def test_rows_are_cleared_is_reflected_in_saved_file(self):
temp_dir = self.get_temp_dir()
metrics_mngr = metrics_manager.ScalarMetricsManager(temp_dir,
prefix='foo')
metrics_mngr.update_metrics(0, _create_dummy_metrics())
metrics_mngr.update_metrics(5, _create_dummy_metrics())
metrics_mngr.update_metrics(10, _create_dummy_metrics())
filename = os.path.join(temp_dir, 'foo.metrics.csv')
with tf.io.gfile.GFile(filename, 'r') as csvfile:
num_lines_before = len(csvfile.readlines())
# The CSV file should have 4 lines, one for the fieldnames, and 3 for each
# call to `update_metrics`.
self.assertEqual(num_lines_before, 4)
metrics_mngr.clear_rounds_after(last_valid_round_num=7)
with tf.io.gfile.GFile(filename, 'r') as csvfile:
num_lines_after = len(csvfile.readlines())
# The CSV file should have 3 lines, one for the fieldnames, and 2 for the
# calls to `update_metrics` with round_nums less <= 7.
self.assertEqual(num_lines_after, 3)
def test_rows_are_cleared_and_last_round_num_is_reset(self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics_mngr.update_metrics(0, _create_dummy_metrics())
metrics_mngr.update_metrics(5, _create_dummy_metrics())
metrics_mngr.update_metrics(10, _create_dummy_metrics())
metrics = metrics_mngr.get_metrics()
self.assertEqual(
3, len(metrics.index),
'There should be 3 rows of metrics (for rounds 0, 5, and 10).')
metrics_mngr.clear_rounds_after(last_valid_round_num=7)
metrics = metrics_mngr.get_metrics()
self.assertEqual(
2, len(metrics.index),
'After clearing all rounds after last_valid_round_num=7, should be 2 '
'rows of metrics (for rounds 0 and 5).')
self.assertEqual(5, metrics['round_num'].iloc[-1],
'Last metrics retained are for round 5.')
# The internal state of the manager knows the last round number is 7, so it
# raises an exception if a user attempts to add new metrics at round 7, ...
with self.assertRaises(ValueError):
metrics_mngr.update_metrics(7, _create_dummy_metrics())
# ... but allows a user to add new metrics at a round number greater than 7.
metrics_mngr.update_metrics(8,
_create_dummy_metrics()) # (No exception.)
def test_fn_writes_metrics(self):
experiment_name = 'test_metrics'
iterative_process = _build_federated_averaging_process()
batch = _batch_fn()
federated_data = [[batch]]
def client_datasets_fn(round_num):
del round_num
return federated_data
def evaluate(model):
keras_model = tff.simulation.models.mnist.create_keras_model(
compile_model=True)
model.assign_weights_to(keras_model)
return {'loss': keras_model.evaluate(batch.x, batch.y)}
root_output_dir = self.get_temp_dir()
training_loop.run(
iterative_process=iterative_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate,
total_rounds=1,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
rounds_per_eval=10,
test_fn=evaluate)
results_dir = os.path.join(root_output_dir, 'results', experiment_name)
scalar_manager = metrics_manager.ScalarMetricsManager(results_dir)
metrics = scalar_manager.get_metrics()
self.assertEqual(2, len(metrics.index))
self.assertIn('eval/loss', metrics.columns)
self.assertIn('test/loss', metrics.columns)
self.assertNotIn('train_eval/loss', metrics.columns)
def test_clear_rounds_after_raises_value_error_if_round_num_is_negative(
self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics_mngr.update_metrics(0, _create_dummy_metrics())
with self.assertRaises(ValueError):
metrics_mngr.clear_rounds_after(last_valid_round_num=-1)
def test_reload_of_csvfile(self):
temp_dir = self.get_temp_dir()
metrics_mngr = metrics_manager.ScalarMetricsManager(temp_dir,
prefix='bar')
metrics_mngr.update_metrics(0, _create_dummy_metrics())
metrics_mngr.update_metrics(5, _create_dummy_metrics())
new_metrics_mngr = metrics_manager.ScalarMetricsManager(temp_dir,
prefix='bar')
fieldnames, metrics = new_metrics_mngr.get_metrics()
self.assertCountEqual(fieldnames, ['round_num', 'a/b', 'a/c'])
self.assertLen(metrics, 2,
'There should be 2 rows (for rounds 0 and 5).')
self.assertEqual(5, metrics[-1]['round_num'],
'Last metrics are for round 5.')
self.assertEqual(set(os.listdir(temp_dir)), set(['bar.metrics.csv']))
def test_reload_of_csvfile(self):
temp_dir = self.get_temp_dir()
metrics_mngr = metrics_manager.ScalarMetricsManager(temp_dir,
prefix='bar')
metrics_mngr.update_metrics(0, _create_dummy_metrics())
metrics_mngr.update_metrics(5, _create_dummy_metrics())
new_metrics_mngr = metrics_manager.ScalarMetricsManager(temp_dir,
prefix='bar')
metrics = new_metrics_mngr.get_metrics()
self.assertEqual(
2, len(metrics.index),
'There should be 2 rows of metrics (for rounds 0 and 5).')
self.assertEqual(5, metrics['round_num'].iloc[-1],
'Last metrics are for round 5.')
self.assertEqual(set(os.listdir(temp_dir)), set(['bar.metrics.csv']))
def test_update_metrics_raises_value_error_if_round_num_is_out_of_order(
self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics_mngr.update_metrics(1, _create_dummy_metrics())
with self.assertRaises(ValueError):
metrics_mngr.update_metrics(0, _create_dummy_metrics())
def test_update_metrics_adds_nan_if_previously_seen_metric_not_provided(
self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics_mngr.update_metrics(0,
_create_dummy_metrics_with_extra_column())
metrics_mngr.update_metrics(1, _create_dummy_metrics())
metrics = metrics_mngr.get_metrics()
self.assertTrue(np.isnan(metrics.at[1, 'a/d']))
def test_update_metrics_adds_empty_str_if_previous_column_not_provided(
self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics_mngr.update_metrics(0,
_create_dummy_metrics_with_extra_column())
metrics_mngr.update_metrics(1, _create_dummy_metrics())
_, metrics = metrics_mngr.get_metrics()
self.assertEqual(metrics[1]['a/d'], '')
def test_clear_rounds_after_raises_runtime_error_if_no_metrics(self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
# Clear is allowed with no metrics if no rounds have yet completed.
metrics_mngr.clear_rounds_after(last_valid_round_num=0)
with self.assertRaises(RuntimeError):
# Raise exception with no metrics if no rounds have yet completed.
metrics_mngr.clear_rounds_after(last_valid_round_num=1)
def test_update_metrics_returns_flat_dict(self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
input_data_dict = _create_dummy_metrics()
appended_data_dict = metrics_mngr.update_metrics(0, input_data_dict)
self.assertEqual({
'a/b': 1.0,
'a/c': 2.0,
'round_num': 0.0
}, appended_data_dict)
def test_metrics_are_appended(self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
_, metrics = metrics_mngr.get_metrics()
self.assertEmpty(metrics)
metrics_mngr.update_metrics(0, _create_dummy_metrics())
_, metrics = metrics_mngr.get_metrics()
self.assertLen(metrics, 1)
metrics_mngr.update_metrics(1, _create_dummy_metrics())
_, metrics = metrics_mngr.get_metrics()
self.assertLen(metrics, 2)
def test_metrics_are_appended(self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics = metrics_mngr.get_metrics()
self.assertTrue(metrics.empty)
metrics_mngr.update_metrics(0, _create_dummy_metrics())
metrics = metrics_mngr.get_metrics()
self.assertEqual(1, len(metrics.index))
metrics_mngr.update_metrics(1, _create_dummy_metrics())
metrics = metrics_mngr.get_metrics()
self.assertEqual(2, len(metrics.index))
def test_run_federated(self, run_federated_fn):
total_rounds = 1
shared_args = collections.OrderedDict(
client_epochs_per_round=1,
client_batch_size=10,
clients_per_round=1,
client_datasets_random_seed=1,
total_rounds=total_rounds,
max_batches_per_client=2,
iterative_process_builder=iterative_process_builder,
rounds_per_checkpoint=10,
rounds_per_eval=10,
rounds_per_train_eval=10,
max_eval_batches=2)
root_output_dir = self.get_temp_dir()
exp_name = 'test_run_federated'
shared_args['root_output_dir'] = root_output_dir
shared_args['experiment_name'] = exp_name
run_federated_fn(**shared_args)
results_dir = os.path.join(root_output_dir, 'results', exp_name)
self.assertTrue(tf.io.gfile.exists(results_dir))
scalar_manager = metrics_manager.ScalarMetricsManager(results_dir)
metrics = scalar_manager.get_metrics()
self.assertIn(
'train/loss',
metrics.columns,
msg=
'The output metrics should have a `train/loss` column if training '
'is successful.')
self.assertIn(
'eval/loss',
metrics.columns,
msg=
'The output metrics should have a `train/loss` column if validation'
' metrics computation is successful.')
self.assertIn(
'test/loss',
metrics.columns,
msg='The output metrics should have a `test/loss` column if test '
'metrics computation is successful.')
self.assertLen(
metrics.index,
total_rounds + 1,
msg='The number of rows in the metrics CSV should be the number of '
'training rounds + 1 (as there is an extra row for validation/test set'
'metrics after training has completed.')
def test_constructor_raises_value_error_if_csvfile_is_invalid(self):
dataframe_missing_round_num = pd.DataFrame.from_dict(
_create_dummy_metrics())
temp_dir = self.get_temp_dir()
# This csvfile is 'invalid' in that it was not originally created by an
# instance of ScalarMetricsManager, and is missing a column for
# round_num.
invalid_csvfile = os.path.join(temp_dir, 'foo.metrics.csv.bz2')
utils_impl.atomic_write_to_csv(dataframe_missing_round_num,
invalid_csvfile)
with self.assertRaises(ValueError):
metrics_manager.ScalarMetricsManager(temp_dir, prefix='foo')
def test_update_metrics_adds_column_if_previously_unseen_metric_added(
self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics_mngr.update_metrics(0, _create_dummy_metrics())
fieldnames, metrics = metrics_mngr.get_metrics()
self.assertCountEqual(fieldnames, ['round_num', 'a/b', 'a/c'])
self.assertNotIn('a/d', metrics[0].keys())
metrics_mngr.update_metrics(1,
_create_dummy_metrics_with_extra_column())
fieldnames, metrics = metrics_mngr.get_metrics()
self.assertCountEqual(fieldnames, ['round_num', 'a/b', 'a/c', 'a/d'])
self.assertEqual(metrics[0]['a/d'], '')
def test_constructor_raises_value_error_if_csvfile_is_invalid(self):
metrics_missing_round_num = _create_dummy_metrics()
temp_dir = self.get_temp_dir()
# This csvfile is 'invalid' in that it was not originally created by an
# instance of ScalarMetricsManager, and is missing a column for
# round_num.
invalid_csvfile = os.path.join(temp_dir, 'foo.metrics.csv')
with tf.io.gfile.GFile(invalid_csvfile, 'w') as csvfile:
writer = csv.DictWriter(
csvfile, fieldnames=metrics_missing_round_num.keys())
writer.writeheader()
writer.writerow(metrics_missing_round_num)
with self.assertRaises(ValueError):
metrics_manager.ScalarMetricsManager(temp_dir, prefix='foo')
def _setup_outputs(root_output_dir,
experiment_name,
hparam_dict,
write_metrics_with_bz2=True,
rounds_per_profile=0):
"""Set up directories for experiment loops, write hyperparameters to disk."""
if not experiment_name:
raise ValueError('experiment_name must be specified.')
create_if_not_exists(root_output_dir)
checkpoint_dir = os.path.join(root_output_dir, 'checkpoints', experiment_name)
create_if_not_exists(checkpoint_dir)
checkpoint_mngr = checkpoint_manager.FileCheckpointManager(checkpoint_dir)
results_dir = os.path.join(root_output_dir, 'results', experiment_name)
create_if_not_exists(results_dir)
metrics_mngr = metrics_manager.ScalarMetricsManager(
results_dir, use_bz2=write_metrics_with_bz2)
summary_logdir = os.path.join(root_output_dir, 'logdir', experiment_name)
create_if_not_exists(summary_logdir)
summary_writer = tf.summary.create_file_writer(summary_logdir)
if hparam_dict:
hparam_dict['metrics_file'] = metrics_mngr.metrics_filename
hparams_file = os.path.join(results_dir, 'hparams.csv')
utils_impl.atomic_write_to_csv(pd.Series(hparam_dict), hparams_file)
with summary_writer.as_default():
hp.hparams({k: v for k, v in hparam_dict.items() if v is not None})
logging.info('Writing...')
logging.info(' checkpoints to: %s', checkpoint_dir)
logging.info(' metrics csv to: %s', metrics_mngr.metrics_filename)
logging.info(' summaries to: %s', summary_logdir)
@contextlib.contextmanager
def profiler(round_num):
if (rounds_per_profile > 0 and round_num % rounds_per_profile == 0):
with tf.profiler.experimental.Profile(summary_logdir):
yield
else:
yield
return checkpoint_mngr, metrics_mngr, summary_writer, profiler
def test_rows_are_cleared_is_reflected_in_saved_file(self):
temp_dir = self.get_temp_dir()
metrics_mngr = metrics_manager.ScalarMetricsManager(temp_dir,
prefix='foo')
metrics_mngr.update_metrics(0, _create_dummy_metrics())
metrics_mngr.update_metrics(5, _create_dummy_metrics())
metrics_mngr.update_metrics(10, _create_dummy_metrics())
file_contents_before = utils_impl.atomic_read_from_csv(
os.path.join(temp_dir, 'foo.metrics.csv.bz2'))
self.assertEqual(3, len(file_contents_before.index))
metrics_mngr.clear_rounds_after(last_valid_round_num=7)
file_contents_after = utils_impl.atomic_read_from_csv(
os.path.join(temp_dir, 'foo.metrics.csv.bz2'))
self.assertEqual(2, len(file_contents_after.index))
def test_csvfile_is_saved(self):
temp_dir = self.get_temp_dir()
metrics_manager.ScalarMetricsManager(temp_dir, prefix='foo')
self.assertEqual(set(os.listdir(temp_dir)),
set(['foo.metrics.csv.bz2']))
def test_column_names(self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics_mngr.update_metrics(0, _create_dummy_metrics())
metrics = metrics_mngr.get_metrics()
self.assertEqual(['a/b', 'a/c', 'round_num'], metrics.columns.tolist())
def test_column_names(self):
metrics_mngr = metrics_manager.ScalarMetricsManager(
self.get_temp_dir())
metrics_mngr.update_metrics(0, _create_dummy_metrics())
fieldnames, _ = metrics_mngr.get_metrics()
self.assertCountEqual(['a/b', 'a/c', 'round_num'], fieldnames)