def test_json(monkeypatch, ray_start_4_cpus, make_temp_dir, workers_to_log,
detailed, filename):
if detailed:
monkeypatch.setenv(ENABLE_DETAILED_AUTOFILLED_METRICS_ENV, "1")
config = TestConfig()
num_iters = 5
num_workers = 4
if workers_to_log is None:
num_workers_to_log = num_workers
elif isinstance(workers_to_log, int):
num_workers_to_log = 1
else:
num_workers_to_log = len(workers_to_log)
def train_func():
for i in range(num_iters):
train.report(index=i)
return 1
if filename is None:
# if None, use default value
callback = JsonLoggerCallback(workers_to_log=workers_to_log)
else:
callback = JsonLoggerCallback(filename=filename,
workers_to_log=workers_to_log)
trainer = Trainer(config, num_workers=num_workers, logdir=make_temp_dir)
trainer.start()
trainer.run(train_func, callbacks=[callback])
if filename is None:
assert str(
callback.log_path.name) == JsonLoggerCallback._default_filename
else:
assert str(callback.log_path.name) == filename
with open(callback.log_path, "r") as f:
log = json.load(f)
print(log)
assert len(log) == num_iters
assert len(log[0]) == num_workers_to_log
assert all(len(element) == len(log[0]) for element in log)
assert all(
all(worker["index"] == worker[TRAINING_ITERATION] - 1
for worker in element) for element in log)
assert all(
all(
all(key in worker for key in BASIC_AUTOFILLED_KEYS)
for worker in element) for element in log)
if detailed:
assert all(
all(
all(key in worker for key in DETAILED_AUTOFILLED_KEYS)
for worker in element) for element in log)
else:
assert all(
all(not any(key in worker for key in DETAILED_AUTOFILLED_KEYS)
for worker in element) for element in log)
def train_linear(num_workers=2, use_gpu=False, epochs=3):
trainer = Trainer(backend="torch", num_workers=num_workers, use_gpu=use_gpu)
config = {"lr": 1e-2, "hidden_size": 1, "batch_size": 4, "epochs": epochs}
trainer.start()
results = trainer.run(
train_func, config, callbacks=[JsonLoggerCallback(), TBXLoggerCallback()]
)
trainer.shutdown()
print(results)
return results
def train_linear(num_workers=2):
trainer = Trainer(TorchConfig(backend="gloo"), num_workers=num_workers)
config = {"lr": 1e-2, "hidden_size": 1, "batch_size": 4, "epochs": 3}
trainer.start()
results = trainer.run(
train_func,
config,
callbacks=[JsonLoggerCallback(),
TBXLoggerCallback()])
trainer.shutdown()
print(results)
return results
def train_fashion_mnist(num_workers=2, use_gpu=False):
trainer = Trainer(
backend="torch", num_workers=num_workers, use_gpu=use_gpu)
trainer.start()
result = trainer.run(
train_func=train_func,
config={
"lr": 1e-3,
"batch_size": 64,
"epochs": 4
},
callbacks=[JsonLoggerCallback()])
trainer.shutdown()
print(f"Loss results: {result}")
def train_linear(num_workers=2, use_gpu=False):
datasets = get_datasets()
trainer = Trainer("torch", num_workers=num_workers, use_gpu=use_gpu)
config = {"lr": 1e-2, "hidden_size": 1, "batch_size": 4, "epochs": 3}
trainer.start()
results = trainer.run(
train_func,
config,
dataset=datasets,
callbacks=[JsonLoggerCallback(), TBXLoggerCallback()],
)
trainer.shutdown()
print(results)
return results