writer.py 14.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
# 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.
# =======================================================================
import os
import time
import numpy as np
走神的阿圆's avatar
走神的阿圆 已提交
18
from visualdl.writer.record_writer import RecordFileWriter
走神的阿圆's avatar
走神的阿圆 已提交
19
from visualdl.component.base_component import scalar, image, embedding, audio, histogram, pr_curve, meta_data
20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68


class DummyFileWriter(object):
    """A fake file writer that writes nothing to the disk.
    """

    def __init__(self, logdir):
        self._logdir = logdir

    def get_logdir(self):
        """Returns the directory where event file will be written."""
        return self._logdir

    def add_event(self, event, step=None, walltime=None):
        return

    def add_summary(self, summary, global_step=None, walltime=None):
        return

    def add_graph(self, graph_profile, walltime=None):
        return

    def add_onnx_graph(self, graph, walltime=None):
        return

    def flush(self):
        return

    def close(self):
        return

    def reopen(self):
        return


class LogWriter(object):
    """Log writer to write vdl records to log file.

    The class `LogWriter` provides APIs to create record file and add records to
    it. The class updates log file asynchronously without slowing down training.
    """

    def __init__(self,
                 logdir=None,
                 comment='',
                 max_queue=10,
                 flush_secs=120,
                 filename_suffix='',
                 write_to_disk=True,
走神的阿圆's avatar
走神的阿圆 已提交
69
                 display_name='',
70
                 file_name='',
71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97
                 **kwargs):
        """Create a instance of class `LogWriter` and create a vdl log file with
        given args.

        Args:
            logdir (string): Directory of log file. Default is
                `runs/**current_time**.**comment**`.
            comment (string): Suffix appended to the default `logdir`.It has no
                effect if `logidr` is assigned.
            max_queue (int): Size of queue for pending records.
            flush_secs (int): The duration to flush the pending records in queue
                to disk.
            filename_suffix (string): Suffix added to vdl log file.
            write_to_disk (boolean): Write to disk if it is True.
        """
        if not logdir:
            from datetime import datetime
            current_time = datetime.now().strftime('%b%d_%H-%M-%S')
            if '' != comment:
                comment = '.' + comment
            logdir = os.path.join('runs', current_time + comment)
        self._logdir = logdir
        self._max_queue = max_queue
        self._flush_secs = flush_secs
        self._filename_suffix = filename_suffix
        self._write_to_disk = write_to_disk
        self.kwargs = kwargs
98
        self._file_name = file_name
99 100 101 102 103

        self._file_writer = None
        self._all_writers = {}
        self._get_file_writer()
        self.loggers = {}
走神的阿圆's avatar
走神的阿圆 已提交
104
        self.add_meta(display_name=display_name)
105

106 107 108 109
    @property
    def logdir(self):
        return self._logdir

110 111 112 113 114 115 116 117 118 119 120
    def _get_file_writer(self):
        if not self._write_to_disk:
            self._file_writer = DummyFileWriter(logdir=self._logdir)
            self._all_writers.update({self._logdir: self._file_writer})
            return self._file_writer

        if self._all_writers is {} or self._file_writer is None:
            self._file_writer = RecordFileWriter(
                logdir=self._logdir,
                max_queue_size=self._max_queue,
                flush_secs=self._flush_secs,
121 122
                filename_suffix=self._filename_suffix,
                filename=self._file_name)
123 124 125
            self._all_writers.update({self._logdir: self._file_writer})
        return self._file_writer

126 127 128 129
    @property
    def file_name(self):
        return self._file_writer.get_filename()

走神的阿圆's avatar
走神的阿圆 已提交
130 131 132 133 134 135 136 137 138 139 140
    def add_meta(self, tag='meta_data_tag', display_name='', step=0, walltime=None):
        """Add a meta to vdl record file.

        Args:
            tag (string): Data identifier
            display_name (string): Display name of `runs`.
            step (int): Step of meta.
            walltime (int): Wall time of scalar
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
走神的阿圆's avatar
走神的阿圆 已提交
141
        walltime = round(time.time() * 1000) if walltime is None else walltime
走神的阿圆's avatar
走神的阿圆 已提交
142 143 144 145
        self._get_file_writer().add_record(
            meta_data(tag=tag, display_name=display_name, step=step,
                      walltime=walltime))

146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161
    def add_scalar(self, tag, value, step, walltime=None):
        """Add a scalar to vdl record file.

        Args:
            tag (string): Data identifier
            value (float): Value of scalar
            step (int): Step of scalar
            walltime (int): Wall time of scalar

        Example:
            for index in range(1, 101):
                writer.add_scalar(tag="train/loss", value=index*0.2, step=index)
                writer.add_scalar(tag="train/lr", value=index*0.5, step=index)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
走神的阿圆's avatar
走神的阿圆 已提交
162
        walltime = round(time.time() * 1000) if walltime is None else walltime
163 164 165
        self._get_file_writer().add_record(
            scalar(tag=tag, value=value, step=step, walltime=walltime))

走神的阿圆's avatar
走神的阿圆 已提交
166
    def add_image(self, tag, img, step, walltime=None, dataformats="HWC"):
167 168 169 170
        """Add an image to vdl record file.

        Args:
            tag (string): Data identifier
171
            img (np.ndarray): Image represented by a numpy.array
172 173
            step (int): Step of image
            walltime (int): Wall time of image
174
            dataformats (string): Format of image
175 176 177 178 179 180 181 182 183 184 185

        Example:
            from PIL import Image
            import numpy as np

            I = Image.open("./test.png")
            I_array = np.array(I)
            writer.add_image(tag="lll", img=I_array, step=0)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
走神的阿圆's avatar
走神的阿圆 已提交
186
        walltime = round(time.time() * 1000) if walltime is None else walltime
187
        self._get_file_writer().add_record(
走神的阿圆's avatar
走神的阿圆 已提交
188 189
            image(tag=tag, image_array=img, step=step, walltime=walltime,
                  dataformats=dataformats))
190

191
    def add_embeddings(self, tag, labels, hot_vectors, labels_meta=None, walltime=None):
192 193 194
        """Add embeddings to vdl record file.

        Args:
195
            tag (string): Data identifier
196
            labels (numpy.array or list): A 1D or 2D matrix of labels
197 198
            hot_vectors (numpy.array or list): A matrix which each row is
                feature of labels.
199
            labels_meta (numpy.array or list): Meta data of labels.
200 201
            walltime (int): Wall time of embeddings.

202
        Example 1:
203 204 205 206 207 208 209 210
            hot_vectors = [
            [1.3561076367500755, 1.3116267195134017, 1.6785401875616097],
            [1.1039614644440658, 1.8891609992484688, 1.32030488587171],
            [1.9924524852447711, 1.9358920727142739, 1.2124401279391606],
            [1.4129542689796446, 1.7372166387197474, 1.7317806077076527],
            [1.3913371800587777, 1.4684674577930312, 1.5214136352476377]]

            labels = ["label_1", "label_2", "label_3", "label_4", "label_5"]
211 212 213 214 215 216 217
            # or like this
            # labels = [["label_1", "label_2", "label_3", "label_4", "label_5"]]

            writer.add_embeddings(tag='default',
                                  labels=labels,
                                  vectors=hot_vectors,
                                  walltime=round(time.time() * 1000))
218

219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236
        Example 2:
            hot_vectors = [
            [1.3561076367500755, 1.3116267195134017, 1.6785401875616097],
            [1.1039614644440658, 1.8891609992484688, 1.32030488587171],
            [1.9924524852447711, 1.9358920727142739, 1.2124401279391606],
            [1.4129542689796446, 1.7372166387197474, 1.7317806077076527],
            [1.3913371800587777, 1.4684674577930312, 1.5214136352476377]]

            labels = [["label_a_1", "label_a_2", "label_a_3", "label_a_4", "label_a_5"],
                      ["label_b_1", "label_b_2", "label_b_3", "label_b_4", "label_b_5"]]

            labels_meta = ["label_a", "label_2"]

            writer.add_embeddings(tag='default',
                                  labels=labels,
                                  labels_meta=labels_meta,
                                  vectors=hot_vectors,
                                  walltime=round(time.time() * 1000))
237 238 239 240 241 242 243
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
        if isinstance(hot_vectors, np.ndarray):
            hot_vectors = hot_vectors.tolist()
        if isinstance(labels, np.ndarray):
            labels = labels.tolist()
244 245 246 247

        if isinstance(labels[0], list) and not labels_meta:
            labels_meta = ["label_%d" % i for i in range(len(labels))]

248
        step = 0
走神的阿圆's avatar
走神的阿圆 已提交
249
        walltime = round(time.time() * 1000) if walltime is None else walltime
250 251 252 253
        self._get_file_writer().add_record(
            embedding(
                tag=tag,
                labels=labels,
254
                labels_meta=labels_meta,
255 256 257 258 259 260 261 262 263 264 265 266 267 268
                hot_vectors=hot_vectors,
                step=step,
                walltime=walltime))

    def add_audio(self,
                  tag,
                  audio_array,
                  step,
                  sample_rate=8000,
                  walltime=None):
        """Add an audio to vdl record file.

        Args:
            tag (string): Data identifier
269
            audio (np.ndarray or list): audio represented by a numpy.array
270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293
            step (int): Step of audio
            sample_rate (int): Sample rate of audio
            walltime (int): Wall time of audio

        Example:
            import wave

            CHUNK = 4096
            f = wave.open(audio_path, "rb")
            wavdata = []
            chunk = f.readframes(CHUNK)
            while chunk:
                data = np.frombuffer(chunk, dtype='uint8')
                wavdata.extend(data)
                chunk = f.readframes(CHUNK)
            audio_data = np.array(wavdata)

            writer.add_audio(tag="audio_test",
                             audio_array=audio_data,
                             step=0,
                             sample_rate=8000)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
走神的阿圆's avatar
走神的阿圆 已提交
294
        walltime = round(time.time() * 1000) if walltime is None else walltime
295 296 297 298 299 300 301 302 303 304
        if isinstance(audio_array, list):
            audio_array = np.array(audio_array)
        self._get_file_writer().add_record(
            audio(
                tag=tag,
                audio_array=audio_array,
                sample_rate=sample_rate,
                step=step,
                walltime=walltime))

305 306 307 308 309 310 311 312 313 314
    def add_histogram(self,
                      tag,
                      values,
                      step,
                      walltime=None,
                      buckets=10):
        """Add an histogram to vdl record file.

        Args:
            tag (string): Data identifier
315
            value (np.ndarray or list): value represented by a numpy.array or list
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
            step (int): Step of histogram
            walltime (int): Wall time of audio
            buckets (int): Number of buckets, default is 10

        Example:
            values = np.arange(0, 1000)
            with LogWriter(logdir="./log/histogram_test/train") as writer:
                for index in range(5):
                    writer.add_histogram(tag='default',
                                         values=values+index,
                                         step=index)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
        hist, bin_edges = np.histogram(values, bins=buckets)
走神的阿圆's avatar
走神的阿圆 已提交
331
        walltime = round(time.time() * 1000) if walltime is None else walltime
332 333 334 335 336 337 338 339
        self._get_file_writer().add_record(
            histogram(
                tag=tag,
                hist=hist,
                bin_edges=bin_edges,
                step=step,
                walltime=walltime))

走神的阿圆's avatar
走神的阿圆 已提交
340 341 342 343 344 345 346 347 348 349 350 351
    def add_pr_curve(self,
                     tag,
                     labels,
                     predictions,
                     step,
                     num_thresholds=10,
                     weights=None,
                     walltime=None):
        """Add an precision-recall curve to vdl record file.

        Args:
            tag (string): Data identifier
352 353
            labels (np.ndarray or list): Binary labels for each element.
            predictions (np.ndarray or list): The probability that an element
走神的阿圆's avatar
走神的阿圆 已提交
354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371
                be classified as true.
            step (int): Step of pr curve.
            weights (float): Multiple of data to display on the curve.
            num_thresholds (int): Number of thresholds used to draw the curve.
            walltime (int): Wall time of pr curve.

        Example:
            with LogWriter(logdir="./log/pr_curve_test/train") as writer:
                for index in range(3):
                    labels = np.random.randint(2, size=100)
                    predictions = np.random.rand(100)
                    writer.add_pr_curve(tag='default',
                                        labels=labels,
                                        predictions=predictions,
                                        step=index)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
走神的阿圆's avatar
走神的阿圆 已提交
372
        walltime = round(time.time() * 1000) if walltime is None else walltime
走神的阿圆's avatar
走神的阿圆 已提交
373 374 375 376 377 378 379 380 381 382 383
        self._get_file_writer().add_record(
            pr_curve(
                tag=tag,
                labels=labels,
                predictions=predictions,
                step=step,
                walltime=walltime,
                num_thresholds=num_thresholds,
                weights=weights
                ))

384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407
    def flush(self):
        """Flush all data in cache to disk.
        """
        if self._all_writers is {}:
            return
        for writer in self._all_writers.values():
            writer.flush()

    def close(self):
        """Close all writers after flush data to disk.
        """
        if self._all_writers is {}:
            return
        for writer in self._all_writers.values():
            writer.flush()
            writer.close()
        self._file_writer = None
        self._all_writers = {}

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_val, exc_tb):
        self.close()