Skip to content

GitLab

"vscode:/vscode.git/clone" did not exist on "7ee6089ea84d2b075c38e45f7178f5cee0eaca7e"
Commit 72124364 authored by Hongkun Yu's avatar Hongkun Yu Committed by A. Unique TensorFlower
Browse files

Delete keras_application_models which is no longer needed. 

PiperOrigin-RevId: 263910123
parent a7264343
Show whitespace changes
Inline Side-by-side
Showing with 0 additions and 506 deletions
official/keras_application_models/README.md deleted 100644 → 0
View file @ a7264343
# Keras Application Models Benchmark
## Overview
This provides a single scaffold to benchmark the Keras built-in application [models](https://keras.io/applications/). All the models are for image classification applications, and include:
- Xception
- VGG16
- VGG19
- ResNet50
- InceptionV3
- InceptionResNetV2
- MobileNet
- DenseNet
- NASNet
## Dataset
Synthetic dataset is used for the benchmark.
## Callbacks
Two custom callbacks are provided for model benchmarking: ExamplesPerSecondCallback and LoggingMetricCallback. For each callback, `epoch_based` and `batch_based` options are available to set the benchmark level. Check [model_callbacks.py](model_callbacks.py) for more details.
## Running Code
To benchmark a model, use `--model` to specify the model name. To perform the benchmark with eager execution, issue the following command:
```
python benchmark_main.py --model resnet50 --eager
```
Note that, if eager execution is enabled, only one GPU is utilized even if multiple GPUs are provided and multi_gpu_model is used.
To use distribution strategy in the benchmark, run the following:
```
python benchmark_main.py --model resnet50 --dist_strat
```
Currently, only one of the --eager and --dist_strat arguments can be defined, as DistributionStrategy is not supported in Eager execution now.
Arguments:
* `--model`: Which model to be benchmarked. The model name is defined as the keys of `MODELS` in [benchmark_main.py](benchmark_main.py).
* `--callbacks`: To specify a list of callbacks.
Use the `--help` or `-h` flag to get a full list of possible arguments.
official/keras_application_models/benchmark_main.py deleted 100644 → 0
View file @ a7264343
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# 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.
# ==============================================================================
"""Benchmark on the keras built-in application models."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
# pylint: disable=g-bad-import-order
import numpy as np
from absl import app as absl_app
from absl import flags
import tensorflow as tf
# pylint: enable=g-bad-import-order
from official.keras_application_models import dataset
from official.keras_application_models import model_callbacks
from official.utils.flags import core as flags_core
from official.utils.logs import logger
from official.utils.misc import distribution_utils
# Define a dictionary that maps model names to their model classes inside Keras
MODELS = {
"vgg16": tf.keras.applications.VGG16,
"vgg19": tf.keras.applications.VGG19,
"inceptionv3": tf.keras.applications.InceptionV3,
"xception": tf.keras.applications.Xception,
"resnet50": tf.keras.applications.ResNet50,
"inceptionresnetv2": tf.keras.applications.InceptionResNetV2,
"mobilenet": tf.keras.applications.MobileNet,
"densenet121": tf.keras.applications.DenseNet121,
"densenet169": tf.keras.applications.DenseNet169,
"densenet201": tf.keras.applications.DenseNet201,
"nasnetlarge": tf.keras.applications.NASNetLarge,
"nasnetmobile": tf.keras.applications.NASNetMobile,
}
def run_keras_model_benchmark(_):
"""Run the benchmark on keras model."""
# Ensure a valid model name was supplied via command line argument
if FLAGS.model not in MODELS.keys():
raise AssertionError("The --model command line argument should "
"be a key in the `MODELS` dictionary.")
# Check if eager execution is enabled
if FLAGS.eager:
tf.logging.info("Eager execution is enabled...")
tf.enable_eager_execution()
# Load the model
tf.logging.info("Benchmark on {} model...".format(FLAGS.model))
keras_model = MODELS[FLAGS.model]
# Get dataset
dataset_name = "ImageNet"
if FLAGS.use_synthetic_data:
tf.logging.info("Using synthetic dataset...")
dataset_name += "_Synthetic"
train_dataset = dataset.generate_synthetic_input_dataset(
FLAGS.model, FLAGS.batch_size)
val_dataset = dataset.generate_synthetic_input_dataset(
FLAGS.model, FLAGS.batch_size)
model = keras_model(weights=None)
else:
tf.logging.info("Using CIFAR-10 dataset...")
dataset_name = "CIFAR-10"
ds = dataset.Cifar10Dataset(FLAGS.batch_size)
train_dataset = ds.train_dataset
val_dataset = ds.test_dataset
model = keras_model(
weights=None, input_shape=ds.input_shape, classes=ds.num_classes)
num_gpus = flags_core.get_num_gpus(FLAGS)
distribution = None
# Use distribution strategy
if FLAGS.dist_strat:
distribution = distribution_utils.get_distribution_strategy(
distribution_strategy=FLAGS.distribution_strategy,
num_gpus=num_gpus)
elif num_gpus > 1:
# Run with multi_gpu_model
# If eager execution is enabled, only one GPU is utilized even if multiple
# GPUs are provided.
if FLAGS.eager:
tf.logging.warning(
"{} GPUs are provided, but only one GPU is utilized as "
"eager execution is enabled.".format(num_gpus))
model = tf.keras.utils.multi_gpu_model(model, gpus=num_gpus)
# Adam optimizer and some other optimizers doesn't work well with
# distribution strategy (b/113076709)
# Use GradientDescentOptimizer here
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.001)
model.compile(loss="categorical_crossentropy",
optimizer=optimizer,
metrics=["accuracy"],
distribute=distribution)
# Create benchmark logger for benchmark logging
run_params = {
"batch_size": FLAGS.batch_size,
"synthetic_data": FLAGS.use_synthetic_data,
"train_epochs": FLAGS.train_epochs,
"num_train_images": FLAGS.num_train_images,
"num_eval_images": FLAGS.num_eval_images,
}
benchmark_logger = logger.get_benchmark_logger()
benchmark_logger.log_run_info(
model_name=FLAGS.model,
dataset_name=dataset_name,
run_params=run_params,
test_id=FLAGS.benchmark_test_id)
# Create callbacks that log metric values about the training and evaluation
callbacks = model_callbacks.get_model_callbacks(
FLAGS.callbacks,
batch_size=FLAGS.batch_size,
metric_logger=benchmark_logger)
# Train and evaluate the model
history = model.fit(
train_dataset,
epochs=FLAGS.train_epochs,
callbacks=callbacks,
validation_data=val_dataset,
steps_per_epoch=int(np.ceil(FLAGS.num_train_images / FLAGS.batch_size)),
validation_steps=int(np.ceil(FLAGS.num_eval_images / FLAGS.batch_size))
)
tf.logging.info("Logging the evaluation results...")
for epoch in range(FLAGS.train_epochs):
eval_results = {
"accuracy": history.history["val_acc"][epoch],
"loss": history.history["val_loss"][epoch],
tf.GraphKeys.GLOBAL_STEP: (epoch + 1) * np.ceil(
FLAGS.num_eval_images/FLAGS.batch_size)
}
benchmark_logger.log_evaluation_result(eval_results)
# Clear the session explicitly to avoid session delete error
tf.keras.backend.clear_session()
def define_keras_benchmark_flags():
"""Add flags for keras built-in application models."""
flags_core.define_base(hooks=False)
flags_core.define_performance()
flags_core.define_image()
flags_core.define_benchmark()
flags.adopt_module_key_flags(flags_core)
flags_core.set_defaults(
data_format="channels_last",
use_synthetic_data=True,
batch_size=32,
train_epochs=2)
flags.DEFINE_enum(
name="model", default=None,
enum_values=MODELS.keys(), case_sensitive=False,
help=flags_core.help_wrap(
"Model to be benchmarked."))
flags.DEFINE_integer(
name="num_train_images", default=1000,
help=flags_core.help_wrap(
"The number of synthetic images for training. The default value is "
"1000."))
flags.DEFINE_integer(
name="num_eval_images", default=50,
help=flags_core.help_wrap(
"The number of synthetic images for evaluation. The default value is "
"50."))
flags.DEFINE_boolean(
name="eager", default=False, help=flags_core.help_wrap(
"To enable eager execution. Note that if eager execution is enabled, "
"only one GPU is utilized even if multiple GPUs are provided and "
"multi_gpu_model is used."))
flags.DEFINE_boolean(
name="dist_strat", default=False, help=flags_core.help_wrap(
"To enable distribution strategy for model training and evaluation. "
"Number of GPUs used for distribution strategy can be set by the "
"argument --num_gpus."))
flags.DEFINE_list(
name="callbacks",
default=["ExamplesPerSecondCallback", "LoggingMetricCallback"],
help=flags_core.help_wrap(
"A list of (case insensitive) strings to specify the names of "
"callbacks. For example: `--callbacks ExamplesPerSecondCallback,"
"LoggingMetricCallback`"))
@flags.multi_flags_validator(
["eager", "dist_strat"],
message="Both --eager and --dist_strat were set. Only one can be "
"defined, as DistributionStrategy is not supported in Eager "
"execution currently.")
# pylint: disable=unused-variable
def _check_eager_dist_strat(flag_dict):
return not(flag_dict["eager"] and flag_dict["dist_strat"])
def main(_):
with logger.benchmark_context(FLAGS):
run_keras_model_benchmark(FLAGS)
if __name__ == "__main__":
tf.logging.set_verbosity(tf.logging.INFO)
define_keras_benchmark_flags()
FLAGS = flags.FLAGS
absl_app.run(main)
official/keras_application_models/dataset.py deleted 100644 → 0
View file @ a7264343
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# 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.
# ==============================================================================
"""Prepare dataset for keras model benchmark."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import tensorflow as tf
from official.utils.misc import model_helpers # pylint: disable=g-bad-import-order
# Default values for dataset.
_NUM_CHANNELS = 3
_NUM_CLASSES = 1000
def _get_default_image_size(model):
"""Provide default image size for each model."""
image_size = (224, 224)
if model in ["inceptionv3", "xception", "inceptionresnetv2"]:
image_size = (299, 299)
elif model in ["nasnetlarge"]:
image_size = (331, 331)
return image_size
def generate_synthetic_input_dataset(model, batch_size):
"""Generate synthetic dataset."""
image_size = _get_default_image_size(model)
image_shape = (batch_size,) + image_size + (_NUM_CHANNELS,)
label_shape = (batch_size, _NUM_CLASSES)
dataset = model_helpers.generate_synthetic_data(
input_shape=tf.TensorShape(image_shape),
label_shape=tf.TensorShape(label_shape),
)
return dataset
class Cifar10Dataset(object):
"""CIFAR10 dataset, including train and test set.
Each sample consists of a 32x32 color image, and label is from 10 classes.
"""
def __init__(self, batch_size):
"""Initializes train/test datasets.
Args:
batch_size: int, the number of batch size.
"""
self.input_shape = (32, 32, 3)
self.num_classes = 10
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
y_train, y_test = y_train.astype(np.int64), y_test.astype(np.int64)
y_train = tf.keras.utils.to_categorical(y_train, self.num_classes)
y_test = tf.keras.utils.to_categorical(y_test, self.num_classes)
self.train_dataset = tf.data.Dataset.from_tensor_slices(
(x_train, y_train)).shuffle(2000).batch(batch_size).repeat()
self.test_dataset = tf.data.Dataset.from_tensor_slices(
(x_test, y_test)).shuffle(2000).batch(batch_size).repeat()
official/keras_application_models/model_callbacks.py deleted 100644 → 0
View file @ a7264343
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# 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.
# ==============================================================================
"""Callbacks for Keras built-in application models.
Note that, in the callbacks, the global_step is initialized in the __init__ of
each callback rather than on_train_begin. As on_train_begin gets called in
the fit_loop, and it will be reset with each call to fit(). To keep the
global_step persistent across all training sessions, it should be initialized in
the __init__.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import time
import tensorflow as tf # pylint: disable=g-bad-import-order
from official.utils.logs import logger
# Metrics to log after each batch and epoch
_PER_BATCH_METRICS = {
"loss": "train_loss",
"acc": "train_accuracy",
}
_PER_EPOCH_METRICS = {
"loss": "train_loss",
"acc": "train_accuracy",
"val_loss": "loss",
"val_acc": "accuracy"
}
class ExamplesPerSecondCallback(tf.keras.callbacks.Callback):
"""ExamplesPerSecond callback.
This callback records the average_examples_per_sec and
current_examples_per_sec during training.
"""
def __init__(self, batch_size, every_n_steps=1, metric_logger=None):
self._batch_size = batch_size
self._every_n_steps = every_n_steps
self._logger = metric_logger or logger.BaseBenchmarkLogger()
self._global_step = 0 # Initialize it in __init__
super(ExamplesPerSecondCallback, self).__init__()
def on_train_begin(self, logs=None):
self._train_start_time = time.time()
self._last_recorded_time = time.time()
def on_batch_end(self, batch, logs=None):
"""Log the examples_per_sec metric every_n_steps."""
self._global_step += 1
current_time = time.time()
if self._global_step % self._every_n_steps == 0:
average_examples_per_sec = self._batch_size * (
self._global_step / (current_time - self._train_start_time))
self._logger.log_metric(
"average_examples_per_sec", average_examples_per_sec,
global_step=self._global_step)
current_examples_per_sec = self._batch_size * (
self._every_n_steps / (current_time - self._last_recorded_time))
self._logger.log_metric(
"current_examples_per_sec", current_examples_per_sec,
global_step=self._global_step)
self._last_recorded_time = current_time # Update last_recorded_time
class LoggingMetricCallback(tf.keras.callbacks.Callback):
"""LoggingMetric callback.
Log the predefined _PER_BATCH_METRICS after each batch, and log the predefined
_PER_EPOCH_METRICS after each epoch.
"""
def __init__(self, metric_logger=None):
self._logger = metric_logger or logger.BaseBenchmarkLogger()
self._per_batch_metrics = _PER_BATCH_METRICS
self._per_epoch_metrics = _PER_EPOCH_METRICS
self._global_step = 0 # Initialize it in __init__
super(LoggingMetricCallback, self).__init__()
def on_batch_end(self, batch, logs=None):
"""Log metrics after each batch."""
self._global_step += 1
for metric in _PER_BATCH_METRICS:
self._logger.log_metric(
_PER_BATCH_METRICS[metric],
logs.get(metric),
global_step=self._global_step)
def on_epoch_end(self, epoch, logs=None):
"""Log metrics after each epoch."""
for metric in _PER_EPOCH_METRICS:
self._logger.log_metric(
_PER_EPOCH_METRICS[metric],
logs.get(metric),
global_step=self._global_step)
def get_model_callbacks(name_list, **kwargs):
"""Factory for getting a list of TensorFlow hooks for training by name.
Args:
name_list: a list of strings to name desired callback classes. Allowed:
ExamplesPerSecondCallback, LoggingMetricCallback, which are defined
as keys in CALLBACKS.
**kwargs: a dictionary of arguments to the callbacks.
Returns:
list of instantiated callbacks, ready to be used in a classifier.train call.
Raises:
ValueError: if an unrecognized name is passed.
"""
if not name_list:
return []
callbacks = []
for name in name_list:
callback_name = CALLBACKS.get(name.strip().lower())
if callback_name is None:
raise ValueError(
"Unrecognized training callback requested: {}".format(name))
else:
callbacks.append(callback_name(**kwargs))
return callbacks
def get_examples_per_second_callback(
every_n_steps=1, batch_size=32, metric_logger=None, **kwargs): # pylint: disable=unused-argument
"""Function to get ExamplesPerSecondCallback."""
return ExamplesPerSecondCallback(
batch_size=batch_size, every_n_steps=every_n_steps,
metric_logger=metric_logger or logger.get_benchmark_logger())
def get_logging_metric_callback(metric_logger=None, **kwargs): # pylint: disable=unused-argument
"""Function to get LoggingMetricCallback."""
return LoggingMetricCallback(
metric_logger=metric_logger or logger.get_benchmark_logger())
# A dictionary to map the callback name and its corresponding function
CALLBACKS = {
"examplespersecondcallback": get_examples_per_second_callback,
"loggingmetriccallback": get_logging_metric_callback,
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment