def predict_titles(titles):
lr_model, data, vectorizer = train()
print(lr_model.predict(vectorizer.transform(titles).toarray()))
print([
max(i) for i in lr_model.predict_proba(
vectorizer.transform(titles).toarray())
])
def knock57():
lr, data, vectorizer = train()
inverse_vectorizer_vocabulary_ = {
v: k
for k, v in vectorizer.vocabulary_.items()
}
for cnt, class_name in enumerate(lr.classes_):
lr.coef_[cnt]
print(class_name)
for i in heapq.nlargest(10, lr.coef_[cnt]):
index1 = np.where(lr.coef_[cnt] == i)
print(inverse_vectorizer_vocabulary_[index1[0][0]], ":", i)
for i in heapq.nsmallest(10, lr.coef_[cnt]):
index1 = np.where(lr.coef_[cnt] == i)
print(inverse_vectorizer_vocabulary_[index1[0][0]], ":", i)
print()
def knock56():
lr, data, vectorizer = train()
# 一行目がprecision 、二行目がrecall 、、、
# 一列目がカテゴリ1、二行目がカテゴリ2 、、、
pprint(
precision_recall_fscore_support(y_true=data[1][1],
y_pred=lr.predict(data[1][0])))
# precision, recall, f-score, support の順番
print(
"macro:",
precision_recall_fscore_support(y_true=data[1][1],
y_pred=lr.predict(data[1][0]),
average="macro"))
print(
"micro:",
precision_recall_fscore_support(y_true=data[1][1],
y_pred=lr.predict(data[1][0]),
average="micro"))
def knock55():
lr, data, vectorizer = train()
print(confusion_matrix(y_true=data[0][1], y_pred=lr.predict(data[0][0])))
print(confusion_matrix(y_true=data[1][1], y_pred=lr.predict(data[1][0])))
valid = open('valid.feature.txt')
valid_ftr, valid_label = read_data(valid)
test = open('test.feature.txt')
test_ftr, test_label = read_data(test)
vectorizer = vectorize(train_ftr)
joblib.dump(vectorizer, 'vectorizer.pkl')
x_train = vectorizer.transform(train_ftr)
y_train = train_label
regularization = []
for c in range(-5,5):
# train with regularization
model_name = 'model_reg_10**' + str(c) + '.pkl'
train(x_train, y_train, model_name, c)
reg = '10e' + str(c)
# calculate accuracy on train, valid, test
acc_train = accuracy(train_ftr, train_label, model_name,'vectorizer.pkl')
acc_valid = accuracy(valid_ftr, valid_label, model_name,'vectorizer.pkl')
acc_test= accuracy(test_ftr, test_label, model_name,'vectorizer.pkl')
regularization.append(reg + '\t' + str(round(acc_train,6)) + '\t' + str(round(acc_valid,6)) + '\t' + str(round(acc_test,6)))
reg_val, acc_tr, acc_val, acc_ts = [],[],[],[]
for i in regularization:
temp = i.strip().split('\t')
reg_val.append(float(temp[0]))
acc_tr.append(float(temp[1]))
acc_val.append(float(temp[2]))
def knock54():
lr, data, vectorizer = train()
print("train", lr.score(data[0][0], data[0][1]))
print("vaild", lr.score(data[1][0], data[1][1]))