def run_imagenet(flags_obj):
"""Run ResNet ImageNet training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
input_fns = []
if flags_obj.use_synthetic_data:
input_fns.append(get_synth_input_fn())
# means using boxes
if flags_obj.model_method == 1:
input_fns.append(input_fn)
elif flags_obj.model_method > 1:
input_fns.append(input_fn)
input_fns.append(box_cond_input_fn)
input_fns.append(box_marg_input_fn)
else:
raise ValueError('invalid input for model method')
resnet_run_loop.resnet_main(
flags_obj,
imagenet_model_fn,
input_fns,
DATASET_NAME,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def main(argv):
parser = resnet_run_loop.ResnetArgParser(
resnet_size_choices=[18, 34, 50, 101, 152, 200])
parser.set_defaults(
train_epochs=90,
version=1
)
flags = parser.parse_args(args=argv[2:])
seed = int(argv[1])
print('Setting random seed = ', seed)
print('special seeding')
mlperf_log.resnet_print(key=mlperf_log.RUN_SET_RANDOM_SEED, value=seed)
random.seed(seed)
tf.set_random_seed(seed)
numpy.random.seed(seed)
mlperf_log.resnet_print(key=mlperf_log.PREPROC_NUM_TRAIN_EXAMPLES,
value=_NUM_IMAGES['train'])
mlperf_log.resnet_print(key=mlperf_log.PREPROC_NUM_EVAL_EXAMPLES,
value=_NUM_IMAGES['validation'])
input_function = flags.use_synthetic_data and get_synth_input_fn() or input_fn
resnet_run_loop.resnet_main(seed,
flags, imagenet_model_fn, input_function,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def run_cifar(flags_obj):
"""Run ResNet CIFAR-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
if flags_obj.image_bytes_as_serving_input:
tf.logging.fatal('--image_bytes_as_serving_input cannot be set to True '
'for CIFAR. This flag is only applicable to ImageNet.')
return
## start a server for a specific task
server = tf.train.Server(cluster,job_name=flags_obj.job_name,task_index=flags_obj.task_index)
# input_function = (flags_obj.use_synthetic_data and get_synth_input_fn(flags_core.get_tf_dtype(flags_obj)) orinput_fn)
if flags_obj.job_name == "ps":
server.join()
elif flags_obj.job_name == "worker":
## Between-graph replication
with tf.device(tf.train.replica_device_setter(worker_device="/job:worker/task:%d" % flags_obj.task_index, cluster=cluster)):
## count the number of updates
global_step = tf.get_variable('global_step',[],initializer = tf.constant_initializer(0),trainable = False)
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn(flags_core.get_tf_dtype(flags_obj)) or input_fn)
# run training
resnet_run_loop.resnet_main(flags_obj, cifar10_model_fn, input_function, DATASET_NAME,shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
def run_cifar(flags_obj):
"""Run ResNet CIFAR-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
if flags_obj.image_bytes_as_serving_input:
tf.logging.fatal(
'--image_bytes_as_serving_input cannot be set to True '
'for CIFAR. This flag is only applicable to ImageNet.')
return
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn(
flags_core.get_tf_dtype(flags_obj)) or input_fn)
#################### My Changes #########################
"""
purpose -- val by steps not by epochs
change -- add the argument train_size=_NUM_IMAGES(['train'])
"""
resnet_run_loop.resnet_main(flags_obj,
cifar10_model_fn,
input_function,
DATASET_NAME,
_NUM_IMAGES['train'],
shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
def main(flags_obj):
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn()
or input_fn)
resnet_run_loop.resnet_main(
flags_obj, imagenet_model_fn, input_function,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def main(argv):
parser = resnet_run_loop.ResnetArgParser(
resnet_size_choices=[18, 34, 50, 101, 152, 200])
flags = parser.parse_args(args=argv[1:])
input_function = flags.use_synthetic_data and get_synth_input_fn(
) or input_fn
resnet_run_loop.resnet_main(flags, imagenet_model_fn, input_function)
def main(flags_obj):
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn()
or input_fn)
resnet_run_loop.resnet_main(flags_obj,
cifar10_model_fn,
input_function,
shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
def run_cifar(flags_obj):
"""Run ResNet CIFAR-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn()
or input_fn)
resnet_run_loop.resnet_main(flags_obj,
cifar10_model_fn,
input_function,
DATASET_NAME,
shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
# After training, compute training loss&validation accuracy, and return them to the Monitor
# First get event file
train_args = []
eval_args = []
train_file = ''
eval_file = ''
train_dir = flags_obj.md
for r, d, f in os.walk(train_dir):
for file in f:
if 'tfevents' in file:
if 'eval' in r:
eval_file = r + '/' + file
else:
train_file = r + '/' + file
if flags_obj.status == 'init':
# collect and plot, then connect server
# show model_directory files and find events file
train_args = analyzer.training_args(train_file, int(flags_obj.bs))
#TODO uptrend define to fill abnormal
abnormal = 0
'''
for r, d, f in os.walk(eval_dir):
for file in f:
if 'tfevents' in file:
# This is the training events file
eval_args = analyzer.validation_args(file)
break
'''
if len(train_args) > 0: # and len(eval_args) > 0:
# connect server and return the data
sendback(flags_obj, train_args, eval_args, abnormal)
elif flags_obj.status == 'train':
# just save the data/evaluate
# TODO evaluate model performance
# train_args = analyzer.training_args(train_file, int(flags_obj.bs))
train_args = []
abnormal = 0
sendback(flags_obj, train_args, eval_args, abnormal)
pass
def run_cifar(flags_obj):
"""Run ResNet CIFAR-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn()
or input_fn)
resnet_run_loop.resnet_main(
flags_obj, cifar10_model_fn, input_function, DATASET_NAME,
shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
def run_imagenet(flags_obj):
"""Run ResNet ImageNet training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn()
or input_fn)
resnet_run_loop.resnet_main(
flags_obj, imagenet_model_fn, input_function, DATASET_NAME,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def run_imagenet(flags_obj):
"""Run ResNet ImageNet training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn()
or input_fn)
resnet_run_loop.resnet_main(
flags_obj, imagenet_model_fn, input_function, DATASET_NAME,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def main(argv):
parser = resnet_run_loop.ResnetArgParser(
resnet_size_choices=[18, 34, 50, 101, 152, 200])
parser.set_defaults(
train_epochs=100
)
flags = parser.parse_args(args=argv[1:])
input_function = flags.use_synthetic_data and get_synth_input_fn() or input_fn
resnet_run_loop.resnet_main(flags, imagenet_model_fn, input_function)
def run_zj(flags_obj):
"""Run ResNet ZJ-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
input_function = input_fn
resnet_run_loop.resnet_main(flags_obj,
zj_model_fn,
input_function,
DATASET_NAME,
shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
def run_cifar(flags_obj):
"""Run ResNet CIFAR-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn()
or input_fn)
##### INPUT
##### Specifically the "shape" that is passed in with the values being at the top of the file
resnet_run_loop.resnet_main(
flags_obj, cifar10_model_fn, input_function, DATASET_NAME,
shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
def run_imagenet(flags_obj):
"""Run ResNet ImageNet training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
input_function = (flags_obj.use_synthetic_data and get_synth_input_fn()
or input_fn)
##### INPUT
##### specifically the shape that is defined here with the constants being up top
resnet_run_loop.resnet_main(
flags_obj, imagenet_model_fn, input_function, DATASET_NAME,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def main(argv):
parser = resnet_run_loop.ResnetArgParser()
# Set defaults that are reasonable for this model.
parser.set_defaults(data_dir='/tmp/cifar10_data',
model_dir='/tmp/cifar10_model',
resnet_size=32,
train_epochs=250,
epochs_between_evals=10,
batch_size=128)
flags = parser.parse_args(args=argv[1:])
input_function = flags.use_synthetic_data and get_synth_input_fn() or input_fn
resnet_run_loop.resnet_main(flags, cifar10_model_fn, input_function)
def main(argv):
parser = resnet_run_loop.ResnetArgParser(
resnet_size_choices=[18, 34, 50, 101, 152, 200])
parser.set_defaults(train_epochs=100)
flags = parser.parse_args(args=argv[1:])
input_function = flags.use_synthetic_data and get_synth_input_fn(
) or input_fn
resnet_run_loop.resnet_main(
flags,
imagenet_model_fn,
input_function,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def run_imagenet(flags_obj):
"""Run ResNet ImageNet training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
Returns:
Dict of results of the run. Contains the keys `eval_results` and
`train_hooks`. `eval_results` contains accuracy (top_1) and
accuracy_top_5. `train_hooks` is a list the instances of hooks used during
training.
"""
input_function = (flags_obj.use_synthetic_data and
get_synth_input_fn(flags_core.get_tf_dtype(flags_obj)) or
input_fn)
if flags_obj.use_dali:
input_function = dali_pipeline.dali_input_fn
result = resnet_run_loop.resnet_main(
flags_obj, imagenet_model_fn, input_function, DATASET_NAME,
shape=[DEFAULT_IMAGE_SIZE, DEFAULT_IMAGE_SIZE, NUM_CHANNELS])
return result
def main(argv):
parser = resnet_run_loop.ResnetArgParser()
# Set defaults that are reasonable for this model.
parser.set_defaults(data_dir='/tmp/cifar10_data',
model_dir='/tmp/cifar10_model',
resnet_size=32,
train_epochs=250,
epochs_between_evals=10,
batch_size=128)
flags = parser.parse_args(args=argv[1:])
input_function = flags.use_synthetic_data and get_synth_input_fn() or input_fn
resnet_run_loop.resnet_main(
flags, cifar10_model_fn, input_function,
shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
def run_cifar(flags_obj):
"""Run ResNet CIFAR-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
if flags_obj.image_bytes_as_serving_input:
tf.logging.fatal('--image_bytes_as_serving_input cannot be set to True '
'for CIFAR. This flag is only applicable to ImageNet.')
return
input_function = (flags_obj.use_synthetic_data and
get_synth_input_fn(flags_core.get_tf_dtype(flags_obj)) or
input_fn)
resnet_run_loop.resnet_main(
flags_obj, cifar10_model_fn, input_function, DATASET_NAME,
shape=[_HEIGHT, _WIDTH, _NUM_CHANNELS])
def main(argv):
parser = resnet_run_loop.ResnetArgParser(
resnet_size_choices=[18, 34, 50, 101, 152, 200])
parser.set_defaults(
train_epochs=100
)
flags = parser.parse_args(args=argv[2:])
seed = int(argv[1])
print('Setting random seed = ', seed)
print('special seeding')
random.seed(seed)
tf.set_random_seed(seed)
numpy.random.seed(seed)
input_function = flags.use_synthetic_data and get_synth_input_fn() or input_fn
resnet_run_loop.resnet_main(seed,
flags, imagenet_model_fn, input_function,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def main(argv):
parser = resnet_run_loop.ResnetArgParser(
resnet_size_choices=[18, 34, 50, 101, 152, 200])
parser.set_defaults(train_epochs=100)
flags = parser.parse_args(args=argv[1:])
#procid = os.environ['SLURM_PROCID']
#procid = os.environ['ALPS_APP_PE']
#flags.model_dir = flags.model_dir + '/' + procid
#flags.benchmark_log_dir = flags.benchmark_log_dir + '/' + procid
#flags.export_dir = flags.export_dir + '/' + procid
input_function = flags.use_synthetic_data and get_synth_input_fn(
) or input_fn
resnet_run_loop.resnet_main(
flags,
imagenet_model_fn,
input_function,
_NUM_IMAGES['train'],
_NUM_IMAGES['validation'],
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def main(argv):
parser = resnet_run_loop.ResnetArgParser(
resnet_size_choices=[18, 34, 50, 101, 152, 200])
parser.set_defaults(train_epochs=100)
flags = parser.parse_args(args=argv[2:])
seed = int(argv[1])
print('Setting random seed = ', seed)
print('special seeding')
random.seed(seed)
tf.set_random_seed(seed)
numpy.random.seed(seed)
input_function = flags.use_synthetic_data and get_synth_input_fn(
) or input_fn
resnet_run_loop.resnet_main(
seed,
flags,
imagenet_model_fn,
input_function,
shape=[_DEFAULT_IMAGE_SIZE, _DEFAULT_IMAGE_SIZE, _NUM_CHANNELS])
def run_resnet(flags_obj):
"""Run ResNet CIFAR-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
Returns:
Dictionary of results. Including final accuracy.
"""
if flags_obj.image_bytes_as_serving_input:
tf.compat.v1.logging.fatal(
'--image_bytes_as_serving_input cannot be set to True for CIFAR. '
'This flag is only applicable to ImageNet.')
return
result = resnet_run_loop.resnet_main(flags_obj,
resnet_model_fn,
input_fn,
DATASET_NAME,
shape=[HEIGHT, WIDTH, NUM_CHANNELS])
return result
def run_cifar(flags_obj):
"""Run ResNet CIFAR-10 training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
Returns:
Dictionary of results. Including final accuracy.
"""
if flags_obj.image_bytes_as_serving_input:
tf.logging.fatal('--image_bytes_as_serving_input cannot be set to True '
'for CIFAR. This flag is only applicable to ImageNet.')
return
input_function = (flags_obj.use_synthetic_data and
get_synth_input_fn(flags_core.get_tf_dtype(flags_obj)) or
input_fn)
result = resnet_run_loop.resnet_main(
flags_obj, cifar10_model_fn, input_function, DATASET_NAME,
shape=[HEIGHT, WIDTH, NUM_CHANNELS])
return result
def main(unused_argv):
input_function = FLAGS.use_synthetic_data and get_synth_input_fn(
) or input_fn
resnet_run_loop.resnet_main(FLAGS, imagenet_model_fn, input_function)
