def test(**kwargs):
# ---------------------- 更新参数 ----------------------
opt = DefaultConfig()
opt.update(**kwargs)
opt.printf()
# ---------------------- 数据处理 ----------------------
# 获取数据
train, test = get_test_data(opt)
gc.collect()
# # 获取样本
# test_sample = get_sample(train, test, load=True)
# gc.collect()
# # 获取特征
# test_feat = get_feat(train, test_sample)
# gc.collect()
# 保存特征至文件
# test_feat.to_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_{}.hdf'.format(test.shape[0]), 'w', complib='blosc', complevel=5)
test_feat = pd.read_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_{}.hdf'.format(test.shape[0]))
test_feat = get_feat(train, test_feat)
gc.collect()
test_feat.to_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_{}_filter.hdf'.format(test.shape[0]), 'w', complib='blosc', complevel=5)
# ---------------------- 载入模型 ----------------------
# opt['model_name'] = 'lgb_1_90_all.pkl'
# gbm0, use_feat0 = load_model(opt)
opt['model_name'] = 'lgb_2017-09-23#20:14:52_0.58893.pkl'
gbm1, use_feat1 = load_model(opt)
# opt['model_name'] = 'lgb_2_300_top15.pkl'
# gbm2, use_feat2 = load_model(opt)
# opt['model_name'] = 'lgb_3_300_top10.pkl'
# gbm3, use_feat3 = load_model(opt)
# opt['model_name'] = 'lgb_4_300_top5.pkl'
# gbm4, use_feat4 = load_model(opt)
# ---------------------- 保存预测结果 -------------------
# test_feat.loc[:, 'pred'] = gbm0.predict(test_feat[use_feat0])
# gc.collect()
# res = test_feat[['orderid', 'geohashed_end_loc', 'pred']].sort_values(by=['orderid', 'pred'], ascending=False).groupby('orderid').head(25)
# res[['orderid', 'geohashed_end_loc']].to_hdf('/home/xuwenchao/dyj-storage/sample_25_{}_filter_leak_sample.hdf'.format(test.shape[0]), 'w', complib='blosc', complevel=5)
# gc.collect()
# test_feat.loc[:, 'pred'] = gbm1.predict(test_feat[use_feat1])
# test_feat[['orderid', 'geohashed_end_loc', 'pred']].to_hdf('/home/xuwenchao/dyj-storage/pred/pred_{}_0.58820.hdf'.format(test.shape[0]), 'w', complib='blosc', complevel=5)
res = predict(test_feat, use_feat1, gbm1)
test_feat[['orderid', 'geohashed_end_loc', 'pred']].to_hdf('/home/xuwenchao/dyj-storage/pred/pred_{}_0.58893.hdf'.format(test.shape[0]), 'w', complib='blosc', complevel=5)
gc.collect()
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
res_path = '{}/day{}_{}_wc_sample_0.58893.csv'.format(opt['result_dir'], opt['test_startday'], cur_time)
res.to_csv(res_path, index=False)
print('保存测试结果至：', res_path)
def val(**kwargs):
# ---------------------- 更新参数 ----------------------
opt = DefaultConfig()
opt.update(**kwargs)
opt.printf()
# ---------------------- 数据处理 ----------------------
# 获取数据
# train1, train2, train_test = get_train_data(opt)
# 获取样本
# train_sample = get_sample(train1, train2, load=True)
# 获取特征
# train_feat = get_feat(train_test, train_sample)
# gc.collect()
# train_feat.to_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_{}.hdf'.format(opt['startday']), 'w', complib='blosc', complevel=5)
train_feat = pd.read_hdf(
'/home/xuwenchao/dyj-storage/all-feat/feat_23_24_label.hdf')
# ---------------------- 载入模型 ----------------------
# opt['model_name'] = 'lgb_1_90_all.pkl'
# gbm0, use_feat0 = load_model(opt)
opt['model_name'] = 'lgb_1_2017-09-15#19:50:48_0.58820.pkl'
gbm, use_feat = load_model(opt)
opt['model_name'] = 'lgb_2017-09-23#20:14:52_0.58893.pkl'
gbm1, use_feat1 = load_model(opt)
# gbm2, use_feat2 = load_model(opt)
# opt['model_name'] = 'lgb_2017-09-03#23:24:26_0.57836.pkl'
# gbm3, use_feat3 = load_model(opt)
# opt['model_name'] = ''
# gbm4, use_feat4 = load_model(opt)
# ---------------------- 评估 -------------------------
train_feat.loc[:, 'pred'] = gbm.predict(train_feat[use_feat])
gc.collect()
train_feat[['orderid', 'geohashed_end_loc', 'pred']].to_csv(
'/home/xuwenchao/dyj-storage/pred/pred_23_24_0.58820.csv', index=None)
train_feat.loc[:, 'pred'] = gbm1.predict(train_feat[use_feat1])
gc.collect()
train_feat[['orderid', 'geohashed_end_loc', 'pred']].to_csv(
'/home/xuwenchao/dyj-storage/pred/pred_23_24_0.58893.csv', index=None)
def val(**kwargs):
# ---------------------- 更新参数 ----------------------
opt = DefaultConfig()
opt.update(**kwargs)
opt.printf()
# ---------------------- 数据处理 ----------------------
# 获取数据
# train1, train2, train_test = get_train_data(opt)
# 获取样本
# train_sample = get_sample(train1, train2, load=True)
# 获取特征
# train_feat = get_feat(train_test, train_sample)
# gc.collect()
# train_feat.to_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_{}.hdf'.format(opt['startday']), 'w', complib='blosc', complevel=5)
train_feat = pd.read_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_23_24_label.hdf')
# ---------------------- 载入模型 ----------------------
# opt['model_name'] = 'lgb_1_90_all.pkl'
# gbm0, use_feat0 = load_model(opt)
opt['model_name'] = 'lgb_1_2017-09-15#19:50:48_0.58820.pkl'
gbm, use_feat = load_model(opt)
opt['model_name'] = 'lgb_2017-09-23#20:14:52_0.58893.pkl'
gbm1, use_feat1 = load_model(opt)
# gbm2, use_feat2 = load_model(opt)
# opt['model_name'] = 'lgb_2017-09-03#23:24:26_0.57836.pkl'
# gbm3, use_feat3 = load_model(opt)
# opt['model_name'] = ''
# gbm4, use_feat4 = load_model(opt)
# ---------------------- 评估 -------------------------
train_feat.loc[:, 'pred'] = gbm.predict(train_feat[use_feat])
gc.collect()
train_feat[['orderid', 'geohashed_end_loc', 'pred']].to_csv('/home/xuwenchao/dyj-storage/pred/pred_23_24_0.58820.csv', index=None)
train_feat.loc[:, 'pred'] = gbm1.predict(train_feat[use_feat1])
gc.collect()
train_feat[['orderid', 'geohashed_end_loc', 'pred']].to_csv('/home/xuwenchao/dyj-storage/pred/pred_23_24_0.58893.csv', index=None)
def train(**kwargs):
# ---------------------- 更新参数 ----------------------
opt = DefaultConfig()
opt.update(**kwargs)
opt.printf()
# ---------------------- 数据处理 ----------------------
# 获取数据
train1, train2 = get_train_data(opt)
# 获取样本
# train_sample = get_sample(train1, train2, load=True)
# 获取特征
# train_feat = get_feat(train1, train_sample)
# 获取标签
# train_all = get_label(train_feat, opt)
# gc.collect()
# train_all.to_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_23_24_label.hdf', 'w', complib='blosc', complevel=5)
train_all = pd.read_hdf(
'/home/xuwenchao/dyj-storage/all-feat/feat_23_24_label.hdf')
print(train_all.shape)
# 取出需要用的特征
# opt['model_name'] = 'lgb_1_2017-09-15#19:50:48_0.58820.pkl'
# gbm, use_feat = load_model(opt)
# predictors_100 = pd.DataFrame(data={'feature_name': gbm.feature_name(), 'feature_importance': gbm.feature_importance()})
# predictors_100 = predictors_100.sort_values(by=['feature_importance'], ascending=False)['feature_name'].values[:100]
# use_feat = list(predictors_100) + ['orderid', 'geohashed_end_loc', 'label'] + ['sloc_eloc_common_eloc_count', 'sloc_eloc_common_sloc_count', 'sloc_eloc_common_conn1_count', 'sloc_eloc_common_conn2_count', 'sloc_eloc_common_eloc_rate', 'sloc_eloc_common_sloc_rate', 'sloc_eloc_common_conn1_rate', 'sloc_eloc_common_conn2_rate', 'user_sloc_eloc_common_eloc_count', 'user_sloc_eloc_common_sloc_count', 'user_sloc_eloc_common_conn1_count', 'user_sloc_eloc_common_conn2_count', 'user_sloc_eloc_common_eloc_rate', 'user_sloc_eloc_common_sloc_rate', 'user_sloc_eloc_common_conn1_rate', 'user_sloc_eloc_common_conn2_rate']
# train_all = train_all[use_feat]
# gc.collect()
# -------------------- 训练第一层 ------------------------
# ********* 准备数据 **********
# 划分验证集
train, val = train_test_split(train_all, test_size=0.1)
# 定义使用哪些特征
# opt['model_name'] = 'lgb_1_2017-09-15#19:50:48_0.58820.pkl'
# gbm, use_feat = load_model(opt)
filters = set([
'orderid', 'userid', 'biketype', 'geohashed_start_loc', 'bikeid',
'starttime', 'geohashed_end_loc', 'label'
])
predictors = list(
filter(lambda x: x not in filters, train_all.columns.tolist()))
# predictors = pd.DataFrame(data={'feature_name': gbm.feature_name(), 'feature_importance': gbm.feature_importance()})
# predictors = predictors.sort_values(by=['feature_importance'], ascending=False)['feature_name'].values[:100]
# use_feat = list(predictors) + ['orderid', 'geohashed_end_loc'] + ['sloc_eloc_common_eloc_count', 'sloc_eloc_common_sloc_count', 'sloc_eloc_common_conn1_count', 'sloc_eloc_common_conn2_count', 'sloc_eloc_common_eloc_rate', 'sloc_eloc_common_sloc_rate', 'sloc_eloc_common_conn1_rate', 'sloc_eloc_common_conn2_rate', 'user_sloc_eloc_common_eloc_count', 'user_sloc_eloc_common_sloc_count', 'user_sloc_eloc_common_conn1_count', 'user_sloc_eloc_common_conn2_count', 'user_sloc_eloc_common_eloc_rate', 'user_sloc_eloc_common_sloc_rate', 'user_sloc_eloc_common_conn1_rate', 'user_sloc_eloc_common_conn2_rate']
# predictors = list(predictors_100) + ['sloc_eloc_common_eloc_count', 'sloc_eloc_common_sloc_count', 'sloc_eloc_common_conn1_count', 'sloc_eloc_common_conn2_count', 'sloc_eloc_common_eloc_rate', 'sloc_eloc_common_sloc_rate', 'sloc_eloc_common_conn1_rate', 'sloc_eloc_common_conn2_rate', 'user_sloc_eloc_common_eloc_count', 'user_sloc_eloc_common_sloc_count', 'user_sloc_eloc_common_conn1_count', 'user_sloc_eloc_common_conn2_count', 'user_sloc_eloc_common_eloc_rate', 'user_sloc_eloc_common_sloc_rate', 'user_sloc_eloc_common_conn1_rate', 'user_sloc_eloc_common_conn2_rate']
print('使用的特征：{}维\n'.format(len(predictors)), predictors)
# 定义数据集
X_train = train[predictors]
y_train = train['label']
X_val = val[predictors]
y_val = val['label']
del train, val
gc.collect()
# ********* LightGBM *********
# 数据集
lgb_train = lgb.Dataset(X_train, y_train)
lgb_val = lgb.Dataset(X_val, y_val, reference=lgb_train)
# 配置
params = {
'objective': 'binary',
'metric': {'auc', 'binary_logloss'},
'is_unbalance': True,
'num_leaves': opt['lgb_leaves'],
'learning_rate': opt['lgb_lr'],
'feature_fraction': 0.886,
'bagging_fraction': 0.886,
'bagging_freq': 5
}
gc.collect()
# ********** 开始训练 *********
gbm1 = lgb.train(params,
lgb_train,
num_boost_round=1200,
valid_sets=[lgb_train, lgb_val],
early_stopping_rounds=5)
gc.collect()
# # ********* 保存模型 *********
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
# save_path = '{}/{}_{}_{:.5f}.pkl'.format(opt['model_dir'], 'lgb', cur_time, score[0])
save_path = '{}/{}_{}.pkl'.format(opt['model_dir'], 'lgb', cur_time)
with open(save_path, 'wb') as fout:
pickle.dump(gbm1, fout)
print('保存模型：', save_path)
gc.collect()
# # ********* 评估 *********
# # 在训练集上看效果
del X_train, y_train, X_val, y_val
gc.collect()
score = get_score(train_all, predictors, gbm1, opt)
print('训练集分数：{}'.format(score))
import sys
sys.exit(0)
# save_path = '{}/{}.pkl'.format(opt['model_dir'], 'lgb_1_300_top25')
# with open(save_path, 'wb') as fout:
# pickle.dump(gbm1, fout)
# print('保存模型(第一层)：', save_path)
# ********* save predict *****
# train_all[['orderid', 'geohashed_end_loc', 'pred']].to_hdf('/home/xuwenchao/dyj-storage/train2324_80_pred_res.hdf', 'w', complib='blosc', complevel=5)
# print('Save train_pred_res.hdf successful!!!')
# import sys
# sys.exit(0)
# -------------------- 训练第二层 ------------------------
# opt['model_name'] = 'lgb_1_300_top25.pkl'
# gbm1, use_feat1 = load_model(opt)
# train_all.loc[:, 'pred'] = gbm1.predict(train_all[use_feat1])
# 去掉重要性较低的特征，筛选出排名前十的候选样本，重新训练模型（后期可以载入模型finetune，
# 尤其是对于样本量较少的情况，甚至可以选前5，
# 但15可以覆盖99.5%的原始label，10可以覆盖98%的原始label，
# 这两者可能会好一些，备选方案：5(+finetune)，10(+finetune)，15(+finetune)）
predictors = pd.DataFrame(
data={
'feature_name': gbm1.feature_name(),
'feature_importance': gbm1.feature_importance()
})
predictors = predictors[
predictors['feature_importance'] > 0]['feature_name'].values
print('第二层使用的特征：{}维\n'.format(len(predictors)), predictors)
train_all = train_all.sort_values(
by=['orderid', 'pred'], ascending=False).groupby('orderid').head(15)
# train_all = rank(train_all, 'orderid', 'pred', ascending=False)
del train_all['pred']
print('第二层数据：', train_all.shape)
# ********* 准备数据 **********
# 划分验证集
train, val = train_test_split(train_all, test_size=0.1)
# 定义数据集
X_train = train[predictors]
y_train = train['label']
X_val = val[predictors]
y_val = val['label']
del train, val
gc.collect()
# 数据集
lgb_train = lgb.Dataset(X_train, y_train)
lgb_val = lgb.Dataset(X_val, y_val, reference=lgb_train)
# ********** 开始训练 *********
gbm2 = lgb.train(params,
lgb_train,
num_boost_round=1200,
valid_sets=[lgb_train, lgb_val],
early_stopping_rounds=5
# init_model=gbm1 # finetune
)
# ********* 评估 *********
# 在训练集上看效果
score = get_score(train_all, predictors, gbm2, opt)
print('训练集分数(第二层)：{}'.format(score))
# ********* 保存模型 *********
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
save_path = '{}/{}_{}_{:.5f}.pkl'.format(opt['model_dir'], 'lgb_2',
cur_time, score[0])
with open(save_path, 'wb') as fout:
pickle.dump(gbm2, fout)
print('保存模型(第二层)：', save_path)
# save_path = '{}/{}.pkl'.format(opt['model_dir'], 'lgb_2_300_top15')
# with open(save_path, 'wb') as fout:
# pickle.dump(gbm2, fout)
# print('保存模型(第二层)：', save_path)
import sys
sys.exit(0)
# -------------------- 训练第三层 ------------------------
# 筛选出排名前五的候选样本
predictors = pd.DataFrame(
data={
'feature_name': gbm2.feature_name(),
'feature_importance': gbm2.feature_importance()
})
predictors = predictors[
predictors['feature_importance'] > 0]['feature_name'].values
print('第三层使用的特征：{}维\n'.format(len(predictors)), predictors)
train_all = train_all.sort_values(
by=['orderid', 'pred'], ascending=False).groupby('orderid').head(10)
# train_all = rank(train_all, 'orderid', 'pred', ascending=False)
del train_all['pred']
print('第三层数据：', train_all.shape)
# ********* 准备数据 **********
# 划分验证集
train, val = train_test_split(train_all, test_size=0.1)
# 定义数据集
X_train = train[predictors]
y_train = train['label']
X_val = val[predictors]
y_val = val['label']
del train, val
gc.collect()
# 数据集
lgb_train = lgb.Dataset(X_train, y_train)
lgb_val = lgb.Dataset(X_val, y_val, reference=lgb_train)
# ********** 开始训练 *********
gbm3 = lgb.train(params,
lgb_train,
num_boost_round=1200,
valid_sets=[lgb_train, lgb_val],
early_stopping_rounds=5
# init_model=gbm2 # finetune
)
# ********* 评估 *********
# 在训练集上看效果
score = get_score(train_all, predictors, gbm3, opt)
print('训练集分数(第三层)：{}'.format(score))
# ********* 保存模型 *********
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
save_path = '{}/{}_{}_{:.5f}.pkl'.format(opt['model_dir'], 'lgb_3',
cur_time, score[0])
with open(save_path, 'wb') as fout:
pickle.dump(gbm3, fout)
print('保存模型(第三层)：', save_path)
save_path = '{}/{}.pkl'.format(opt['model_dir'], 'lgb_3_300_top10')
with open(save_path, 'wb') as fout:
pickle.dump(gbm3, fout)
print('保存模型(第三层)：', save_path)
# -------------------- 训练第四层 ------------------------
# 筛选出排名前三的候选样本
predictors = pd.DataFrame(
data={
'feature_name': gbm3.feature_name(),
'feature_importance': gbm3.feature_importance()
})
predictors = predictors[
predictors['feature_importance'] > 0]['feature_name'].values
print('第四层使用的特征：{}维\n'.format(len(predictors)), predictors)
train_all = train_all.sort_values(
by=['orderid', 'pred'], ascending=False).groupby('orderid').head(5)
# train_all = rank(train_all, 'orderid', 'pred', ascending=False)
del train_all['pred']
print('第四层数据：', train_all.shape)
# ********* 准备数据 **********
# 划分验证集
train, val = train_test_split(train_all, test_size=0.1)
# 定义数据集
X_train = train[predictors]
y_train = train['label']
X_val = val[predictors]
y_val = val['label']
del train, val
gc.collect()
# 数据集
lgb_train = lgb.Dataset(X_train, y_train)
lgb_val = lgb.Dataset(X_val, y_val, reference=lgb_train)
# ********** 开始训练 *********
gbm4 = lgb.train(params,
lgb_train,
num_boost_round=1200,
valid_sets=[lgb_train, lgb_val],
early_stopping_rounds=5
# init_model=gbm3 # finetune
)
# ********* 评估 *********
# 在训练集上看效果
score = get_score(train_all, predictors, gbm4, opt)
print('训练集分数(第四层)：{}'.format(score))
# ********* 保存模型 *********
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
save_path = '{}/{}_{}_{:.5f}.pkl'.format(opt['model_dir'], 'lgb_4',
cur_time, score[0])
with open(save_path, 'wb') as fout:
pickle.dump(gbm4, fout)
print('保存模型(第四层)：', save_path)
save_path = '{}/{}.pkl'.format(opt['model_dir'], 'lgb_4_300_top5')
with open(save_path, 'wb') as fout:
pickle.dump(gbm4, fout)
print('保存模型(第四层)：', save_path)
def train(**kwargs):
# ---------------------- 更新参数 ----------------------
opt = DefaultConfig()
opt.update(**kwargs)
opt.printf()
# ---------------------- 数据处理 ----------------------
# 获取数据
# train1, train2 = get_train_data(opt)
# 获取样本
# train_sample = get_sample(train1, train2, load=True)
# 获取特征
# train_feat = get_feat(train1, train_sample)
# 获取标签
# train_all = get_label(train_feat, opt)
# gc.collect()
# train_all.to_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_23_24_label.hdf', 'w', complib='blosc', complevel=5)
train_all = pd.read_hdf('/home/xuwenchao/dyj-storage/all-feat/feat_23_24_label.hdf')
print(train_all.shape)
# 取出需要用的特征
# opt['model_name'] = 'lgb_1_2017-09-15#19:50:48_0.58820.pkl'
# gbm, use_feat = load_model(opt)
# predictors_100 = pd.DataFrame(data={'feature_name': gbm.feature_name(), 'feature_importance': gbm.feature_importance()})
# predictors_100 = predictors_100.sort_values(by=['feature_importance'], ascending=False)['feature_name'].values[:100]
# use_feat = list(predictors_100) + ['orderid', 'geohashed_end_loc', 'label'] + ['sloc_eloc_common_eloc_count', 'sloc_eloc_common_sloc_count', 'sloc_eloc_common_conn1_count', 'sloc_eloc_common_conn2_count', 'sloc_eloc_common_eloc_rate', 'sloc_eloc_common_sloc_rate', 'sloc_eloc_common_conn1_rate', 'sloc_eloc_common_conn2_rate', 'user_sloc_eloc_common_eloc_count', 'user_sloc_eloc_common_sloc_count', 'user_sloc_eloc_common_conn1_count', 'user_sloc_eloc_common_conn2_count', 'user_sloc_eloc_common_eloc_rate', 'user_sloc_eloc_common_sloc_rate', 'user_sloc_eloc_common_conn1_rate', 'user_sloc_eloc_common_conn2_rate']
# train_all = train_all[use_feat]
# gc.collect()
# -------------------- 训练第一层 ------------------------
# ********* 准备数据 **********
# 划分验证集
train, val = train_test_split(train_all, test_size=0.1)
# 定义使用哪些特征
# opt['model_name'] = 'lgb_1_2017-09-15#19:50:48_0.58820.pkl'
# gbm, use_feat = load_model(opt)
filters = set(['orderid', 'userid', 'biketype', 'geohashed_start_loc', 'bikeid', 'starttime', 'geohashed_end_loc', 'label'])
predictors = list(filter(lambda x: x not in filters, train_all.columns.tolist()))
# predictors = pd.DataFrame(data={'feature_name': gbm.feature_name(), 'feature_importance': gbm.feature_importance()})
# predictors = predictors.sort_values(by=['feature_importance'], ascending=False)['feature_name'].values[:100]
# use_feat = list(predictors) + ['orderid', 'geohashed_end_loc'] + ['sloc_eloc_common_eloc_count', 'sloc_eloc_common_sloc_count', 'sloc_eloc_common_conn1_count', 'sloc_eloc_common_conn2_count', 'sloc_eloc_common_eloc_rate', 'sloc_eloc_common_sloc_rate', 'sloc_eloc_common_conn1_rate', 'sloc_eloc_common_conn2_rate', 'user_sloc_eloc_common_eloc_count', 'user_sloc_eloc_common_sloc_count', 'user_sloc_eloc_common_conn1_count', 'user_sloc_eloc_common_conn2_count', 'user_sloc_eloc_common_eloc_rate', 'user_sloc_eloc_common_sloc_rate', 'user_sloc_eloc_common_conn1_rate', 'user_sloc_eloc_common_conn2_rate']
# predictors = list(predictors_100) + ['sloc_eloc_common_eloc_count', 'sloc_eloc_common_sloc_count', 'sloc_eloc_common_conn1_count', 'sloc_eloc_common_conn2_count', 'sloc_eloc_common_eloc_rate', 'sloc_eloc_common_sloc_rate', 'sloc_eloc_common_conn1_rate', 'sloc_eloc_common_conn2_rate', 'user_sloc_eloc_common_eloc_count', 'user_sloc_eloc_common_sloc_count', 'user_sloc_eloc_common_conn1_count', 'user_sloc_eloc_common_conn2_count', 'user_sloc_eloc_common_eloc_rate', 'user_sloc_eloc_common_sloc_rate', 'user_sloc_eloc_common_conn1_rate', 'user_sloc_eloc_common_conn2_rate']
print('使用的特征：{}维\n'.format(len(predictors)), predictors)
# 定义数据集
X_train = train[predictors]
y_train = train['label']
X_val = val[predictors]
y_val = val['label']
del train, val
gc.collect()
# ********* LightGBM *********
# 数据集
lgb_train = lgb.Dataset(X_train, y_train)
lgb_val = lgb.Dataset(X_val, y_val, reference=lgb_train)
# 配置
params = {
'objective': 'binary',
'metric': {'auc', 'binary_logloss'},
'is_unbalance': True,
'num_leaves': opt['lgb_leaves'],
'learning_rate': opt['lgb_lr'],
'feature_fraction': 0.886,
'bagging_fraction': 0.886,
'bagging_freq': 5
}
gc.collect()
# ********** 开始训练 *********
gbm1 = lgb.train(
params,
lgb_train,
num_boost_round=1200,
valid_sets=[lgb_train, lgb_val],
early_stopping_rounds=5
)
gc.collect()
# # ********* 保存模型 *********
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
# save_path = '{}/{}_{}_{:.5f}.pkl'.format(opt['model_dir'], 'lgb', cur_time, score[0])
save_path = '{}/{}_{}.pkl'.format(opt['model_dir'], 'lgb', cur_time)
with open(save_path, 'wb') as fout:
pickle.dump(gbm1, fout)
print('保存模型：', save_path)
gc.collect()
# # ********* 评估 *********
# # 在训练集上看效果
del X_train, y_train, X_val, y_val
gc.collect()
score = get_score(train_all, predictors, gbm1, opt)
print('训练集分数：{}'.format(score))
import sys
sys.exit(0)
# save_path = '{}/{}.pkl'.format(opt['model_dir'], 'lgb_1_300_top25')
# with open(save_path, 'wb') as fout:
# pickle.dump(gbm1, fout)
# print('保存模型(第一层)：', save_path)
# ********* save predict *****
# train_all[['orderid', 'geohashed_end_loc', 'pred']].to_hdf('/home/xuwenchao/dyj-storage/train2324_80_pred_res.hdf', 'w', complib='blosc', complevel=5)
# print('Save train_pred_res.hdf successful!!!')
# import sys
# sys.exit(0)
# -------------------- 训练第二层 ------------------------
# opt['model_name'] = 'lgb_1_300_top25.pkl'
# gbm1, use_feat1 = load_model(opt)
# train_all.loc[:, 'pred'] = gbm1.predict(train_all[use_feat1])
# 去掉重要性较低的特征，筛选出排名前十的候选样本，重新训练模型（后期可以载入模型finetune，尤其是对于样本量较少的情况，甚至可以选前5，但15可以覆盖99.5%的原始label，10可以覆盖98%的原始label，这两者可能会好一些，备选方案：5(+finetune)，10(+finetune)，15(+finetune)）
predictors = pd.DataFrame(data={'feature_name': gbm1.feature_name(), 'feature_importance': gbm1.feature_importance()})
predictors = predictors[predictors['feature_importance']>0]['feature_name'].values
print('第二层使用的特征：{}维\n'.format(len(predictors)), predictors)
train_all = train_all.sort_values(by=['orderid', 'pred'], ascending=False).groupby('orderid').head(15)
# train_all = rank(train_all, 'orderid', 'pred', ascending=False)
del train_all['pred']
print('第二层数据：', train_all.shape)
# ********* 准备数据 **********
# 划分验证集
train, val = train_test_split(train_all, test_size=0.1)
# 定义数据集
X_train = train[predictors]
y_train = train['label']
X_val = val[predictors]
y_val = val['label']
del train, val
gc.collect()
# 数据集
lgb_train = lgb.Dataset(X_train, y_train)
lgb_val = lgb.Dataset(X_val, y_val, reference=lgb_train)
# ********** 开始训练 *********
gbm2 = lgb.train(
params,
lgb_train,
num_boost_round=1200,
valid_sets=[lgb_train, lgb_val],
early_stopping_rounds=5
# init_model=gbm1 # finetune
)
# ********* 评估 *********
# 在训练集上看效果
score = get_score(train_all, predictors, gbm2, opt)
print('训练集分数(第二层)：{}'.format(score))
# ********* 保存模型 *********
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
save_path = '{}/{}_{}_{:.5f}.pkl'.format(opt['model_dir'], 'lgb_2', cur_time, score[0])
with open(save_path, 'wb') as fout:
pickle.dump(gbm2, fout)
print('保存模型(第二层)：', save_path)
# save_path = '{}/{}.pkl'.format(opt['model_dir'], 'lgb_2_300_top15')
# with open(save_path, 'wb') as fout:
# pickle.dump(gbm2, fout)
# print('保存模型(第二层)：', save_path)
import sys
sys.exit(0)
# -------------------- 训练第三层 ------------------------
# 筛选出排名前五的候选样本
predictors = pd.DataFrame(data={'feature_name': gbm2.feature_name(), 'feature_importance': gbm2.feature_importance()})
predictors = predictors[predictors['feature_importance']>0]['feature_name'].values
print('第三层使用的特征：{}维\n'.format(len(predictors)), predictors)
train_all = train_all.sort_values(by=['orderid', 'pred'], ascending=False).groupby('orderid').head(10)
# train_all = rank(train_all, 'orderid', 'pred', ascending=False)
del train_all['pred']
print('第三层数据：', train_all.shape)
# ********* 准备数据 **********
# 划分验证集
train, val = train_test_split(train_all, test_size=0.1)
# 定义数据集
X_train = train[predictors]
y_train = train['label']
X_val = val[predictors]
y_val = val['label']
del train, val
gc.collect()
# 数据集
lgb_train = lgb.Dataset(X_train, y_train)
lgb_val = lgb.Dataset(X_val, y_val, reference=lgb_train)
# ********** 开始训练 *********
gbm3 = lgb.train(
params,
lgb_train,
num_boost_round=1200,
valid_sets=[lgb_train, lgb_val],
early_stopping_rounds=5
# init_model=gbm2 # finetune
)
# ********* 评估 *********
# 在训练集上看效果
score = get_score(train_all, predictors, gbm3, opt)
print('训练集分数(第三层)：{}'.format(score))
# ********* 保存模型 *********
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
save_path = '{}/{}_{}_{:.5f}.pkl'.format(opt['model_dir'], 'lgb_3', cur_time, score[0])
with open(save_path, 'wb') as fout:
pickle.dump(gbm3, fout)
print('保存模型(第三层)：', save_path)
save_path = '{}/{}.pkl'.format(opt['model_dir'], 'lgb_3_300_top10')
with open(save_path, 'wb') as fout:
pickle.dump(gbm3, fout)
print('保存模型(第三层)：', save_path)
# -------------------- 训练第四层 ------------------------
# 筛选出排名前三的候选样本
predictors = pd.DataFrame(data={'feature_name': gbm3.feature_name(), 'feature_importance': gbm3.feature_importance()})
predictors = predictors[predictors['feature_importance']>0]['feature_name'].values
print('第四层使用的特征：{}维\n'.format(len(predictors)), predictors)
train_all = train_all.sort_values(by=['orderid', 'pred'], ascending=False).groupby('orderid').head(5)
# train_all = rank(train_all, 'orderid', 'pred', ascending=False)
del train_all['pred']
print('第四层数据：', train_all.shape)
# ********* 准备数据 **********
# 划分验证集
train, val = train_test_split(train_all, test_size=0.1)
# 定义数据集
X_train = train[predictors]
y_train = train['label']
X_val = val[predictors]
y_val = val['label']
del train, val
gc.collect()
# 数据集
lgb_train = lgb.Dataset(X_train, y_train)
lgb_val = lgb.Dataset(X_val, y_val, reference=lgb_train)
# ********** 开始训练 *********
gbm4 = lgb.train(
params,
lgb_train,
num_boost_round=1200,
valid_sets=[lgb_train, lgb_val],
early_stopping_rounds=5
# init_model=gbm3 # finetune
)
# ********* 评估 *********
# 在训练集上看效果
score = get_score(train_all, predictors, gbm4, opt)
print('训练集分数(第四层)：{}'.format(score))
# ********* 保存模型 *********
cur_time = datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')
save_path = '{}/{}_{}_{:.5f}.pkl'.format(opt['model_dir'], 'lgb_4', cur_time, score[0])
with open(save_path, 'wb') as fout:
pickle.dump(gbm4, fout)
print('保存模型(第四层)：', save_path)
save_path = '{}/{}.pkl'.format(opt['model_dir'], 'lgb_4_300_top5')
with open(save_path, 'wb') as fout:
pickle.dump(gbm4, fout)
print('保存模型(第四层)：', save_path)
def test_stack(**kwargs):
opt = DefaultConfig()
opt.update(**kwargs)
logger = Logger()
result_dir = '/home/dyj/'
resmat = [result_dir+'TextCNN1_2017-07-27#12:30:16_test_res.pt',\
result_dir+'TextCNN2_2017-07-27#12:22:42_test_res.pt', \
result_dir+'RNN1_2017-07-27#12:35:51_test_res.pt',\
result_dir+'RNN2_2017-07-27#11:33:24_test_res.pt',\
result_dir+'RCNN1_2017-07-27#11:30:42_test_res.pt',\
result_dir+'RCNNcha_2017-07-27#16:00:33_test_res.pt',\
result_dir+'FastText4_2017-07-28#17:20:21_test_res.pt',\
result_dir+'FastText1_2017-07-29#10:47:46_test_res.pt']
opt['stack_num'] = len(resmat)
test_dataset = Stack_Dataset(resmat=resmat, test=True)
test_loader = data.DataLoader(test_dataset,
shuffle=False,
batch_size=opt['batch_size'])
test_idx = np.load(opt['test_idx'])
topic_idx = np.load(opt['topic_idx'])
logger.info('Using model {}'.format(opt['model']))
Model = getattr(models, opt['model'])
model = Model(opt)
print model
if opt['load']:
if opt.get('load_name', None) is None:
model = load_model(model,
model_dir=opt['model_dir'],
model_name=opt['model'])
else:
model = load_model(model, model_dir=opt['model_dir'], model_name=opt['model'], \
name=opt['load_name'])
if opt['device'] != None:
torch.cuda.set_device(opt['device'])
if opt['cuda']:
model.cuda()
logger.info('Start testing...')
model.eval()
predict_label_list = []
res = torch.Tensor(opt['test_num'], opt['class_num'])
for i, batch in enumerate(test_loader, 0):
batch_size = batch[0].size(0)
resmat = batch
resmat = [Variable(ii) for ii in resmat]
if opt['cuda']:
resmat = [ii.cuda() for ii in resmat]
logit = model(resmat)
if opt.get('save_resmat', False):
res[i * opt['batch_size']:i * opt['batch_size'] +
batch_size] = logit.data.cpu()
predict_label_list += [list(ii) for ii in logit.topk(5, 1)[1].data]
if opt.get('save_resmat', False):
torch.save(
res, '{}/{}_{}_test_res.pt'.format(
opt['result_dir'], opt['model'],
datetime.datetime.now().strftime('%Y-%m-%d#%H:%M:%S')))
return
lines = []
for qid, top5 in zip(test_idx, predict_label_list):
topic_ids = [topic_idx[i] for i in top5]
lines.append('{},{}'.format(qid, ','.join(topic_ids)))
if opt.get('load_name', None) is None:
write_result(lines,
model_dir=opt['model_dir'],
model_name=opt['model'],
result_dir=opt['result_dir'])
else:
write_result(lines, model_dir=opt['model_dir'], model_name=opt['model'], \
name=opt['load_name'], result_dir=opt['result_dir'])
def train_stack(**kwargs):
opt = DefaultConfig()
opt.update(**kwargs)
vis = Visualizer(opt['model'])
logger = Logger()
result_dir = '/home/dyj/'
resmat = [(result_dir + 'RNN10_cal_res.pt', 10),\
(result_dir + 'TextCNN10_char.pt', 10),\
(result_dir + 'TextCNN10_top1.pt', 10),\
(result_dir + 'TextCNN10_top1_char.pt', 10),\
(result_dir + 'FastText10_res.pt', 10),\
('/mnt/result/results/TextCNN5_12h.pt', 5),\
('/mnt/result/results/RNN1_char.pt', 1)
]
label = result_dir + 'label.pt'
opt['stack_num'] = len(resmat)
train_dataset = Stack_Dataset(resmat=resmat, label=label)
train_loader = data.DataLoader(train_dataset,
shuffle=True,
batch_size=opt['batch_size'])
logger.info('Using model {}'.format(opt['model']))
Model = getattr(models, opt['model'])
model = Model(opt)
print model
if opt['use_self_loss']:
Loss = getattr(models, opt['loss_function'])
else:
Loss = getattr(nn, opt['loss_function'])
if opt['load']:
if opt.get('load_name', None) is None:
model = load_model(model,
model_dir=opt['model_dir'],
model_name=opt['model'])
else:
model = load_model(model, model_dir=opt['model_dir'], model_name=opt['model'], \
name=opt['load_name'])
if opt['device'] != None:
torch.cuda.set_device(opt['device'])
if opt['cuda']:
model.cuda()
loss_function = Loss()
optimizer = torch.optim.Adam(model.parameters(), lr=opt['lr'])
logger.info('Start running...')
steps = 0
model.train()
for epoch in range(opt['base_epoch'] + 1, opt['epochs'] + 1):
for i, batch in enumerate(train_loader, 1):
resmat, label = batch[0:-1], batch[-1]
resmat, label = [Variable(ii) for ii in resmat], Variable(label)
if opt['cuda']:
resmat, label = [ii.cuda() for ii in resmat], label.cuda()
optimizer.zero_grad()
logit = model(resmat)
loss = loss_function(logit, label)
loss.backward()
optimizer.step()
steps += 1
if steps % opt['log_interval'] == 0:
corrects = ((logit.data >
opt['threshold']) == (label.data).byte()).sum()
accuracy = 100.0 * corrects / (opt['batch_size'] *
opt['class_num'])
log_info = 'Steps[{:>8}] (epoch[{:>2}] / batch[{:>5}]) - loss: {:.6f}, acc: {:.4f} % ({} / {})'.format( \
steps, epoch, i, loss.data[0], accuracy, \
corrects, opt['batch_size'] * opt['class_num'])
logger.info(log_info)
vis.plot('loss', loss.data[0])
precision, recall, score = get_score(logit.data.cpu(),
label.data.cpu())
logger.info('Precision {}, Recall {}, Score {}'.format(
precision, recall, score))
vis.plot('score', score)
logger.info('Training epoch {} finished!'.format(epoch))
#save_model(model, model_dir=opt['model_dir'], model_name=opt['model'], epoch=epoch)
if epoch == 3:
for param_group in optimizer.param_groups:
param_group['lr'] = opt['lr'] * opt['lr_decay']
save_model(model,
model_dir=opt['model_dir'],
model_name=opt['model'],
epoch=epoch)
def test(**kwargs):
opt = DefaultConfig()
opt.update(**kwargs)
logger = Logger()
prefix = ''
if opt['use_double_length']: prefix += '_2'
print prefix
if opt['use_char']:
logger.info('Load char data starting...')
opt['embed_num'] = opt['char_embed_num']
embed_mat = np.load(opt['char_embed'])
test_title = np.load(opt['test_title_char' + prefix])
test_desc = np.load(opt['test_desc_char' + prefix])
logger.info('Load char data finished!')
elif opt['use_word']:
logger.info('Load word data starting...')
opt['embed_num'] = opt['word_embed_num']
embed_mat = np.load(opt['word_embed'])
test_title = np.load(opt['test_title_word' + prefix])
test_desc = np.load(opt['test_desc_word' + prefix])
logger.info('Load word data finished!')
elif opt['use_char_word']:
logger.info('Load char-word data starting...')
embed_mat_char = np.load(opt['char_embed'])
embed_mat_word = np.load(opt['word_embed'])
embed_mat = np.vstack((embed_mat_char, embed_mat_word))
test_title = np.load(opt['test_title_char' + prefix])
test_desc = np.load(opt['test_desc_word' + prefix])
logger.info('Load char-word data finished!')
elif opt['use_word_char']:
logger.info('Load word-char data starting...')
embed_mat_char = np.load(opt['char_embed'])
embed_mat_word = np.load(opt['word_embed'])
embed_mat = np.vstack((embed_mat_char, embed_mat_word))
test_title = np.load(opt['test_title_word' + prefix])
test_desc = np.load(opt['test_desc_char' + prefix])
logger.info('Load word-char data finished!')
test_idx = np.load(opt['test_idx'])
topic_idx = np.load(opt['topic_idx'])
test_dataset = Dataset(test=True, title=test_title, desc=test_desc)
test_loader = data.DataLoader(test_dataset,
shuffle=False,
batch_size=opt['batch_size'])
logger.info('Using model {}'.format(opt['model']))
Model = getattr(models, opt['model'])
model = Model(embed_mat, opt)
if opt['load']:
if opt.get('load_name', None) is None:
model = load_model(model,
model_dir=opt['model_dir'],
model_name=opt['model'])
else:
model = load_model(model, model_dir=opt['model_dir'], model_name=opt['model'], \
name=opt['load_name'])
if opt['device'] != None:
torch.cuda.set_device(opt['device'])
if opt['cuda']:
model.cuda()
logger.info('Start testing...')
model.eval()
predict_label_list = []
res = torch.Tensor(opt['test_num'], opt['class_num'])
for i, batch in enumerate(test_loader, 0):
batch_size = batch[0].size(0)
title, desc = batch
title, desc = Variable(title), Variable(desc)
if opt['cuda']:
title, desc = title.cuda(), desc.cuda()
logit = model(title, desc)
if opt.get('save_resmat', False):
res[i * opt['batch_size']:i * opt['batch_size'] +
batch_size] = logit.data.cpu()
predict_label_list += [list(ii) for ii in logit.topk(5, 1)[1].data]
if opt.get('save_resmat', False):
torch.save(res, '{}/{}_test_res.pt'.format(opt['result_dir'],
opt['model']))
return
lines = []
for qid, top5 in zip(test_idx, predict_label_list):
topic_ids = [topic_idx[i] for i in top5]
lines.append('{},{}'.format(qid, ','.join(topic_ids)))
if opt.get('load_name', None) is None:
write_result(lines,
model_dir=opt['model_dir'],
model_name=opt['model'],
result_dir=opt['result_dir'])
else:
write_result(lines, model_dir=opt['model_dir'], model_name=opt['model'], \
name=opt['load_name'], result_dir=opt['result_dir'])
def train(**kwargs):
opt = DefaultConfig()
opt.update(**kwargs)
vis = Visualizer(opt['model'])
logger = Logger()
prefix = ''
if opt['use_double_length']: prefix += '_2'
print prefix
if opt['use_char']:
logger.info('Load char data starting...')
opt['embed_num'] = opt['char_embed_num']
embed_mat = np.load(opt['char_embed'])
train_title = np.load(opt['train_title_char' + prefix])
train_desc = np.load(opt['train_desc_char' + prefix])
train_label = np.load(opt['train_label'])
val_title = np.load(opt['val_title_char' + prefix])
val_desc = np.load(opt['val_desc_char' + prefix])
val_label = np.load(opt['val_label'])
logger.info('Load char data finished!')
elif opt['use_word']:
logger.info('Load word data starting...')
opt['embed_num'] = opt['word_embed_num']
embed_mat = np.load(opt['word_embed'])
train_title = np.load(opt['train_title_word' + prefix])
train_desc = np.load(opt['train_desc_word' + prefix])
train_label = np.load(opt['train_label'])
val_title = np.load(opt['val_title_word' + prefix])
val_desc = np.load(opt['val_desc_word' + prefix])
val_label = np.load(opt['val_label'])
logger.info('Load word data finished!')
elif opt['use_char_word']:
logger.info('Load char-word data starting...')
embed_mat_char = np.load(opt['char_embed'])
embed_mat_word = np.load(opt['word_embed'])
embed_mat = np.vstack((embed_mat_char, embed_mat_word))
train_title = np.load(opt['train_title_char' + prefix])
train_desc = np.load(opt['train_desc_word' + prefix])
train_label = np.load(opt['train_label'])
val_title = np.load(opt['val_title_char' + prefix])
val_desc = np.load(opt['val_desc_word' + prefix])
val_label = np.load(opt['val_label'])
logger.info('Load char-word data finished!')
elif opt['use_word_char']:
logger.info('Load word-char data starting...')
embed_mat_char = np.load(opt['char_embed'])
embed_mat_word = np.load(opt['word_embed'])
embed_mat = np.vstack((embed_mat_char, embed_mat_word))
train_title = np.load(opt['train_title_word' + prefix])
train_desc = np.load(opt['train_desc_char' + prefix])
train_label = np.load(opt['train_label'])
val_title = np.load(opt['val_title_word' + prefix])
val_desc = np.load(opt['val_desc_char' + prefix])
val_label = np.load(opt['val_label'])
logger.info('Load word-char data finished!')
train_dataset = Dataset(title=train_title,
desc=train_desc,
label=train_label,
class_num=opt['class_num'])
train_loader = data.DataLoader(train_dataset,
shuffle=True,
batch_size=opt['batch_size'])
val_dataset = Dataset(title=val_title,
desc=val_desc,
label=val_label,
class_num=opt['class_num'])
val_loader = data.DataLoader(val_dataset,
shuffle=False,
batch_size=opt['batch_size'])
logger.info('Using model {}'.format(opt['model']))
Model = getattr(models, opt['model'])
model = Model(embed_mat, opt)
print model
loss_weight = torch.ones(opt['class_num'])
if opt['boost']:
if opt['base_layer'] != 0:
cal_res = torch.load('{}/{}/layer_{}_cal_res_3.pt'.format(
opt['model_dir'], opt['model'], opt['base_layer']),
map_location=lambda storage, loc: storage)
logger.info('Load cal_res successful!')
loss_weight = torch.load('{}/{}/layer_{}_loss_weight_3.pt'.format(
opt['model_dir'], opt['model'], opt['base_layer'] + 1),
map_location=lambda storage, loc: storage)
else:
cal_res = torch.zeros(opt['val_num'], opt['class_num'])
print 'cur_layer:', opt['base_layer'] + 1, \
'loss_weight:', loss_weight.mean(), loss_weight.max(), loss_weight.min(), loss_weight.std()
if opt['use_self_loss']:
Loss = getattr(models, opt['loss_function'])
else:
Loss = getattr(nn, opt['loss_function'])
if opt['load']:
if opt.get('load_name', None) is None:
model = load_model(model,
model_dir=opt['model_dir'],
model_name=opt['model'])
else:
model = load_model(model, model_dir=opt['model_dir'], model_name=opt['model'], \
name=opt['load_name'])
if opt['cuda'] and opt['device'] != None:
torch.cuda.set_device(opt['device'])
if opt['cuda']:
model.cuda()
loss_weight = loss_weight.cuda()
# import sys
# precision, recall, score = eval(val_loader, model, opt, save_res=True)
# print precision, recall, score
# sys.exit()
loss_function = Loss(weight=loss_weight + 1 - loss_weight.mean())
optimizer = torch.optim.Adam(model.parameters(), lr=opt['lr'])
logger.info('Start running...')
steps = 0
model.train()
base_epoch = opt['base_epoch']
for epoch in range(1, opt['epochs'] + 1):
for i, batch in enumerate(train_loader, 0):
title, desc, label = batch
title, desc, label = Variable(title), Variable(desc), Variable(
label).float()
if opt['cuda']:
title, desc, label = title.cuda(), desc.cuda(), label.cuda()
optimizer.zero_grad()
logit = model(title, desc)
loss = loss_function(logit, label)
loss.backward()
optimizer.step()
steps += 1
if steps % opt['log_interval'] == 0:
corrects = ((logit.data >
opt['threshold']) == (label.data).byte()).sum()
accuracy = 100.0 * corrects / (opt['batch_size'] *
opt['class_num'])
log_info = 'Steps[{:>8}] (epoch[{:>2}] / batch[{:>5}]) - loss: {:.6f}, acc: {:.4f} % ({} / {})'.format( \
steps, epoch + base_epoch, (i+1), loss.data[0], accuracy, \
corrects, opt['batch_size'] * opt['class_num'])
logger.info(log_info)
vis.plot('loss', loss.data[0])
precision, recall, score = eval(batch,
model,
opt,
isBatch=True)
vis.plot('score', score)
logger.info('Training epoch {} finished!'.format(epoch + base_epoch))
precision, recall, score = eval(val_loader, model, opt)
log_info = 'Epoch[{}] - score: {:.6f} (precision: {:.4f}, recall: {:.4f})'.format( \
epoch + base_epoch, score, precision, recall)
vis.log(log_info)
save_model(model, model_dir=opt['model_dir'], model_name=opt['model'], \
epoch=epoch+base_epoch, score=score)
if epoch + base_epoch == 2:
model.opt['static'] = False
elif epoch + base_epoch == 4:
for param_group in optimizer.param_groups:
param_group['lr'] = opt['lr'] * opt['lr_decay']
elif epoch + base_epoch >= 5:
if opt['boost']:
res, truth = eval(val_loader, model, opt, return_res=True)
ori_score = get_score(cal_res, truth)
cal_res += res
cur_score = get_score(cal_res, truth)
logger.info('Layer {}: {}, Layer {}: {}'.format(
opt['base_layer'], ori_score, opt['base_layer'] + 1,
cur_score))
loss_weight = get_loss_weight(cal_res, truth)
torch.save(
cal_res, '{}/{}/layer_{}_cal_res_3.pt'.format(
opt['model_dir'], opt['model'], opt['base_layer'] + 1))
logger.info('Save cal_res successful!')
torch.save(
loss_weight, '{}/{}/layer_{}_loss_weight_3.pt'.format(
opt['model_dir'], opt['model'], opt['base_layer'] + 2))
break