Sklearn 0.23.2代码规范

1# -*- coding:utf-8 -*-
2""" sklearn train demo """
3import os
4import argparse
5import time
6from sklearn.model_selection import train_test_split
7from sklearn.metrics import f1_score
8from sklearn import svm
9import pandas as pd
10import numpy as np
11from rudder_autosearch.sdk.amaas_tools import AMaasTools
12
13def parse_arg():
14    """parse arguments"""
15    parser = argparse.ArgumentParser(description='Sklearn iris Example')
16    parser.add_argument('--train_dir', type=str, default='./train_data',
17                        help='input data dir for training (default: ./train_data)')
18    parser.add_argument('--test_dir', type=str, default='./test_data',
19                        help='input data dir for test (default: ./test_data)')
20    parser.add_argument('--output_dir', type=str, default='./output',
21                        help='output dir for auto_search job (default: ./output)')
22    parser.add_argument('--job_id', type=str, default="job-1234",
23                        help='auto_search job id')
24    parser.add_argument('--trial_id', type=str, default="0-0",
25                        help='auto_search id of a single trial')
26    parser.add_argument('--metric', type=str, default="f1_score",
27                        help='evaluation metric of the model')
28    parser.add_argument('--data_sampling_scale', type=float, default=1.0,
29                        help='sampling ratio of the dataset for auto_search (default: 1.0)')
30    parser.add_argument('--kernel', type=str, default='linear',
31                        help='kernel function (default: "linear")')
32    parser.add_argument('--C', type=float, default=1,
33                        help='penalty term (default: 1)')
34    parser.add_argument('--gamma', type=float, default=0.5,
35                        help='parameter of the kernel (default: 0.5)')
36
37    args = parser.parse_args()
38    args.output_dir = os.path.join(args.output_dir, args.job_id, args.trial_id)
39    if not os.path.exists(args.output_dir):
40        os.makedirs(args.output_dir)
41    print("job_id: {}, trial_id: {}".format(args.job_id, args.trial_id))
42    return args
43
44def load_data(train_dir, data_sampling_scale):
45    """ load data """
46    # 共150条数据，训练120条，测试30条，进行2,8分进行模型训练
47    # 每条数据类型为 x{nbarray} [6.4, 3.1, 5.5, 1.8]
48    # 上传的数据储存在./train_data和./test_data中
49    inputdata = pd.read_csv(train_dir + "/iris.csv")
50    target = inputdata["Species"]
51    inputdata = inputdata.drop(columns=["Species"])
52    # 切分，测试训练2,8分
53    x_train, x_test, y_train, y_test = train_test_split(inputdata, target, test_size=0.2, random_state=0)
54    train_data = np.concatenate([x_train, y_train.ravel().reshape([-1, 1])], axis=1)
55    np.random.seed(0)
56    np.random.shuffle(train_data)
57    train_data = train_data[0:int(data_sampling_scale * len(train_data))]
58    x_train, y_train = train_data[:, 0:-1], train_data[:, -1]
59    return (x_train, x_test), (y_train, y_test)
60
61def save_model(model, output_dir):
62    """ save model with pickle format """
63    import pickle
64    with open(output_dir + '/clf.pickle', 'wb') as f:
65        pickle.dump(model, f)
66
67def save_model_joblib(model, output_dir):
68    """ save model with joblib format """
69    try:
70        import joblib
71    except:
72        from sklearn.externals import joblib
73    joblib.dump(model, output_dir + '/clf.pkl')
74
75def evaluate(model, x_test, y_test):
76    """evaluate"""
77    # 多分类f1_score指标
78    predict = model.predict(x_test)
79    f1 = f1_score(y_test, predict, average="micro")
80    print("f1_score: %f" % f1)
81    return f1
82
83def report_final(args, metric):
84    """report_final_result"""
85    # 结果上报sdk
86    amaas_tools = AMaasTools(args.job_id, args.trial_id)
87    metric_dict = {args.metric: metric}
88    for i in range(3):
89        flag, ret_msg = amaas_tools.report_final_result(metric=metric_dict,
90                                                             export_model_path=args.output_dir,
91                                                             checkpoint_path="")
92        print("End Report, metric:{}, ret_msg:{}".format(metric, ret_msg))
93        if flag:
94            break
95        time.sleep(1)
96    assert flag, "Report final result to manager failed! Please check whether manager'address or manager'status " \
97                 "is ok! "
98
99def main():
100    """ main """
101    # 获取参数
102    args = parse_arg()
103    # 加载数据集
104    (x_train, x_test), (y_train, y_test) = load_data(args.train_dir, args.data_sampling_scale)
105    # 模型定义
106    model = svm.SVC(C=args.C, kernel=args.kernel, gamma=args.gamma)
107    # 模型训练
108    model.fit(x_train, y_train)
109    # 模型保存
110    save_model(model, args.output_dir)
111    # 模型评估
112    f1 = evaluate(model, x_test, y_test)
113    # 上报结果
114    report_final(args, metric=f1)
115
116if __name__ == '__main__':
117    main()

1#搜索算法参数
2search_strategy:
3  algo: RANDOM_SEARCH #搜索策略：随机搜索
4
5#单次训练时数据的采样比例，单位%
6data_sampling_scale: 100 #｜（0，100] int类型
7
8#最大搜索次数
9max_trial_num: 10 # ｜>0 int类型
10
11#评价指标参数
12metrics:
13  name: f1_score #评价指标 ｜ 任意字符串 str类型
14  goal: MAXIMIZE #最大值/最小值 ｜ str类型   MAXIMIZE or MINIMIZE   必须为这两个之一（也即支持大写）
15  expected_value: 100 #早停标准值，评价指标超过该值则结束整个超参搜索，单位% ｜无限制 int类型
16
17#搜索参数空间
18search_space:
19  kernel: #核函数
20    htype: choice
21    value: ["linear", "rbf"]
22  C: #惩罚项
23    htype: loguniform
24    value: [0.001, 1000]
25  gamma: #核函数参数
26    htype: loguniform
27    value: [0.0001, 1]