update robustMT to tg (9065ffbb) · Commits · 钟慕尧 / deepchem

deepchem/models/tensorflow_models/robust_multitask.py

+307 −270

File changed.

Preview size limit exceeded, changes collapsed.

deepchem/models/tests/test_overfit.py

+0 −2

Original line number	Diff line number	Diff line
		@@ -598,7 +598,6 @@ class TestOverfit(test_util.TensorFlowTestCase):

		# Fit trained model
		model.fit(dataset, nb_epoch=25)
		model.save()

		# Eval model on train
		scores = model.evaluate(dataset, [classification_metric])
		@@ -766,7 +765,6 @@ class TestOverfit(test_util.TensorFlowTestCase):

		# Fit trained model
		model.fit(dataset, nb_epoch=25)
		model.save()

		# Eval model on train
		scores = model.evaluate(dataset, [regression_metric])

deepchem/molnet/init.py

+1 −1

Original line number	Diff line number	Diff line
		@@ -33,4 +33,4 @@ from deepchem.molnet.dnasim import motif_density
		from deepchem.molnet.dnasim import simulate_single_motif_detection

		from deepchem.molnet.run_benchmark import run_benchmark
		from deepchem.molnet.run_benchmark_low_data import run_benchmark_low_data
		#from deepchem.molnet.run_benchmark_low_data import run_benchmark_low_data

deepchem/molnet/run_benchmark_models.py

+15 −13

Original line number	Diff line number	Diff line
		@@ -95,18 +95,18 @@ def benchmark_classification(train_dataset,
		learning_rate = hyper_parameters['learning_rate']

		# Building tensorflow MultiTaskDNN model
		model = deepchem.models.TensorflowMultiTaskClassifier(
		model = deepchem.models.MultiTaskClassifier(
		len(tasks),
		n_features,
		layer_sizes=layer_sizes,
		weight_init_stddevs=weight_init_stddevs,
		bias_init_consts=bias_init_consts,
		dropouts=dropouts,
		penalty=penalty,
		penalty_type=penalty_type,
		weight_decay_penalty=penalty,
		weight_decay_penalty_type=penalty_type,
		batch_size=batch_size,
		learning_rate=learning_rate,
		seed=seed)
		random_seed=seed)

		elif model_name == 'tf_robust':
		layer_sizes = hyper_parameters['layer_sizes']
		@@ -137,11 +137,11 @@ def benchmark_classification(train_dataset,
		bypass_weight_init_stddevs=bypass_weight_init_stddevs,
		bypass_bias_init_consts=bypass_bias_init_consts,
		bypass_dropouts=bypass_dropouts,
		penalty=penalty,
		penalty_type=penalty_type,
		weight_decay_penalty=penalty,
		weight_decay_penalty_type=penalty_type,
		batch_size=batch_size,
		learning_rate=learning_rate,
		seed=seed)
		random_seed=seed)

		elif model_name == 'logreg':
		penalty = hyper_parameters['penalty']
		@@ -461,15 +461,15 @@ def benchmark_regression(train_dataset,
		nb_epoch = hyper_parameters['nb_epoch']
		learning_rate = hyper_parameters['learning_rate']

		model = deepchem.models.TensorflowMultiTaskRegressor(
		model = deepchem.models.MultiTaskRegressor(
		len(tasks),
		n_features,
		layer_sizes=layer_sizes,
		weight_init_stddevs=weight_init_stddevs,
		bias_init_consts=bias_init_consts,
		dropouts=dropouts,
		penalty=penalty,
		penalty_type=penalty_type,
		weight_decay_penalty=penalty,
		weight_decay_penalty_type=penalty_type,
		batch_size=batch_size,
		learning_rate=learning_rate,
		seed=seed)
		@@ -486,15 +486,15 @@ def benchmark_regression(train_dataset,
		learning_rate = hyper_parameters['learning_rate']
		fit_transformers = [hyper_parameters['fit_transformers'](train_dataset)]

		model = deepchem.models.TensorflowMultiTaskFitTransformRegressor(
		model = deepchem.models.MultiTaskFitTransformRegressor(
		n_tasks=len(tasks),
		n_features=n_features,
		layer_sizes=layer_sizes,
		weight_init_stddevs=weight_init_stddevs,
		bias_init_consts=bias_init_consts,
		dropouts=dropouts,
		penalty=penalty,
		penalty_type=penalty_type,
		weight_decay_penalty=penalty,
		weight_decay_penalty_type=penalty_type,
		batch_size=batch_size,
		learning_rate=learning_rate,
		fit_transformers=fit_transformers,
		@@ -774,6 +774,7 @@ def benchmark_regression(train_dataset,
		return train_scores, valid_scores, test_scores


		'''
		def low_data_benchmark_classification(train_dataset,
		valid_dataset,
		n_features,
		@@ -883,3 +884,4 @@ def low_data_benchmark_classification(train_dataset,
		valid_dataset, metric, n_pos, n_neg, n_trials=n_eval_trials)

		return valid_scores
		'''

examples/factors/FACTORS_tf_robust.py

+18 −13

Original line number	Diff line number	Diff line
		@@ -37,14 +37,20 @@ metric = dc.metrics.Metric(dc.metrics.pearson_r2_score, task_averager=np.mean)
		all_results = []
		for trial in range(num_trials):
		model = dc.models.RobustMultitaskRegressor(
		len(FACTORS_tasks), train_dataset.get_data_shape()[0],
		layer_sizes=[1000]n_layers, bypass_layer_sizes=[100]n_bypass_layers,
		dropouts=[.25]n_layers, bypass_dropouts=[.25]n_bypass_layers,
		weight_init_stddevs=[.02]n_layers, bias_init_consts=[1.]n_layers,
		len(FACTORS_tasks),
		train_dataset.get_data_shape()[0],
		layer_sizes=[1000] * n_layers,
		bypass_layer_sizes=[100] * n_bypass_layers,
		dropouts=[.25] * n_layers,
		bypass_dropouts=[.25] * n_bypass_layers,
		weight_init_stddevs=[.02] * n_layers,
		bias_init_consts=[1.] * n_layers,
		bypass_weight_init_stddevs=[.02] * n_bypass_layers,
		bypass_bias_init_consts=[1.] * n_bypass_layers,
		learning_rate=.0003, penalty=.0001, penalty_type="l2",
		optimizer="adam", batch_size=100, logdir="FACTORS_tf_bypass")
		learning_rate=.0003,
		weight_decay_penalty=.0001,
		weight_decay_penalty_type="l2",
		batch_size=100)

		print("Fitting Model")
		model.fit(train_dataset, nb_epoch=nb_epoch)
		@@ -57,9 +63,8 @@ for trial in range(num_trials):
		test_score, test_task_scores = model.evaluate(
		test_dataset, [metric], transformers, per_task_metrics=True)

		all_results.append((train_score, train_task_scores,
		valid_score, valid_task_scores,
		test_score, test_task_scores))
		all_results.append((train_score, train_task_scores, valid_score,
		valid_task_scores, test_score, test_task_scores))

		print("Scores for trial %d" % trial)
		print("----------------------------------------------------------------")
		@@ -79,8 +84,8 @@ for trial in range(num_trials):
		print("####################################################################")

		for trial in range(num_trials):
		(train_score, train_task_scores, valid_score, valid_task_scores,
		test_score, test_task_scores) = all_results[trial]
		(train_score, train_task_scores, valid_score, valid_task_scores, test_score,
		test_task_scores) = all_results[trial]

		print("Scores for trial %d" % trial)
		print("----------------------------------------------------------------")

Admin message