num_cluster = clustering_control_param['num_training_cluster']
num_pos_cluster = num_cluster
num_neg_cluster = num_cluster
clustering_testdata = clustering_control_param['training_clustering_method']
clustered_pos = clustering_testdata(pos, num_pos_cluster)
clustered_neg = clustering_testdata(neg, num_neg_cluster)
X_train, Y_train = clustered_pos + clustered_neg, [1] * num_pos_cluster + [0] * num_neg_cluster
X_train, Y_train = np.array(X_train), np.array(Y_train)
trian_vec = vectorizer.fit_transform(X_train)
# clustering test data
from candidate_content import get_candidate
expanding_pos_content, expanding_neg_content = get_candidate()
expanding_pos_content, expanding_neg_content = np.array(expanding_pos_content), np.array(expanding_neg_content)
expanding_pos_content_vec, expanding_neg_content_vec = vectorizer.transform(
expanding_pos_content), vectorizer.transform(expanding_neg_content)
# 加载测试资料
from Utils import load_test_data
X_test, Y_test = load_test_data()
# 下面代码临时执行，需要用时再执行，不用时注释，用来产生扩展的test data
# from candidate_content import get_candidate_dynamic
# get_candidate_dynamic(X_test, neg, 5, 'neg')
# get_candidate_dynamic(X_test, pos, 5, 'pos')
# exit()
# 临时执行代码结束，为了更好的组织代码结构，将此文件另存为一份dynamic_classifer.py
# 加载模型
clf = pickle.load(open("./acc_tmp/predict/classifier.p", "rb"))
logger.info('成功加载分类器模型')
# 向量化
from customed_vectorizer import StemmedTfidfVectorizer
from parameters import vectorizer_param as param
vectorizer = StemmedTfidfVectorizer(**param)
neg_clustered_texts_vec = vectorizer.fit_transform(neg_clustered_texts)
neg_extantion_content_vec = vectorizer.fit_transform(neg_extantion_content)
pos_clustered_texts_vec = vectorizer.fit_transform(pos_clustered_texts)
pos_extantion_content_vec = vectorizer.fit_transform(pos_extantion_content)
logger.info('向量化完成')
# 预测
predict_neg_clustered_texts_vec = clf.predict_proba(neg_clustered_texts_vec)[:, 1]
predict_neg_extantion_content_vec = clf.predict_proba(neg_extantion_content_vec)[:, 1]
predict_pos_clustered_texts_vec = clf.predict_proba(pos_clustered_texts_vec)[:, 1]
predict_pos_extantion_content_vec = clf.predict_proba(pos_extantion_content_vec)[:, 1]
from Utils import load_test_data
text, _ = load_test_data()
predict_testdata_without_clustering = clf.predict_proba(vectorizer.fit_transform(text))[:, 1]
logger.info('完成预测，即将保存')
# 保存结果
pickle.dump(predict_neg_clustered_texts_vec, open("./data/predict_dynamics/predict_neg_clustered_texts_vec.p", "wb"))
pickle.dump(predict_neg_extantion_content_vec, open("./data/predict_dynamics/predict_neg_extantion_content_vec.p", "wb"))
pickle.dump(predict_pos_clustered_texts_vec, open("./data/predict_dynamics/predict_pos_clustered_texts_vec.p", "wb"))
pickle.dump(predict_pos_extantion_content_vec, open("./data/predict_dynamics/predict_pos_extantion_content_vec.p", "wb"))
pickle.dump(predict_testdata_without_clustering, open("./data/predict_dynamics/predict_testdata_without_clustering.p", "wb"))
logger.info('完成保存')
plt.xlabel('pos distance')
plt.ylabel('neg distance')
plt.show()
plt.plot(range(0, length), expand_with_pos, 'r-', alpha=0.8)
plt.scatter(range(0,length), expand_with_pos, marker='o', c=colors)
plt.plot(range(0, length), pos_expand_content, 'r-', alpha=0.4)
plt.scatter(range(0,length), expand_with_neg, marker='o', c=colors)
plt.axhline(0.5, color='black')
plt.plot(range(0, length), expand_with_neg, 'g-', alpha=0.8)
plt.plot(range(0, length), neg_expand_content, 'g-', alpha=0.4)
plt.show()
import pickle
from Utils import load_test_data
_, true = load_test_data()
length = len(true)
colors = ['red' if true[i] == 1 else 'green' for i in range(0, length)]
g = {'c': colors, 'alpha': 0.7}
expand_with_pos = pickle.load(open("./data/predict_dynamics/predict_pos_clustered_texts_vec.p", "rb"))
expand_with_neg = pickle.load(open("./data/predict_dynamics/predict_neg_clustered_texts_vec.p", "rb"))
pos_expand_content = pickle.load(open("./data/predict_dynamics/predict_pos_extantion_content_vec.p", "rb"))
neg_expand_content = pickle.load(open("./data/predict_dynamics/predict_neg_extantion_content_vec.p", "rb"))
analysis_dynamics(expand_with_pos, expand_with_neg, pos_expand_content, neg_expand_content)
# predict
predict1 =[0 if ele==True else 1 for ele in ((1 - expand_with_pos) > expand_with_neg)]
from analysis import analysis_result as ar
ar(predict1,true)
ar(predict_lable_testdata_without_clustering,true)
else:
pos.append(X_train[i])
num_cluster = clustering_control_param['num_training_cluster']
num_pos_cluster = num_cluster
num_neg_cluster = num_cluster
clustering_testdata = clustering_control_param[
'training_clustering_method']
clustered_pos = clustering_testdata(pos, num_pos_cluster)
clustered_neg = clustering_testdata(neg, num_neg_cluster)
X_train, Y_train = clustered_pos + clustered_neg, [
1
] * num_pos_cluster + [0] * num_neg_cluster
X_train, Y_train = np.array(X_train), np.array(Y_train)
from Utils import load_test_data
X_test, Y_test = load_test_data()
if parameters['clustering_test_data'] == True and clustering_control_param[
'use_additional_texts'] == False:
clustering_test_data_method = clustering_control_param[
'clustering_test_data_method']
X_test, X_test_labels = clustering_test_data_method(
X_test, clustering_control_param['num_test_cluster'])
elif parameters[
'clustering_test_data'] == True and clustering_control_param[
'use_additional_texts'] == True:
# #另一种方法，clustering_texts_using_trainingset
clustering_test_data_method = clustering_control_param[
'clustering_test_data_method']
cluster_size = clustering_control_param['cluster_size']
if clustering_control_param['additional_texts'] == 'test_data':
alpha=0.6,
color='b')
plt.xlabel('Result')
plt.ylabel('Scores')
plt.title('Experiment analysis')
plt.xticks(index + bar_width,
('Accuracy', 'F', 'Precision', 'Recall', 'F'))
plt.ylim(0, 1)
plt.tight_layout()
plt.show()
if __name__ == "__main__":
# TRUE LABELS
from Utils import load_test_data
_, true = load_test_data()
# Predict labels
predict_label_1 = pickle.load(
open("./acc_tmp/predict/predict_label_1.p", "rb"))
predict_label_2 = pickle.load(
open("./acc_tmp/predict/predict_label_2.p", "rb"))
predict_label_3 = pickle.load(
open("./acc_tmp/predict/predict_label_3.p", "rb"))
predict_label_4 = pickle.load(
open("./acc_tmp/predict/predict_label_4.p", "rb"))
analysis_result(predict_label_1, true)
analysis_result(predict_label_2, true)
analysis_result(predict_label_3, true)
analysis_result(predict_label_4, true)
predict_without_clustering = pickle.load(
def train_model(clc_factory, X, Y):
print('start train_model...')
# 设置随机状态，来得到确定性的行为
cv = ShuffleSplit(n=len(X), n_iter=1, test_size=0.0, random_state=0)
# accuracy
scores = []
# AUC
pr_scores = []
# F1 score
f1 = []
# 暂时先用for吧，仅仅迭代一次
for train, test in cv:
# 测试方法一: do not clustering test data and training data
# '''
X_train, Y_train = X[train], Y[train]
X_test, Y_test = load_test_data()
X_test, Y_test = np.array(X_test), np.array(Y_test)
# clf = pickle.load(open("./acc_tmp/clf_all_data_noclustering.p", "rb"))
clf = clc_factory()
clf.fit(X_train, Y_train)
pickle.dump(clf, open("./acc_tmp/clf.p", "wb"))
# accuracy
test_score = clf.score(X_test, Y_test)
scores.append(test_score)
proba = clf.predict_proba(X_test)
precision, recall, pr_thresholds = precision_recall_curve(Y_test, proba[:, 1])
# AUC
aera_uc = auc(recall, precision)
pr_scores.append(aera_uc)
#F1_score
f1.append(f1_score(Y_test, clf.predict(X_test), average='macro'))
summary = (np.mean(scores), np.mean(pr_scores), np.mean(f1))
# Area Under Curve （曲线下面的面积）
print('正确率(Accuracy)：%.3f\nP/R AUC值：%.3f\nF值(Macro-F score)：%.3f' % (summary))
# 画图
pl.clf()
pl.plot(recall, precision, label='Precision-Recall curve')
pl.xlabel('Recall')
pl.ylabel('Precision')
pl.ylim([0.0, 1.05])
pl.xlim([0.0, 1.0])
pl.title('Precision-Recall Curve (AUC=%0.3f)' % aera_uc)
pl.legend(loc="lower left")
pl.show()
# '''
'''
# 测试方法2, do not use clustering in test data, changed to specified test data
X_train, Y_train = X[train], Y[train]
clf = clc_factory()
clf.fit(X_train, Y_train)
pickle.dump(clf, open("./acc_tmp/clf_all_data.p", "wb"))
X_test, Y_test=load_test_data()
X_test,X_test_labels=build_clustered_testdata(X_test)
# print(X_test_labels,len(X_test_labels))
X_test, Y_test=np.array(X_test), np.array(Y_test)
true_labels=Y_test
predict_labels=np.array(sentiment_map_cluster2tweets(clf.predict(X_test),X_test_labels))
precision,recall,fbeta_score,support=precision_recall_fscore_support(true_labels, predict_labels, average='binary')
print('精确度(Precision):%.3f\n召回率：%.3f\nF值: %.3f'%(precision,recall,fbeta_score))
# '''
'''
X_train, Y_train = np.array(X_train), np.array(Y_train)
trian_vec = vectorizer.fit_transform(X_train)
# clustering test data
from candidate_content import get_candidate
expanding_pos_content, expanding_neg_content = get_candidate()
expanding_pos_content, expanding_neg_content = np.array(
expanding_pos_content), np.array(expanding_neg_content)
expanding_pos_content_vec, expanding_neg_content_vec = vectorizer.transform(
expanding_pos_content), vectorizer.transform(expanding_neg_content)
# 加载测试资料
from Utils import load_test_data
X_test, Y_test = load_test_data()
# 下面代码临时执行，需要用时再执行，不用时注释，用来产生扩展的test data
from candidate_content import get_candidate_dynamic
# get_candidate_dynamic(X_test, neg, 8, 'neg')
# get_candidate_dynamic(X_test, pos, 8, 'pos')
# exit()
# 临时执行代码结束，为了更好的组织代码结构，将此文件另存为一份dynamic_classifer.py
from test_data_clustering import expand_text_list as expanding_method
expanded_texts_with_pos, expanded_texts_with_neg = expanding_method(
X_test,
expanding_pos_content), expanding_method(X_test, expanding_neg_content)
expanded_texts_vec_with_pos, expanded_texts_vec_with_neg = vectorizer.transform(
expanded_texts_with_pos), vectorizer.transform(expanded_texts_with_neg)
def train_model(clc_factory, X, Y):
print('start train_model...')
# 设置随机状态，来得到确定性的行为
cv = ShuffleSplit(n=len(X), n_iter=1, test_size=0.0, random_state=0)
# accuracy
scores = []
# AUC
pr_scores = []
# F1 score
f1 = []
# 暂时先用for吧，仅仅迭代一次
for train, test in cv:
# 测试方法一: do not clustering test data and training data
# '''
X_train, Y_train = X[train], Y[train]
X_test, Y_test = load_test_data()
X_test, Y_test = np.array(X_test), np.array(Y_test)
# clf = pickle.load(open("./acc_tmp/clf_all_data_noclustering.p", "rb"))
clf = clc_factory()
clf.fit(X_train, Y_train)
pickle.dump(clf, open("./acc_tmp/clf.p", "wb"))
# accuracy
test_score = clf.score(X_test, Y_test)
scores.append(test_score)
proba = clf.predict_proba(X_test)
precision, recall, pr_thresholds = precision_recall_curve(
Y_test, proba[:, 1])
# AUC
aera_uc = auc(recall, precision)
pr_scores.append(aera_uc)
#F1_score
f1.append(f1_score(Y_test, clf.predict(X_test), average='macro'))
summary = (np.mean(scores), np.mean(pr_scores), np.mean(f1))
# Area Under Curve （曲线下面的面积）
print('正确率(Accuracy)：%.3f\nP/R AUC值：%.3f\nF值(Macro-F score)：%.3f' %
(summary))
# 画图
pl.clf()
pl.plot(recall, precision, label='Precision-Recall curve')
pl.xlabel('Recall')
pl.ylabel('Precision')
pl.ylim([0.0, 1.05])
pl.xlim([0.0, 1.0])
pl.title('Precision-Recall Curve (AUC=%0.3f)' % aera_uc)
pl.legend(loc="lower left")
pl.show()
# '''
'''
# 测试方法2, do not use clustering in test data, changed to specified test data
X_train, Y_train = X[train], Y[train]
clf = clc_factory()
clf.fit(X_train, Y_train)
pickle.dump(clf, open("./acc_tmp/clf_all_data.p", "wb"))
X_test, Y_test=load_test_data()
X_test,X_test_labels=build_clustered_testdata(X_test)
# print(X_test_labels,len(X_test_labels))
X_test, Y_test=np.array(X_test), np.array(Y_test)
true_labels=Y_test
predict_labels=np.array(sentiment_map_cluster2tweets(clf.predict(X_test),X_test_labels))
precision,recall,fbeta_score,support=precision_recall_fscore_support(true_labels, predict_labels, average='binary')
print('精确度(Precision):%.3f\n召回率：%.3f\nF值: %.3f'%(precision,recall,fbeta_score))
# '''
'''
X_train, Y_train = clustered_pos + clustered_neg, [1] * num_pos_cluster + [0] * num_neg_cluster
X_train, Y_train = np.array(X_train), np.array(Y_train)
trian_vec = vectorizer.fit_transform(X_train)
# clustering test data
from candidate_content import get_candidate
expanding_pos_content, expanding_neg_content = get_candidate()
expanding_pos_content, expanding_neg_content = np.array(expanding_pos_content), np.array(expanding_neg_content)
expanding_pos_content_vec, expanding_neg_content_vec = vectorizer.transform(
expanding_pos_content), vectorizer.transform(expanding_neg_content)
# 加载测试资料
from Utils import load_test_data
X_test, Y_test = load_test_data()
# 下面代码临时执行，需要用时再执行，不用时注释，用来产生扩展的test data
from candidate_content import get_candidate_dynamic
# get_candidate_dynamic(X_test, neg, 8, 'neg')
# get_candidate_dynamic(X_test, pos, 8, 'pos')
# exit()
# 临时执行代码结束，为了更好的组织代码结构，将此文件另存为一份dynamic_classifer.py
from test_data_clustering import expand_text_list as expanding_method
expanded_texts_with_pos, expanded_texts_with_neg = expanding_method(X_test,
expanding_pos_content), expanding_method(X_test,
expanding_neg_content)
expanded_texts_vec_with_pos, expanded_texts_vec_with_neg = vectorizer.transform(
expanded_texts_with_pos), vectorizer.transform(expanded_texts_with_neg)