writer.py 20.7 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.
# =======================================================================
P
Peter Pan 已提交
15

16 17 18
import os
import time
import numpy as np
走神的阿圆's avatar
走神的阿圆 已提交
19
from visualdl.writer.record_writer import RecordFileWriter
走神的阿圆's avatar
走神的阿圆 已提交
20
from visualdl.server.log import logger
走神的阿圆's avatar
走神的阿圆 已提交
21
from visualdl.utils.img_util import merge_images
F
Feiyu Chan 已提交
22
from visualdl.utils.figure_util import figure_to_image
走神的阿圆's avatar
走神的阿圆 已提交
23 24
from visualdl.component.base_component import scalar, image, embedding, audio, \
    histogram, pr_curve, roc_curve, meta_data, text
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 69 70 71 72 73


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
74
                 display_name='',
75
                 file_name='',
76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102
                 **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
103
        self._file_name = file_name
104 105 106 107 108

        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
109
        self.add_meta(display_name=display_name)
110

111 112 113 114
    @property
    def logdir(self):
        return self._logdir

115 116 117 118 119 120 121 122 123 124 125
    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,
126 127
                filename_suffix=self._filename_suffix,
                filename=self._file_name)
128 129 130
            self._all_writers.update({self._logdir: self._file_writer})
        return self._file_writer

131 132 133 134
    @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
135 136 137 138 139 140 141 142 143 144 145
    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
146
        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
147 148 149 150
        self._get_file_writer().add_record(
            meta_data(tag=tag, display_name=display_name, step=step,
                      walltime=walltime))

151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
    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
167
        walltime = round(time.time() * 1000) if walltime is None else walltime
168 169 170
        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
171
    def add_image(self, tag, img, step, walltime=None, dataformats="HWC"):
172 173 174 175
        """Add an image to vdl record file.

        Args:
            tag (string): Data identifier
176
            img (np.ndarray): Image represented by a numpy.array
177 178
            step (int): Step of image
            walltime (int): Wall time of image
179
            dataformats (string): Format of image
180 181 182 183 184 185 186 187 188 189 190

        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
191
        walltime = round(time.time() * 1000) if walltime is None else walltime
192
        self._get_file_writer().add_record(
走神的阿圆's avatar
走神的阿圆 已提交
193 194
            image(tag=tag, image_array=img, step=step, walltime=walltime,
                  dataformats=dataformats))
195

F
Feiyu Chan 已提交
196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
    def add_figure(self, tag, figure, step, walltime=None):
        """Add an figure to vdl record file.

        Args:
            tag (string): Data identifier
            figure (matplotlib.figure.Figure): Image represented by a Figure
            step (int): Step of image
            walltime (int): Wall time of image
            dataformats (string): Format of image

        Example:
            form matplotlib import pyplot as plt
            import numpy as np

            x = np.arange(100)
            y = x ** 2 + 1
            plt.plot(x, y)
            fig = plt.gcf()
            writer.add_figure(tag="lll", figure=fig, step=0)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
        walltime = round(time.time() * 1000) if walltime is None else walltime
        img = figure_to_image(figure)
        self._get_file_writer().add_record(
            image(tag=tag, image_array=img, step=step, walltime=walltime))

走神的阿圆's avatar
走神的阿圆 已提交
223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242
    def add_text(self, tag, text_string, step=None, walltime=None):
        """Add an text to vdl record file.
        Args:
            tag (string): Data identifier
            text_string (string): Value of text
            step (int): Step of text
            walltime (int): Wall time of text
        Example:
            for index in range(1, 101):
                writer.add_text(tag="train/loss", text_string=str(index) + 'text', step=index)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
        walltime = round(
            time.time() * 1000) if walltime is None else walltime
        self._get_file_writer().add_record(
            text(
                tag=tag, text_string=text_string, step=step,
                walltime=walltime))

走神的阿圆's avatar
走神的阿圆 已提交
243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272
    def add_image_matrix(self, tag, imgs, step, rows=-1, scale=1.0, walltime=None, dataformats="HWC"):
        """Add an image to vdl record file.

        Args:
            tag (string): Data identifier
            imgs (np.ndarray): Image represented by a numpy.array
            step (int): Step of image
            rows (int): Number of rows, -1 means as close as possible to the square
            scale (float): Image zoom scale
            walltime (int): Wall time of image
            dataformats (string): Format of image

        Example:
            from PIL import Image
            import numpy as np

            I = Image.open("./test.png")
            I_array = np.array([I, I, I])
            writer.add_image_matrix(tag="lll", imgs=I_array, step=0)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
        walltime = round(time.time() * 1000) if walltime is None else walltime
        img = merge_images(imgs=imgs, dataformats=dataformats, scale=scale, rows=rows)
        self.add_image(tag=tag,
                       img=img,
                       step=step,
                       walltime=walltime,
                       dataformats=dataformats)

走神的阿圆's avatar
走神的阿圆 已提交
273 274 275
    def add_embeddings(self, tag, mat=None, metadata=None,
                       metadata_header=None, walltime=None, labels=None,
                       hot_vectors=None, labels_meta=None):
276 277 278
        """Add embeddings to vdl record file.

        Args:
279
            tag (string): Data identifier
走神的阿圆's avatar
走神的阿圆 已提交
280
            mat (numpy.array or list): A matrix which each row is
281
                feature of labels.
走神的阿圆's avatar
走神的阿圆 已提交
282 283
            metadata (numpy.array or list): A 1D or 2D matrix of labels
            metadata_header (numpy.array or list): Meta data of labels.
284
            walltime (int): Wall time of embeddings.
走神的阿圆's avatar
走神的阿圆 已提交
285 286 287 288 289 290
            labels (numpy.array or list): Obsolete parameter, use `metadata` to
                replace it.
            hot_vectors (numpy.array or list): Obsolete parameter, use `mat` to
                replace it.
            labels_meta (numpy.array or list): Obsolete parameter, use
                `metadata_header` to replace it.
291
        Example 1:
走神的阿圆's avatar
走神的阿圆 已提交
292
            mat = [
293 294 295 296 297 298
            [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]]

走神的阿圆's avatar
走神的阿圆 已提交
299
            metadata = ["label_1", "label_2", "label_3", "label_4", "label_5"]
300
            # or like this
走神的阿圆's avatar
走神的阿圆 已提交
301
            # metadata = [["label_1", "label_2", "label_3", "label_4", "label_5"]]
302 303

            writer.add_embeddings(tag='default',
走神的阿圆's avatar
走神的阿圆 已提交
304 305
                                  metadata=metadata,
                                  mat=mat,
306
                                  walltime=round(time.time() * 1000))
307

308
        Example 2:
走神的阿圆's avatar
走神的阿圆 已提交
309
            mat = [
310 311 312 313 314 315
            [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]]

走神的阿圆's avatar
走神的阿圆 已提交
316
            metadata = [["label_a_1", "label_a_2", "label_a_3", "label_a_4", "label_a_5"],
317 318
                      ["label_b_1", "label_b_2", "label_b_3", "label_b_4", "label_b_5"]]

走神的阿圆's avatar
走神的阿圆 已提交
319
            metadata_header = ["label_a", "label_2"]
320 321

            writer.add_embeddings(tag='default',
走神的阿圆's avatar
走神的阿圆 已提交
322 323 324
                                  metadata=metadata,
                                  metadata_header=metadata_header,
                                  mat=mat,
325
                                  walltime=round(time.time() * 1000))
326 327 328
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
329
        if (mat is None) and hot_vectors:
走神的阿圆's avatar
走神的阿圆 已提交
330 331 332 333
            mat = hot_vectors
            logger.warning('Parameter `hot_vectors` in function '
                           '`add_embeddings` will be deprecated in '
                           'future, use `mat` instead.')
334
        if (metadata is None) and labels:
走神的阿圆's avatar
走神的阿圆 已提交
335 336 337 338
            metadata = labels
            logger.warning(
                'Parameter `labels` in function `add_embeddings` will be '
                'deprecated in future, use `metadata` instead.')
339
        if (metadata_header is None) and labels_meta:
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
            metadata_header = labels_meta
            logger.warning(
                'Parameter `labels_meta` in function `add_embeddings` will be'
                ' deprecated in future, use `metadata_header` instead.')
        if isinstance(mat, np.ndarray):
            mat = mat.tolist()
        if isinstance(metadata, np.ndarray):
            metadata = metadata.tolist()

        if isinstance(metadata[0], list) and not metadata_header:
            metadata_header = ["label_%d" % i for i in range(len(metadata))]
351

352
        step = 0
走神的阿圆's avatar
走神的阿圆 已提交
353
        walltime = round(time.time() * 1000) if walltime is None else walltime
354 355 356
        self._get_file_writer().add_record(
            embedding(
                tag=tag,
走神的阿圆's avatar
走神的阿圆 已提交
357 358 359
                labels=metadata,
                labels_meta=metadata_header,
                hot_vectors=mat,
360 361 362 363 364 365 366 367 368 369 370 371 372
                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
373
            audio (np.ndarray or list): audio represented by a numpy.array
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
            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
398
        walltime = round(time.time() * 1000) if walltime is None else walltime
399 400 401 402 403 404 405 406 407 408
        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))

409 410 411 412 413 414 415 416 417 418
    def add_histogram(self,
                      tag,
                      values,
                      step,
                      walltime=None,
                      buckets=10):
        """Add an histogram to vdl record file.

        Args:
            tag (string): Data identifier
419
            value (np.ndarray or list): value represented by a numpy.array or list
420 421 422 423 424 425 426 427 428 429 430 431 432 433 434
            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
435
        walltime = round(time.time() * 1000) if walltime is None else walltime
436 437 438 439 440 441 442 443
        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
444 445 446 447 448 449 450 451 452 453 454 455
    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
456 457
            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
458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475
                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
476
        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
477 478 479 480 481 482 483 484 485 486
        self._get_file_writer().add_record(
            pr_curve(
                tag=tag,
                labels=labels,
                predictions=predictions,
                step=step,
                walltime=walltime,
                num_thresholds=num_thresholds,
                weights=weights
                ))
P
Peter Pan 已提交
487

P
Peter Pan 已提交
488
    def add_roc_curve(self,
P
Peter Pan 已提交
489 490 491 492 493 494 495
                      tag,
                      labels,
                      predictions,
                      step,
                      num_thresholds=10,
                      weights=None,
                      walltime=None):
P
Peter Pan 已提交
496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529
        """Add an ROC curve to vdl record file.
        Args:
            tag (string): Data identifier
            labels (numpy.ndarray or list): Binary labels for each element.
            predictions (numpy.ndarray or list): The probability that an element
                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/roc_curve_test/train") as writer:
                for index in range(3):
                    labels = np.random.randint(2, size=100)
                    predictions = np.random.rand(100)
                    writer.add_roc_curve(tag='default',
                                        labels=labels,
                                        predictions=predictions,
                                        step=index)
        """
        if '%' in tag:
            raise RuntimeError("% can't appear in tag!")
        walltime = round(time.time() * 1000) if walltime is None else walltime
        self._get_file_writer().add_record(
            roc_curve(
                tag=tag,
                labels=labels,
                predictions=predictions,
                step=step,
                walltime=walltime,
                num_thresholds=num_thresholds,
                weights=weights
                ))

530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553
    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()