reader.py 12.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# Copyright (c) 2020 VisualDL Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# =======================================================================
走神的阿圆's avatar
走神的阿圆 已提交
15
import collections
走神的阿圆's avatar
走神的阿圆 已提交
16
from functools import partial  # noqa: F401
17

18
from visualdl.component import components
19 20
from visualdl.io import bfile
from visualdl.proto import record_pb2
21 22
from visualdl.reader.record_reader import RecordReader
from visualdl.server.data_manager import default_data_manager
23 24
from visualdl.utils.string_util import decode_tag
from visualdl.utils.string_util import encode_tag
25 26


27
def is_VDLRecord_file(path, check=False):
28 29 30 31 32 33
    """Determine whether it is a VDL log file according to the file name.

    File name of a VDL log file must contain `vdlrecords`.

    Args:
        path: File name to determine.
34
        check: Check file is valid or not.
35 36 37 38

    Returns:
        True if the file is a VDL log file, otherwise false.
    """
P
Peter Pan 已提交
39
    if "vdlrecords" not in path:
40 41 42 43 44 45 46 47 48
        return False
    if check:
        _reader = RecordReader(filepath=path)
        meta_data = _reader.get_next()
        record = record_pb2.Record()
        record.ParseFromString(meta_data)
        if 'meta_data_tag' != record.values[0].tag:
            return False
    return True
49 50 51 52 53 54 55


class LogReader(object):
    """Log reader to read vdl log, support for frontend api in lib.py.

    """

56
    def __init__(self, logdir='', file_path=''):
57 58 59 60 61
        """Instance of LogReader

        Args:
            logdir: The dir include vdl log files, multiple subfolders allowed.
        """
62 63 64 65 66
        if isinstance(logdir, str):
            self.dir = [logdir]
        else:
            self.dir = logdir

67 68 69 70
        self.reader = None
        self.readers = {}
        self.walks = None
        self._tags = {}
走神的阿圆's avatar
走神的阿圆 已提交
71 72
        self.name2tags = {}
        self.tags2name = {}
73 74 75

        self.file_readers = {}

走神的阿圆's avatar
走神的阿圆 已提交
76
        # {'run': {'scalar': {'tag1': data, 'tag2': data}}}
P
Peter Pan 已提交
77
        self._log_datas = collections.defaultdict(
78 79
            lambda: collections.defaultdict(lambda: collections.defaultdict(
                list)))
走神的阿圆's avatar
走神的阿圆 已提交
80

81
        if file_path:
82 83
            self._log_data = collections.defaultdict(lambda: collections.
                                                     defaultdict(list))
84
            self.get_file_reader(file_path=file_path)
走神的阿圆's avatar
走神的阿圆 已提交
85 86 87 88 89 90 91 92 93 94 95
            remain = self.get_remain()
            self.read_log_data(remain=remain)
            components_name = components.keys()
            for name in components_name:
                exec("self.get_%s=partial(self.get_data, '%s')" % (name, name))
        elif logdir:
            self.data_manager = default_data_manager
            self.load_new_data(update=True)
            self._a_tags = {}

            self._model = ""
走神的阿圆's avatar
走神的阿圆 已提交
96

97 98 99 100 101 102
    @property
    def model(self):
        return self._model

    @model.setter
    def model(self, model_path):
走神的阿圆's avatar
走神的阿圆 已提交
103 104 105
        self._model = model_path
        with bfile.BFile(model_path, 'rb') as bfp:
            if not bfp.isfile(model_path):
106 107 108
                print(
                    "Model path %s should be file path, please check this path."
                    % model_path)
109
            else:
走神的阿圆's avatar
走神的阿圆 已提交
110 111 112
                if bfile.exists(model_path):
                    self._model = model_path
                else:
113 114
                    print("Model path %s is invalid, please check this path." %
                          model_path)
115

116 117 118 119
    @property
    def logdir(self):
        return self.dir

走神的阿圆's avatar
走神的阿圆 已提交
120 121 122 123 124 125 126 127
    def parsing_from_proto(self, component, proto_datas):
        data = []
        if 'scalar' == component:
            for item in proto_datas:
                data.append([item.id, item.tag, item.timestamp, item.value])
            return data
        return proto_datas

走神的阿圆's avatar
走神的阿圆 已提交
128 129 130 131
    def _get_log_tags(self):
        component_keys = self._log_data.keys()
        log_tags = {}
        for key in component_keys:
走神的阿圆's avatar
走神的阿圆 已提交
132 133 134 135
            _tags = list(self._log_data[key].keys())
            tags = list(map(lambda x: encode_tag(x), _tags))
            log_tags[key] = tags

走神的阿圆's avatar
走神的阿圆 已提交
136 137
        return log_tags

走神的阿圆's avatar
走神的阿圆 已提交
138
    def get_log_data(self, component, run, tag):
139 140 141
        if (run in self._log_datas.keys()
                and component in self._log_datas[run].keys()
                and tag in self._log_datas[run][component].keys()):
走神的阿圆's avatar
走神的阿圆 已提交
142 143 144 145 146
            return self._log_datas[run][component][tag]
        else:
            file_path = bfile.join(run, self.walks[run])
            reader = self._get_file_reader(file_path=file_path, update=False)
            remain = self.get_remain(reader=reader)
147 148
            data = self.read_log_data(
                remain=remain, update=False)[component][tag]
走神的阿圆's avatar
走神的阿圆 已提交
149 150 151 152
            data = self.parsing_from_proto(component, data)
            self._log_datas[run][component][tag] = data
            return data

走神的阿圆's avatar
走神的阿圆 已提交
153 154 155 156
    def get_tags(self):
        return self._get_log_tags()

    def get_data(self, component, tag):
走神的阿圆's avatar
走神的阿圆 已提交
157
        return self._log_data[component][decode_tag(tag)]
走神的阿圆's avatar
走神的阿圆 已提交
158

159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
    def parse_from_bin(self, record_bin):
        """Register to self._tags by component type.

        Args:
            record_bin: Binary data from vdl log file.
        """
        record = record_pb2.Record()
        record.ParseFromString(record_bin)
        value = record.values[0]
        tag = decode_tag(value.tag)
        path = bfile.join(self.reader.dir, tag)

        if path not in self._tags.keys():
            value_type = value.WhichOneof("one_value")
            if "value" == value_type:
                component = "scalar"
            elif "image" == value_type:
                component = "image"
            elif "embeddings" == value_type:
                component = "embeddings"
            elif "audio" == value_type:
                component = "audio"
181 182
            elif "histogram" == value_type:
                component = "histogram"
走神的阿圆's avatar
走神的阿圆 已提交
183 184
            elif "pr_curve" == value_type:
                component = "pr_curve"
P
Peter Pan 已提交
185 186
            elif "roc_curve" == value_type:
                component = "roc_curve"
走神的阿圆's avatar
走神的阿圆 已提交
187 188 189
            elif "meta_data" == value_type:
                self.update_meta_data(record)
                component = "meta_data"
走神的阿圆's avatar
走神的阿圆 已提交
190 191
            elif "text" == value_type:
                component = "text"
走神的阿圆's avatar
走神的阿圆 已提交
192 193
            elif "hparam" == value_type:
                component = "hyper_parameters"
194 195 196 197 198 199
            else:
                raise TypeError("Invalid value type `%s`." % value_type)
            self._tags[path] = component

        return self._tags[path], self.reader.dir, tag, value

走神的阿圆's avatar
走神的阿圆 已提交
200 201 202 203 204 205
    def update_meta_data(self, record):
        meta = record.values[0].meta_data
        if meta.display_name:
            self.name2tags[meta.display_name] = self.reader.dir
            self.tags2name[self.reader.dir] = meta.display_name

206 207
    def get_all_walk(self):
        self.walks = {}
208 209 210
        for dir in self.dir:
            for root, dirs, files in bfile.walk(dir):
                self.walks.update({root: files})
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249

    def logs(self, update=False):
        """Get logs.

        Every dir(means `run` in vdl) has only one log(meads `actual log file`).

        Returns:
            walks: A dict like {"exp1": "vdlrecords.1587375595.log",
                                "exp2": "vdlrecords.1587375685.log"}
        """
        if self.walks is None or update is True:
            self.get_all_walk()

            walks_temp = {}
            for run, tags in self.walks.items():
                tags_temp = [tag for tag in tags if is_VDLRecord_file(tag)]
                tags_temp.sort(reverse=True)
                if len(tags_temp) > 0:
                    walks_temp.update({run: tags_temp[0]})
            self.walks = walks_temp
        return self.walks

    def get_log_reader(self, dir, log):
        """Get log reader for every vdl log file.

        Get instance of class RecordReader base on BFile. Note that each
        `log` may contain multi `tag`.

        Args:
            dir: Dir name of log.
            log: Vdl log file name.
        """
        if self.walks is None:
            self.logs()
        if self.walks.get(dir, None) != log:
            raise FileNotFoundError("Can't find file %s.", (dir + "/" + log))

        filepath = bfile.join(dir, log)
        if filepath not in self.readers.keys():
走神的阿圆's avatar
走神的阿圆 已提交
250
            self.register_reader(filepath, dir)
251 252 253
        self.reader = self.readers[filepath]
        return self.reader

254
    def get_file_reader(self, file_path):
走神的阿圆's avatar
走神的阿圆 已提交
255 256 257 258 259
        """Get file reader for specified vdl log file.

        Get instance of class RecordReader base on BFile.

        Args:
260
            file_path: Vdl log file path.
走神的阿圆's avatar
走神的阿圆 已提交
261
        """
走神的阿圆's avatar
走神的阿圆 已提交
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280
        return self._get_file_reader(file_path, True)

    def _get_file_reader(self, file_path, update=True):
        if update:
            self.register_reader(file_path)
            self.reader = self.readers[file_path]
            self.reader.dir = file_path
            return self.reader
        else:
            reader = RecordReader(filepath=file_path)
            return reader

    def register_reader(self, path, dir=None, update=True):
        if update:
            if path not in list(self.readers.keys()):
                reader = RecordReader(filepath=path, dir=dir)
                self.readers[path] = reader
        else:
            pass
281 282 283 284 285 286 287 288 289 290

    def register_readers(self, update=False):
        """Register all readers for all vdl log files.

        Args:
            update: Need update if `update` is True.
        """
        self.logs(update)
        for dir, path in self.walks.items():
            filepath = bfile.join(dir, path)
走神的阿圆's avatar
走神的阿圆 已提交
291
            self.register_reader(filepath, dir)
292 293 294 295 296 297 298 299 300 301 302 303

    def add_remain(self):
        """Add remain data to data_manager.

        Add remain data to data manager according its component type and tag
        one by one.
        """
        for reader in self.readers.values():
            self.reader = reader
            remain = self.reader.get_remain()
            for item in remain:
                component, dir, tag, record = self.parse_from_bin(item)
走神的阿圆's avatar
走神的阿圆 已提交
304

305 306 307
                self.data_manager.add_item(component, self.reader.dir, tag,
                                           record)

走神的阿圆's avatar
走神的阿圆 已提交
308
    def get_remain(self, reader=None):
309 310
        """Get all remain data by self.reader.
        """
走神的阿圆's avatar
走神的阿圆 已提交
311
        if self.reader is None and reader is None:
312
            raise RuntimeError("Please specify log path!")
313 314
        return self.reader.get_remain(
        ) if reader is None else reader.get_remain()
315

走神的阿圆's avatar
走神的阿圆 已提交
316
    def read_log_data(self, remain, update=True):
走神的阿圆's avatar
走神的阿圆 已提交
317 318 319 320 321
        """Parse data from log file without sampling.

        Args:
            remain: Raw data from log file.
        """
322 323
        _log_data = collections.defaultdict(lambda: collections.defaultdict(
            list))
走神的阿圆's avatar
走神的阿圆 已提交
324 325
        for item in remain:
            component, dir, tag, record = self.parse_from_bin(item)
走神的阿圆's avatar
走神的阿圆 已提交
326 327 328 329
            _log_data[component][tag].append(record)
        if update:
            self._log_data = _log_data
        return _log_data
走神的阿圆's avatar
走神的阿圆 已提交
330 331 332 333 334

    @property
    def log_data(self):
        return self._log_data

335 336 337 338 339 340 341 342 343
    def runs(self, update=True):
        self.logs(update=update)
        return list(self.walks.keys())

    def tags(self):
        if self._tags is None:
            self.add_remain()
        return self._tags

344
    def components(self, update=False):
345 346
        """Get components type used by vdl.
        """
347 348
        if self.logdir is None:
            return set()
349 350
        if update is True:
            self.load_new_data(update=update)
351
        components_set = set(self._tags.values())
352

353
        return components_set
354

355 356 357 358 359 360 361 362 363 364 365 366 367 368
    def component_tabs(self, update=False):
        """Get component tabs used by vdl frontend.
        """
        component_tabs = set()
        if not self.logdir:
            return component_tabs
        if update is True:
            self.load_new_data(update=update)
        for component in set(self._tags.values()):
            if component == 'meta_data':
                continue
            component_tabs.add(component)
        return component_tabs

369 370 371 372 373 374
    def load_new_data(self, update=True):
        """Load remain data.

        Make sure all readers for every vdl log file are registered, load all
        remain data.
        """
375 376 377
        if self.logdir is not None:
            self.register_readers(update=update)
            self.add_remain()
走神的阿圆's avatar
走神的阿圆 已提交
378 379 380 381 382 383

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_val, exc_tb):
        pass