kagle_util.py 8.6 KB
Newer Older
X
xiexionghang 已提交
1 2 3
"""
Util lib
"""
X
xiexionghang 已提交
4 5 6
import os
import time
import datetime
T
tangwei 已提交
7 8
import kagle.utils.kagle_fs as kagle_fs

X
xiexionghang 已提交
9 10

def get_env_value(env_name):
X
xiexionghang 已提交
11 12 13
    """
    get os environment value
    """
X
xiexionghang 已提交
14 15
    return os.popen("echo -n ${" + env_name + "}").read().strip()

X
xiexionghang 已提交
16

X
xiexionghang 已提交
17
def now_time_str():
X
xiexionghang 已提交
18 19 20 21
    """
    get current format str_time
    """
    return "\n" + time.strftime("%Y-%m-%d %H:%M:%S", time.localtime()) + "[0]:"
X
xiexionghang 已提交
22

X
xiexionghang 已提交
23

X
xiexionghang 已提交
24
def get_absolute_path(path, params):
X
xiexionghang 已提交
25 26
    """R
    """
X
xiexionghang 已提交
27 28
    if path.startswith('afs:') or path.startswith('hdfs:'):
        sub_path = path.split('fs:')[1]
T
tangwei 已提交
29
        if ':' in sub_path:  # such as afs://xxx:prot/xxxx
X
xiexionghang 已提交
30 31 32 33 34 35
            return path
        elif 'fs_name' in params:
            return params['fs_name'] + sub_path
    else:
        return path

X
xiexionghang 已提交
36

X
xiexionghang 已提交
37
def make_datetime(date_str, fmt=None):
X
xiexionghang 已提交
38 39 40 41 42 43 44 45
    """
    create a datetime instance by date_string
    Args:
        date_str: such as 2020-01-14
        date_str_format: "%Y-%m-%d"
    Return:
        datetime 
    """
X
xiexionghang 已提交
46
    if fmt is None:
T
tangwei 已提交
47
        if len(date_str) == 8:  # %Y%m%d
X
xiexionghang 已提交
48
            return datetime.datetime.strptime(date_str, '%Y%m%d')
T
tangwei 已提交
49
        if len(date_str) == 12:  # %Y%m%d%H%M
X
xiexionghang 已提交
50 51 52 53 54
            return datetime.datetime.strptime(date_str, '%Y%m%d%H%M')
    return datetime.datetime.strptime(date_str, fmt)


def rank0_print(log_str):
X
xiexionghang 已提交
55 56
    """R
    """
X
xiexionghang 已提交
57 58
    print_log(log_str, {'master': True})

X
xiexionghang 已提交
59

X
xiexionghang 已提交
60
def print_cost(cost, params):
X
xiexionghang 已提交
61 62
    """R
    """
X
xiexionghang 已提交
63
    log_str = params['log_format'] % cost
T
tangwei 已提交
64
    print_log(log_str, params)
X
xiexionghang 已提交
65
    return log_str
T
tangwei 已提交
66

X
xiexionghang 已提交
67

X
xiexionghang 已提交
68
class CostPrinter(object):
X
xiexionghang 已提交
69 70 71
    """
    For count cost time && print cost log
    """
T
tangwei 已提交
72

X
xiexionghang 已提交
73
    def __init__(self, callback, callback_params):
X
xiexionghang 已提交
74 75
        """R
        """
X
xiexionghang 已提交
76 77
        self.reset(callback, callback_params)
        pass
T
tangwei 已提交
78

X
xiexionghang 已提交
79
    def __del__(self):
X
xiexionghang 已提交
80 81
        """R
        """
X
xiexionghang 已提交
82 83 84
        if not self._done:
            self.done()
        pass
T
tangwei 已提交
85

X
xiexionghang 已提交
86
    def reset(self, callback, callback_params):
X
xiexionghang 已提交
87 88
        """R
        """
X
xiexionghang 已提交
89 90 91 92 93
        self._done = False
        self._callback = callback
        self._callback_params = callback_params
        self._begin_time = time.time()
        pass
T
tangwei 已提交
94

X
xiexionghang 已提交
95
    def done(self):
X
xiexionghang 已提交
96 97
        """R
        """
X
xiexionghang 已提交
98
        cost = time.time() - self._begin_time
T
tangwei 已提交
99
        log_str = self._callback(cost, self._callback_params)  # cost(s)
X
xiexionghang 已提交
100 101 102
        self._done = True
        return cost, log_str

X
xiexionghang 已提交
103 104

class PathGenerator(object):
X
xiexionghang 已提交
105 106 107
    """
    generate path with template & runtime variables
    """
T
tangwei 已提交
108

X
xiexionghang 已提交
109
    def __init__(self, config):
X
xiexionghang 已提交
110 111
        """R
        """
T
tangwei 已提交
112
        self._templates = {}
X
xiexionghang 已提交
113 114
        self.add_path_template(config)
        pass
T
tangwei 已提交
115

X
xiexionghang 已提交
116
    def add_path_template(self, config):
X
xiexionghang 已提交
117 118
        """R
        """
X
xiexionghang 已提交
119 120 121 122 123 124
        if 'templates' in config:
            for template in config['templates']:
                self._templates[template['name']] = template['template']
        pass

    def generate_path(self, template_name, param):
X
xiexionghang 已提交
125 126
        """R
        """
X
xiexionghang 已提交
127 128 129 130 131 132 133 134
        if template_name in self._templates:
            if 'time_format' in param:
                str = param['time_format'].strftime(self._templates[template_name])
                return str.format(**param)
            return self._templates[template_name].format(**param)
        else:
            return ""

X
xiexionghang 已提交
135

X
xiexionghang 已提交
136
class TimeTrainPass(object):
X
xiexionghang 已提交
137 138 139 140
    """
    timely pass
    define pass time_interval && start_time && end_time
    """
T
tangwei 已提交
141

X
xiexionghang 已提交
142
    def __init__(self, global_config):
X
xiexionghang 已提交
143 144
        """R
        """
X
xiexionghang 已提交
145 146 147 148 149 150
        self._config = global_config['epoch']
        if '+' in self._config['days']:
            day_str = self._config['days'].replace(' ', '')
            day_fields = day_str.split('+')
            self._begin_day = make_datetime(day_fields[0].strip())
            if len(day_fields) == 1 or len(day_fields[1]) == 0:
T
tangwei 已提交
151 152 153
                # 100 years, meaning to continuous running
                self._end_day = self._begin_day + datetime.timedelta(days=36500)
            else:
X
xiexionghang 已提交
154 155
                # example: 2020212+10 
                run_day = int(day_fields[1].strip())
T
tangwei 已提交
156 157
                self._end_day = self._begin_day + datetime.timedelta(days=run_day)
        else:
X
xiexionghang 已提交
158 159 160 161 162 163
            # example: {20191001..20191031}
            days = os.popen("echo -n " + self._config['days']).read().split(" ")
            self._begin_day = make_datetime(days[0])
            self._end_day = make_datetime(days[len(days) - 1])
        self._checkpoint_interval = self._config['checkpoint_interval']
        self._dump_inference_interval = self._config['dump_inference_interval']
T
tangwei 已提交
164
        self._interval_per_pass = self._config['train_time_interval']  # train N min data per pass
X
xiexionghang 已提交
165 166 167 168 169 170 171 172 173 174

        self._pass_id = 0
        self._inference_pass_id = 0
        self._pass_donefile_handler = None
        if 'pass_donefile_name' in self._config:
            self._train_pass_donefile = global_config['output_path'] + '/' + self._config['pass_donefile_name']
            if kagle_fs.is_afs_path(self._train_pass_donefile):
                self._pass_donefile_handler = kagle_fs.FileHandler(global_config['io']['afs'])
            else:
                self._pass_donefile_handler = kagle_fs.FileHandler(global_config['io']['local_fs'])
T
tangwei 已提交
175

X
xiexionghang 已提交
176 177 178 179 180 181
            last_done = self._pass_donefile_handler.cat(self._train_pass_donefile).strip().split('\n')[-1]
            done_fileds = last_done.split('\t')
            if len(done_fileds) > 4:
                self._base_key = done_fileds[1]
                self._checkpoint_model_path = done_fileds[2]
                self._checkpoint_pass_id = int(done_fileds[3])
T
tangwei 已提交
182
                self._inference_pass_id = int(done_fileds[4])
X
xiexionghang 已提交
183 184 185
                self.init_pass_by_id(done_fileds[0], self._checkpoint_pass_id)

    def max_pass_num_day(self):
X
xiexionghang 已提交
186 187
        """R
        """
X
xiexionghang 已提交
188
        return 24 * 60 / self._interval_per_pass
T
tangwei 已提交
189

X
xiexionghang 已提交
190
    def save_train_progress(self, day, pass_id, base_key, model_path, is_checkpoint):
X
xiexionghang 已提交
191 192
        """R
        """
X
xiexionghang 已提交
193 194 195
        if is_checkpoint:
            self._checkpoint_pass_id = pass_id
            self._checkpoint_model_path = model_path
T
tangwei 已提交
196 197
        done_content = "%s\t%s\t%s\t%s\t%d\n" % (day, base_key,
                                                 self._checkpoint_model_path, self._checkpoint_pass_id, pass_id)
X
xiexionghang 已提交
198 199 200 201
        self._pass_donefile_handler.write(done_content, self._train_pass_donefile, 'a')
        pass

    def init_pass_by_id(self, date_str, pass_id):
X
xiexionghang 已提交
202 203 204 205 206 207
        """
        init pass context with pass_id
        Args:
            date_str: example "20200110"
            pass_id(int): pass_id of date
        """
T
tangwei 已提交
208
        date_time = make_datetime(date_str)
X
xiexionghang 已提交
209 210 211 212 213 214 215
        if pass_id < 1:
            pass_id = 0
        if (date_time - self._begin_day).total_seconds() > 0:
            self._begin_day = date_time
        self._pass_id = pass_id
        mins = self._interval_per_pass * (pass_id - 1)
        self._current_train_time = date_time + datetime.timedelta(minutes=mins)
T
tangwei 已提交
216

X
xiexionghang 已提交
217
    def init_pass_by_time(self, datetime_str):
X
xiexionghang 已提交
218 219 220 221 222
        """
        init pass context with datetime
        Args:
            date_str: example "20200110000" -> "%Y%m%d%H%M"
        """
X
xiexionghang 已提交
223
        self._current_train_time = make_datetime(datetime_str)
X
xiexionghang 已提交
224
        minus = self._current_train_time.hour * 60 + self._current_train_time.minute
X
xiexionghang 已提交
225 226
        self._pass_id = minus / self._interval_per_pass + 1

X
xiexionghang 已提交
227 228 229
    def current_pass(self):
        """R
        """
X
xiexionghang 已提交
230
        return self._pass_id
T
tangwei 已提交
231

X
xiexionghang 已提交
232
    def next(self):
X
xiexionghang 已提交
233 234
        """R
        """
X
xiexionghang 已提交
235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
        has_next = True
        old_pass_id = self._pass_id
        if self._pass_id < 1:
            self.init_pass_by_time(self._begin_day.strftime("%Y%m%d%H%M"))
        else:
            next_time = self._current_train_time + datetime.timedelta(minutes=self._interval_per_pass)
            if (next_time - self._end_day).total_seconds() > 0:
                has_next = False
            else:
                self.init_pass_by_time(next_time.strftime("%Y%m%d%H%M"))
        if has_next and (self._inference_pass_id < self._pass_id or self._pass_id < old_pass_id):
            self._inference_pass_id = self._pass_id - 1
        return has_next

    def is_checkpoint_pass(self, pass_id):
X
xiexionghang 已提交
250 251
        """R
        """
X
xiexionghang 已提交
252 253 254 255 256 257 258
        if pass_id < 1:
            return True
        if pass_id == self.max_pass_num_day():
            return False
        if pass_id % self._checkpoint_interval == 0:
            return True
        return False
T
tangwei 已提交
259

X
xiexionghang 已提交
260
    def need_dump_inference(self, pass_id):
X
xiexionghang 已提交
261 262
        """R
        """
X
xiexionghang 已提交
263 264 265
        return self._inference_pass_id < pass_id and pass_id % self._dump_inference_interval == 0

    def date(self, delta_day=0):
X
xiexionghang 已提交
266 267 268 269 270 271 272
        """
        get train date
        Args:
            delta_day(int): n day afer current_train_date
        Return:
            date(current_train_time + delta_day)
        """
X
xiexionghang 已提交
273 274 275
        return (self._current_train_time + datetime.timedelta(days=delta_day)).strftime("%Y%m%d")

    def timestamp(self, delta_day=0):
X
xiexionghang 已提交
276 277
        """R
        """
X
xiexionghang 已提交
278
        return (self._current_train_time + datetime.timedelta(days=delta_day)).timestamp()