File size: 7,798 Bytes
efb03ac |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 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 69 70 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 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 |
import datetime
import os
from typing import List
import absl
import keras_tuner
import tensorflow as tf
from tensorflow.keras.optimizers import Adam
import tensorflow_transform as tft
from tensorflow_cloud import CloudTuner
from tfx.v1.components import TunerFnResult
from tfx.components.trainer.fn_args_utils import DataAccessor
from tfx.components.trainer.fn_args_utils import FnArgs
from tfx.dsl.io import fileio
from tfx_bsl.tfxio import dataset_options
import tfx.extensions.google_cloud_ai_platform.constants as vertex_const
import tfx.extensions.google_cloud_ai_platform.trainer.executor as vertex_training_const
import tfx.extensions.google_cloud_ai_platform.tuner.executor as vertex_tuner_const
_TRAIN_DATA_SIZE = 128
_EVAL_DATA_SIZE = 128
_TRAIN_BATCH_SIZE = 32
_EVAL_BATCH_SIZE = 32
_CLASSIFIER_LEARNING_RATE = 1e-3
_FINETUNE_LEARNING_RATE = 7e-6
_CLASSIFIER_EPOCHS = 30
_IMAGE_KEY = "image"
_LABEL_KEY = "label"
def INFO(text: str):
absl.logging.info(text)
def _transformed_name(key: str) -> str:
return key + "_xf"
def _get_signature(model):
signatures = {
"serving_default": _get_serve_image_fn(model).get_concrete_function(
tf.TensorSpec(
shape=[None, 224, 224, 3],
dtype=tf.float32,
name=_transformed_name(_IMAGE_KEY),
)
)
}
return signatures
def _get_serve_image_fn(model):
@tf.function
def serve_image_fn(image_tensor):
return model(image_tensor)
return serve_image_fn
def _image_augmentation(image_features):
batch_size = tf.shape(image_features)[0]
image_features = tf.image.random_flip_left_right(image_features)
image_features = tf.image.resize_with_crop_or_pad(image_features, 250, 250)
image_features = tf.image.random_crop(image_features, (batch_size, 224, 224, 3))
return image_features
def _data_augmentation(feature_dict):
image_features = feature_dict[_transformed_name(_IMAGE_KEY)]
image_features = _image_augmentation(image_features)
feature_dict[_transformed_name(_IMAGE_KEY)] = image_features
return feature_dict
def _input_fn(
file_pattern: List[str],
data_accessor: DataAccessor,
tf_transform_output: tft.TFTransformOutput,
is_train: bool = False,
batch_size: int = 200,
) -> tf.data.Dataset:
dataset = data_accessor.tf_dataset_factory(
file_pattern,
dataset_options.TensorFlowDatasetOptions(
batch_size=batch_size, label_key=_transformed_name(_LABEL_KEY)
),
tf_transform_output.transformed_metadata.schema,
)
if is_train:
dataset = dataset.map(lambda x, y: (_data_augmentation(x), y))
return dataset
def _get_hyperparameters() -> keras_tuner.HyperParameters:
hp = keras_tuner.HyperParameters()
hp.Choice("learning_rate", [1e-3, 1e-2], default=1e-3)
return hp
def _build_keras_model(hparams: keras_tuner.HyperParameters) -> tf.keras.Model:
base_model = tf.keras.applications.ResNet50(
input_shape=(224, 224, 3), include_top=False, weights="imagenet", pooling="max"
)
base_model.input_spec = None
base_model.trainable = False
model = tf.keras.Sequential(
[
tf.keras.layers.InputLayer(
input_shape=(224, 224, 3), name=_transformed_name(_IMAGE_KEY)
),
base_model,
tf.keras.layers.Dropout(0.1),
tf.keras.layers.Dense(10, activation="softmax"),
]
)
model.compile(
loss="sparse_categorical_crossentropy",
optimizer=Adam(learning_rate=hparams.get("learning_rate")),
metrics=["sparse_categorical_accuracy"],
)
model.summary(print_fn=INFO)
return model
def cloud_tuner_fn(fn_args: FnArgs) -> TunerFnResult:
TUNING_ARGS_KEY = vertex_tuner_const.TUNING_ARGS_KEY
TRAINING_ARGS_KEY = vertex_training_const.TRAINING_ARGS_KEY
VERTEX_PROJECT_KEY = "project"
VERTEX_REGION_KEY = "region"
tuner = CloudTuner(
_build_keras_model,
max_trials=6,
hyperparameters=_get_hyperparameters(),
project_id=fn_args.custom_config[TUNING_ARGS_KEY][VERTEX_PROJECT_KEY],
region=fn_args.custom_config[TUNING_ARGS_KEY][VERTEX_REGION_KEY],
objective="val_sparse_categorical_accuracy",
directory=fn_args.working_dir,
)
tf_transform_output = tft.TFTransformOutput(fn_args.transform_graph_path)
train_dataset = _input_fn(
fn_args.train_files,
fn_args.data_accessor,
tf_transform_output,
is_train=True,
batch_size=_TRAIN_BATCH_SIZE,
)
eval_dataset = _input_fn(
fn_args.eval_files,
fn_args.data_accessor,
tf_transform_output,
is_train=False,
batch_size=_EVAL_BATCH_SIZE,
)
return TunerFnResult(
tuner=tuner,
fit_kwargs={
"x": train_dataset,
"validation_data": eval_dataset,
"steps_per_epoch": steps_per_epoch,
"validation_steps": fn_args.eval_steps,
},
)
def tuner_fn(fn_args: FnArgs) -> TunerFnResult:
steps_per_epoch = int(_TRAIN_DATA_SIZE / _TRAIN_BATCH_SIZE)
tuner = keras_tuner.RandomSearch(
_build_keras_model,
max_trials=6,
hyperparameters=_get_hyperparameters(),
allow_new_entries=False,
objective=keras_tuner.Objective("val_sparse_categorical_accuracy", "max"),
directory=fn_args.working_dir,
project_name="img_classification_tuning",
)
tf_transform_output = tft.TFTransformOutput(fn_args.transform_graph_path)
train_dataset = _input_fn(
fn_args.train_files,
fn_args.data_accessor,
tf_transform_output,
is_train=True,
batch_size=_TRAIN_BATCH_SIZE,
)
eval_dataset = _input_fn(
fn_args.eval_files,
fn_args.data_accessor,
tf_transform_output,
is_train=False,
batch_size=_EVAL_BATCH_SIZE,
)
return TunerFnResult(
tuner=tuner,
fit_kwargs={
"x": train_dataset,
"validation_data": eval_dataset,
"steps_per_epoch": steps_per_epoch,
"validation_steps": fn_args.eval_steps,
},
)
def run_fn(fn_args: FnArgs):
steps_per_epoch = int(_TRAIN_DATA_SIZE / _TRAIN_BATCH_SIZE)
total_epochs = int(fn_args.train_steps / steps_per_epoch)
if _CLASSIFIER_EPOCHS > total_epochs:
raise ValueError("Classifier epochs is greater than the total epochs")
tf_transform_output = tft.TFTransformOutput(fn_args.transform_output)
train_dataset = _input_fn(
fn_args.train_files,
fn_args.data_accessor,
tf_transform_output,
is_train=True,
batch_size=_TRAIN_BATCH_SIZE,
)
eval_dataset = _input_fn(
fn_args.eval_files,
fn_args.data_accessor,
tf_transform_output,
is_train=False,
batch_size=_EVAL_BATCH_SIZE,
)
INFO("Tensorboard logging to {}".format(fn_args.model_run_dir))
tensorboard_callback = tf.keras.callbacks.TensorBoard(
log_dir=fn_args.model_run_dir, update_freq="batch"
)
if fn_args.hyperparameters:
hparams = keras_tuner.HyperParameters.from_config(fn_args.hyperparameters)
else:
hparams = _get_hyperparameters()
INFO(f"HyperParameters for training: ${hparams.get_config()}")
model = _build_keras_model(hparams)
model.fit(
train_dataset,
epochs=_CLASSIFIER_EPOCHS,
steps_per_epoch=steps_per_epoch,
validation_data=eval_dataset,
validation_steps=fn_args.eval_steps,
callbacks=[tensorboard_callback],
)
model.save(
fn_args.serving_model_dir, save_format="tf", signatures=_get_signature(model)
)
|