def create_trial_instance(trial_def: Type[det.Trial]) -> None:
with tempfile.TemporaryDirectory() as td:
trial_instance = experimental.create_trial_instance(
trial_def=trial_def,
config={"hyperparameters": {"global_batch_size": det.Constant(16)}},
checkpoint_dir=td,
)
check.check_isinstance(trial_instance, det.Trial)
dest="config",
help="Specifies Determined Experiment configuration.",
default="{}",
)
parser.add_argument("--local", action="store_true", help="Specifies local mode")
parser.add_argument("--test", action="store_true", help="Specifies test mode")
parser.add_argument(
"--use-fit",
action="store_true",
help="If true, uses model.fit() instead of model.fit_generator()",
)
args = parser.parse_args()
config = {
"hyperparameters": {
"global_batch_size": det.Constant(value=32),
"dense1": det.Constant(value=128),
},
"searcher": {"name": "single", "metric": "val_accuracy", "max_steps": 40},
}
config.update(json.loads(args.config))
context = init(config, local=args.local, test=args.test, context_dir=str(pathlib.Path.cwd()))
train_images, train_labels = data.load_training_data()
train_images = train_images / 255.0
train_data = _ArrayLikeAdapter(
x=train_images, y=train_labels, batch_size=context.get_per_slot_batch_size()
)
test_images, test_labels = data.load_validation_data()
)
return model
def build_training_data_loader(self):
return np.zeros(1), np.zeros(1)
def build_validation_data_loader(self):
return np.zeros(1), np.zeros(1)
if __name__ == "__main__":
experimental.create(
trial_def=RuntimeErrorTrial,
config={
"description": "keras_runtime_error",
"hyperparameters": {
"global_batch_size": det.Constant(1)
},
"searcher": {
"metric": "accuracy"
},
"data_layer": {
"type": "lfs",
"container_storage_path": "/tmp"
},
},
local=True,
test=True,
context_dir=str(pathlib.Path.cwd()),
)
parser = argparse.ArgumentParser()
parser.add_argument(
"--config",
dest="config",
help="Specifies Determined Experiment configuration.",
default="{}",
)
parser.add_argument(
"--mode", dest="mode", help="Specifies local mode or cluster mode.", default="cluster"
)
args = parser.parse_args()
config = {
"hyperparameters": {
"learning_rate": det.Log(-4.0, -2.0, 10),
"global_batch_size": det.Constant(64),
"hidden_layer_1": det.Constant(250),
"hidden_layer_2": det.Constant(250),
"hidden_layer_3": det.Constant(250),
"dropout": det.Double(0.0, 0.5),
},
"searcher": {
"name": "single",
"metric": "accuracy",
"max_steps": 10,
"smaller_is_better": False,
},
}
config.update(json.loads(args.config))
context = estimator.init(
)
parser.add_argument("--mode",
dest="mode",
help="Specifies local mode or cluster mode.",
default="cluster")
args = parser.parse_args()
config = {
"data": {
"url":
"https://s3-us-west-2.amazonaws.com/determined-ai-test-data/pytorch_mnist.tar.gz"
},
"hyperparameters": {
"learning_rate": det.Log(minval=-3.0, maxval=-1.0, base=10),
"dropout": det.Double(minval=0.2, maxval=0.8),
"global_batch_size": det.Constant(value=64),
"n_filters1": det.Constant(value=32),
"n_filters2": det.Constant(value=32),
},
"searcher": {
"name": "single",
"metric": "validation_error",
"max_steps": 20,
"smaller_is_better": True,
},
}
config.update(json.loads(args.config))
experimental.create(
trial_def=model_def.MNistTrial,
config=config,
action="store_true",
help="Specifies local mode")
parser.add_argument("--test",
action="store_true",
help="Specifies test mode")
args = parser.parse_args()
dataset_url = (
"https://determined-ai-public-datasets.s3-us-west-2.amazonaws.com/"
"PennFudanPed/PennFudanPed.zip")
config = {
"data": {
"url": dataset_url
},
"hyperparameters": {
"learning_rate": det.Constant(value=0.005),
"momentum": det.Constant(value=0.9),
"weight_decay": det.Constant(value=0.0005),
"global_batch_size": det.Constant(value=2),
},
"batches_per_step": 1,
"searcher": {
"name": "single",
"metric": "val_avg_iou",
"max_steps": 16,
"smaller_is_better": False,
},
}
config.update(json.loads(args.config))
experimental.create(
dest="mode",
help="Specifies test mode or submit mode.",
default="submit")
args = parser.parse_args()
config = {
"description": "Resnet50 Imagenet TF Keras",
"searcher": {
"name": "single",
"metric": "val_loss",
"max_steps": 1,
"smaller_is_better": True,
},
"min_validation_period": 1,
"hyperparameters": {
"global_batch_size": det.Constant(value=32),
"learning_rate": det.Constant(value=0.1),
},
}
ctx = keras.init(config=config,
mode=experimental.Mode(args.mode),
context_dir=str(pathlib.Path.cwd()))
lr_schedule = ctx.get_hparam("learning_rate")
if ctx.get_data_config().get("use_tensor_lr", False):
lr_schedule = common.PiecewiseConstantDecayWithWarmup(
batch_size=ctx.get_per_slot_batch_size(),
epoch_size=imagenet_preprocessing.NUM_IMAGES["train"],
warmup_epochs=common.LR_SCHEDULE[0][1],
boundaries=[p[1] for p in common.LR_SCHEDULE[1:]],
multipliers=[p[0] for p in common.LR_SCHEDULE],
model = keras.Sequential([keras.layers.Dense(10)])
model = self.context.wrap_model(model)
model.compile(
optimizer="adam",
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=[
tf.keras.metrics.Accuracy()
], # ERR: this is the wrong accuracy, should be SparseCategoricalAccuracy
)
return model
def build_training_data_loader(self):
return np.zeros(1), np.zeros(1)
def build_validation_data_loader(self):
return np.zeros(1), np.zeros(1)
if __name__ == "__main__":
experimental.create(
trial_def=RuntimeErrorTrial,
config={
"description": "keras_runtime_error",
"hyperparameters": {"global_batch_size": det.Constant(1)},
"searcher": {"metric": "accuracy"},
"data_layer": {"type": "lfs", "container_storage_path": "/tmp"},
},
mode=experimental.Mode.LOCAL,
context_dir=str(pathlib.Path.cwd()),
)
"metric": "acc",
"max_length": {
"batches": 400,
},
"smaller_is_better": True,
},
"data": {
"data_dir": "/tmp/data",
"task": "MRPC",
"model_name_or_path": "bert-base-uncased",
"output_mode": "classification",
"path_to_mrpc": "",
"download_data": True,
},
"hyperparameters": {
"global_batch_size": det.Constant(value=24),
"model_type": det.Constant(value="bert"),
"learning_rate": det.Constant(value=0.00002),
"lr_scheduler_epoch_freq": det.Constant(value=1),
"adam_epsilon": det.Constant(value=1e-8),
"weight_decay": det.Constant(value=0),
"num_warmup_steps": det.Constant(value=0),
"num_training_steps": det.Constant(value=459),
"max_seq_length": det.Constant(value=128),
},
}
experimental.create(
trial_def=model_def.BertPytorch,
mode=experimental.Mode(args.mode),
context_dir=str(pathlib.Path.cwd()),
default="{}",
)
parser.add_argument("--mode",
dest="mode",
help="Specifies local mode or cluster mode.",
default="cluster")
parser.add_argument(
"--use-fit",
action="store_true",
help="If true, uses model.fit() instead of model.fit_generator()",
)
args = parser.parse_args()
config = {
"hyperparameters": {
"global_batch_size": det.Constant(value=32),
"kernel_size": det.Constant(value=3),
"dropout": det.Double(minval=0.0, maxval=0.5),
"activation": det.Constant(value="relu"),
},
"searcher": {
"name": "single",
"metric": "val_accuracy",
"max_steps": 40
},
}
config.update(json.loads(args.config))
context = keras.init(config,
mode=experimental.Mode(args.mode),
context_dir=str(pathlib.Path.cwd()))
