def cross_validation():
_, vocab = train()
X_tensor = np.array([np.sum([np.eye(vocab.vocab_size+1)[wid] for wid in sentence], axis=0) for sentence in vocab.sentences])
Y_tensor = np.array(vocab.labels)
preds = cross_val_predict(LogisticRegression(), X_tensor, Y_tensor, cv=5)
print('Precision: ', precision_score(vocab.labels, preds))
print('Recall: ', recall_score(vocab.labels, preds))
print('F-measure: ', f1_score(vocab.labels, preds))
def labeling():
model, vocab = train()
X_test = np.array([
np.sum([np.eye(vocab.vocab_size + 1)[wid] for wid in sentence], axis=0)
for sentence in vocab.sentences
])
true_label = [label[0] for label in vocab.labels]
pred_label = model.predict(X_test)
pred_prob = model.predict_proba(X_test)
print('true|pred|prob')
[
print('{}\t{}\t{}'.format(true, pred, max(prob)))
for true, pred, prob in zip(true_label, pred_label, pred_prob)
]
return true_label, pred_label, pred_prob
def predict():
model, vocab = train()
X_test = np.array([np.sum([np.eye(vocab.vocab_size+1)[wid] for wid in sentence], axis=0) for sentence in vocab.sentences])
print(model.predict(X_test))
# 余ったものを最後尾に追加
splitList[-2] = splitList[-2] + splitList[-1]
splitList = splitList[:-1]
print('{}分割交差検定を開始します。'.format(len(splitList)))
accuracys, presicions, recalls, F_measures = np.zeros(k * 4).reshape(
4, k) # それぞれk個の要素を持ったnp.arrayである
for i in range(k):
print('i = ' + str(i))
testList = splitList[i]
trainList = [
line for l in splitList[:i] + splitList[i + 1:] for line in l
] # flatten
wdic = train(trainList)
paramdic = {'TP': 0, 'FP': 0, 'TN': 0, 'FN': 0}
for line in testList:
ans, odds = likelihood(line, wdic)
prediction = '+1' if odds > border else '-1'
paramdic[classify('{}\t{}\t{}'.format(ans, prediction, odds))] += 1
accuracys[i], presicions[i], recalls[i], F_measures[i] \
= ret_score(*map(lambda x: paramdic[x], ['TP', 'FP', 'TN', 'FN']))
print('正解率:\t{}'.format(np.average(accuracys)))
print('適合率:\t{}'.format(np.average(presicions)))
print('再現率:\t{}'.format(np.average(recalls)))
print('F1スコア:\t{}'.format(np.average(F_measures)))
def main():
_, vocab = train()
X_tensor = np.sum([np.sum([np.eye(vocab.vocab_size+1)[wid] for wid in sentence], axis=0) for sentence in vocab.sentences], axis=0)
[print(vocab.id2word[idx]) for idx in X_tensor.argsort()[::-1][:10]]
[print(vocab.id2word[idx]) for idx in X_tensor.argsort()[:10]]