t Cross validated Precision: %0.3f (+/- %0.3f)" %
(pr_scores.mean(), pr_scores.std() * 2))
print("\t Cross validated Recall: %0.3f (+/- %0.3f)" %
(re_scores.mean(), re_scores.std() * 2))
print("\t Cross validated F1: %0.3f (+/- %0.3f)" %
(f1_scores.mean(), f1_scores.std() * 2))
return acc_scores.mean()
def main():
global bow_corpus
global word_to_idx
wn.ensure_loaded()
if NEW_CORPUS:
bow_corpus = build_bow_corpus(get_users())
save_corpus(bow_corpus)
else:
bow_corpus = get_corpus()
bow_corpus_tmp = [w[0] for w in bow_corpus.items() if w[1] > 2]
word_to_idx = {k: idx for idx, k in enumerate(bow_corpus_tmp)}
idx_to_word = {idx: k for k, idx in word_to_idx.items()}
users = get_users()
facts = gt.get_fact_topics()
transactions = gt.get_transactions()
users_df = pd.DataFrame([vars(u) for u in users])
print(users_df.describe())
print(users_df[users_df['stance'] == 0].describe())
print(users_df[users_df['stance'] == 1].describe())
print(users_df[users_df['stance'] == 2].describe())
print(users_df[users_df['stance'] == 3].describe())
users_df['f_t'] = users_df['fact'].map(
lambda x: facts[facts['hash'] == x]['true'].values[0])
c_true = users_df['f_t'] == '1'
c_fal = users_df['f_t'] == '0'
c_fal1 = users_df['f_t'] == 0
c_den = users_df['stance'] == 0
c_sup = users_df['stance'] == 1
print(users_df[c_true & c_sup].describe())
print(users_df[c_fal | c_fal1][c_den].describe())
print(users_df[c_fal | c_fal1][c_sup].describe())
print(users_df[c_true & c_den].describe())
print(users_df[users_df['was_correct'] == 1].describe())
print(users_df[users_df['was_correct'] == 0].describe())
print(len([t for u in users for t in u.tweets if u.tweets is not None]))
corpus_analysis(bow_corpus, word_to_idx, idx_to_word)
# temporal_analysis(get_users())
cluster_users_on_tweets(users, word_to_idx, idx_to_word)
def main():
global users
wn.ensure_loaded()
facts = gt.get_fact_topics(DIR)
features = [
'avg_mentions', 'avg_emoticons', 'avg_links', 'avg_questionM',
'avg_personal_pronoun_first', 'avg_sent_pos', 'avg_sent_neg',
'avg_sentiment', 'fr_has_url', 'share_most_freq_author', 'lvl_size',
'avg_followers', 'avg_friends', 'avg_status_cnt', 'avg_reg_age'
]
if NEW_DATA:
users = gt.get_users(DIR)
transactions = gt.get_transactions(DIR)
print(transactions.describe())
tr_hsh = transactions['fact'].values
cond = facts['hash'].isin(tr_hsh)
facts = facts[cond]
facts = pd.DataFrame([
get_features(fact, transactions, users)
for idx, fact in facts.iterrows() if fact['true'] != 'unknown'
])
with open('model_data/castillo_data', 'wb') as tmpfile:
pickle.dump(facts, tmpfile)
else:
with open('model_data/castillo_data', 'rb') as tmpfile:
facts = pickle.load(tmpfile)
print(facts.describe())
X = facts[list(features)].values
y = facts['y'].values
fig = plt.figure()
fig.subplots_adjust(hspace=0.4, wspace=0.4)
for i in range(1, len(features) + 1):
ax = fig.add_subplot(3, 5, i)
sns.boxplot(x="y", y=features[i - 1], data=facts, palette="Set3")
#plt.show()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20)
std_clf = make_pipeline(StandardScaler(),
DecisionTreeClassifier(random_state=42))
std_clf.fit(X_train, y_train)
pred_test_std = std_clf.predict(X_test)
precision, recall, fscore, sup = precision_recall_fscore_support(
y_test, pred_test_std, average='macro')
score = metrics.accuracy_score(y_test, pred_test_std)
print(
"Random split: Accuracy: %0.3f, Precision: %0.3f, Recall: %0.3f, F1 score: %0.3f"
% (score, precision, recall, fscore))
acc_scores = cross_val_score(std_clf, X, y, cv=3)
pr_scores = cross_val_score(std_clf, X, y, scoring='precision', cv=3)
re_scores = cross_val_score(std_clf, X, y, scoring='recall', cv=3)
f1_scores = cross_val_score(std_clf, X, y, scoring='f1', cv=3)
print("\t Cross validated Accuracy: %0.3f (+/- %0.3f)" %
(acc_scores.mean(), acc_scores.std() * 2))
print("\t Cross validated Precision: %0.3f (+/- %0.3f)" %
(pr_scores.mean(), pr_scores.std() * 2))
print("\t Cross validated Recall: %0.3f (+/- %0.3f)" %
(re_scores.mean(), re_scores.std() * 2))
print("\t Cross validated F1: %0.3f (+/- %0.3f)" %
(f1_scores.mean(), f1_scores.std() * 2))