Commit d937f57b authored by Bharath Ramsundar's avatar Bharath Ramsundar
Browse files

Cleaned up MUV simple example

parent 1fa8044a

deepchem/datasets/__init__.py

+0 −23
Original line number Diff line number Diff line
@@ -196,8 +196,6 @@ class Dataset(object): 
        y_batch = y[indices]

        w_batch = w[indices]

        ids_batch = ids[indices]

        #(X_batch, y_batch, w_batch, ids_batch) = self._pad_batch(

        #    X_batch, y_batch, w_batch, ids_batch, shard_batch_size)

        yield (X_batch, y_batch, w_batch, ids_batch)



  @staticmethod
@@ -249,27 +247,6 @@ class Dataset(object): 
      ws.append(np.array(w_b))

    return np.vstack(ws)



  #def _pad_batch(self, X_b, y_b, w_b, ids_b, batch_size):

  #  """Fix batch to have exactly batch_size elements.

 

  #  Due to rounding issues, some batches will not have exactly batch_size

  #  elements. Handle these batches by zero padding all arrays.

  #  """

  #  n, feature_shape = np.shape(X_b)[0], np.shape(X_b)[1:]

  #  _, num_tasks = np.shape(y_b)

  #  if n == batch_size:

  #    return (X_b, y_b, w_b, ids_b)

  #  else:

  #    X_batch = np.zeros((batch_size,) + feature_shape)

  #    y_batch = np.zeros((batch_size, num_tasks))

  #    w_batch = np.zeros((batch_size, num_tasks))

  #    ids_batch = np.zeros((batch_size,), dtype=object)

  #    X_batch[:n] = X_b

  #    y_batch[:n] = y_b

  #    w_batch[:n] = w_b

  #    ids_batch[:n] = ids_b

  #  return X_batch, y_batch, w_batch, ids_batch



  def __len__(self):

    """

    Finds number of elements in dataset.

examples/muv_keras_models.py

+2 −2
Original line number Diff line number Diff line
@@ -50,7 +50,7 @@ valid_dir = os.path.join(base_dir, "valid_dataset") 
test_dir = os.path.join(base_dir, "test_dataset")

model_dir = os.path.join(base_dir, "model")



# Remove existing model directory since TF doesn't overwrite by default...

# Remove existing model directory since keras doesn't overwrite by default...

if os.path.exists(model_dir):

  shutil.rmtree(model_dir)

os.makedirs(model_dir)
@@ -153,7 +153,7 @@ for transformer in transformers: 
for transformer in transformers:

    transformer.transform(test_dataset)



# Fit tensorflow models

# Fit keras models

MUV_task_types = {task: "classification" for task in MUV_tasks}

classification_metric = Metric(metrics.roc_auc_score, np.mean,

                               verbosity=verbosity,

examples/muv_sklearn_models.py

+14 −0
Original line number Diff line number Diff line
@@ -7,6 +7,7 @@ from __future__ import unicode_literals 


import os

import numpy as np

import shutil

from sklearn.linear_model import LogisticRegression

from sklearn.ensemble import RandomForestClassifier

from deepchem.utils.save import load_from_disk
@@ -48,6 +49,10 @@ valid_dir = os.path.join(base_dir, "valid_dataset") 
test_dir = os.path.join(base_dir, "test_dataset")

model_dir = os.path.join(base_dir, "model")



if os.path.exists(model_dir):

  shutil.rmtree(model_dir)

os.makedirs(model_dir)



# Load MUV dataset

print("About to load MUV dataset.")

dataset_file = os.path.join(
@@ -105,14 +110,20 @@ full_dataset = Dataset(data_dir=full_dir, samples=featurized_samples, 
print("full_dataset.get_task_names()")

print(full_dataset.get_task_names())

y = full_dataset.get_labels()

#if os.path.exists(train_dir):

#  shutil.rmtree(train_dir)

train_dataset = Dataset(data_dir=train_dir, samples=train_samples, 

                        featurizers=featurizers, tasks=MUV_tasks,

                        verbosity=verbosity, reload=reload)

y_train  = train_dataset.get_labels()

#if os.path.exists(valid_dir):

#  shutil.rmtree(valid_dir)

valid_dataset = Dataset(data_dir=valid_dir, samples=valid_samples, 

                        featurizers=featurizers, tasks=MUV_tasks,

                        verbosity=verbosity, reload=reload)

y_valid = valid_dataset.get_labels()

#if os.path.exists(test_dir):

#  shutil.rmtree(test_dir)

test_dataset = Dataset(data_dir=test_dir, samples=test_samples, 

                       featurizers=featurizers, tasks=MUV_tasks,

                       verbosity=verbosity, reload=reload)
@@ -161,6 +172,9 @@ params_dict = { 
def model_builder(tasks, task_types, model_params, model_dir, verbosity=None):

  return SklearnModel(tasks, task_types, model_params, model_dir,

                      model_instance=LogisticRegression(class_weight="balanced"),

                      #model_instance=RandomForestClassifier(

                      #    class_weight="balanced",

                      #    n_estimators=500),

                      verbosity=verbosity)

model = SingletaskToMultitask(MUV_tasks, MUV_task_types, params_dict, model_dir,

                              model_builder, verbosity=verbosity)

examples/muv_tf_models.py

+76 −71
Original line number Diff line number Diff line
@@ -86,22 +86,13 @@ featurized_samples = featurizer.featurize( 


assert len(featurized_samples) == len(dataset)

# Train/Valid/Test Split dataset

#print("About to perform train/valid/test split.")

#splitter = RandomSplitter(verbosity=verbosity)

#frac_train, frac_valid, frac_test = .8, .1, .1

#train_samples, valid_samples, test_samples = \

#    splitter.train_valid_test_split(

#        featurized_samples, train_dir, valid_dir, test_dir,

#        log_every_n=1000, reload=reload)

#

#len_train_samples, len_valid_samples, len_test_samples = \

#  len(train_samples), len(valid_samples), len(test_samples)

#assert relative_difference(

#    len(train_samples), frac_train * len(featurized_samples)) < 1e-3

#assert relative_difference(

#    len(valid_samples), frac_valid * len(featurized_samples)) < 1e-3

#assert relative_difference(

#    len(test_samples), frac_test * len(featurized_samples)) < 1e-3

print("About to perform train/valid/test split.")

splitter = RandomSplitter(verbosity=verbosity)

frac_train, frac_valid, frac_test = .8, .1, .1

train_samples, valid_samples, test_samples = \

    splitter.train_valid_test_split(

        featurized_samples, train_dir, valid_dir, test_dir,

        log_every_n=1000, reload=reload)



# Generate datasets

print("About to create datasets")
@@ -112,18 +103,45 @@ full_dataset = Dataset(data_dir=full_dir, samples=featurized_samples, 
                        verbosity=verbosity, reload=reload)

print("len(full_dataset)")

print(len(full_dataset))

# Do train/valid split.

#num_train = 2048

#num_train = 4096

#num_train = 8192

num_train = 12800 

#num_train = 13000

#num_train = 16384 ## BROKEN

num_valid = 1024

X, y, w, ids = full_dataset.to_numpy()



w_flat = w.flatten()

#y_flat = y.flatten()

y = full_dataset.get_labels()

train_dataset = Dataset(data_dir=train_dir, samples=train_samples, 

                        featurizers=featurizers, tasks=MUV_tasks,

                        verbosity=verbosity, reload=reload)

y_train  = train_dataset.get_labels()

valid_dataset = Dataset(data_dir=valid_dir, samples=valid_samples, 

                        featurizers=featurizers, tasks=MUV_tasks,

                        verbosity=verbosity, reload=reload)

y_valid = valid_dataset.get_labels()

test_dataset = Dataset(data_dir=test_dir, samples=test_samples, 

                       featurizers=featurizers, tasks=MUV_tasks,

                       verbosity=verbosity, reload=reload)

y_test = test_dataset.get_labels()

len_train_dataset, len_valid_dataset, len_test_dataset = \

  len(train_dataset), len(valid_dataset), len(test_dataset)



input_transformers = []

output_transformers = []

weight_transformers = [BalancingTransformer(transform_w=True, dataset=train_dataset)]

transformers = input_transformers + output_transformers + weight_transformers

for transformer in transformers:

    transformer.transform(train_dataset)

for transformer in transformers:

    transformer.transform(valid_dataset)

for transformer in transformers:

    transformer.transform(test_dataset)



## Do train/valid split.

##num_train = 2048

##num_train = 4096

##num_train = 8192

#num_train = 12800 

##num_train = 13000

##num_train = 16384 ## BROKEN

#num_valid = 1024

#X, y, w, ids = full_dataset.to_numpy()

#

#w_flat = w.flatten()

##y_flat = y.flatten()





####### DEBUG
@@ -133,30 +151,30 @@ w_flat = w.flatten() 
#    print(weight_nonzero)

#    w[y_flat != 0] = weight_nonzero

####### DEBUG



X, y, w, ids = X[w_flat != 0], y[w_flat != 0], w[w_flat != 0], ids[w_flat != 0]

print("Shape after removing zeros")

print("X.shape")

print(X.shape)

X_train, X_valid = X[:num_train], X[num_train:num_train+num_valid]

y_train, y_valid = y[:num_train], y[num_train:num_train+num_valid]

w_train, w_valid = w[:num_train], w[num_train:num_train+num_valid]

ids_train, ids_valid = ids[:num_train], ids[num_train:num_train+num_valid]





if os.path.exists(train_dir):

  shutil.rmtree(train_dir)

train_dataset = Dataset.from_numpy(train_dir, MUV_tasks, X_train, y_train,

                                   w_train, ids_train)

print("len(train_dataset)")

print(len(train_dataset))

if os.path.exists(valid_dir):

  shutil.rmtree(valid_dir)

valid_dataset = Dataset.from_numpy(valid_dir, MUV_tasks, X_valid, y_valid,

                                   w_valid, ids_valid)

print("len(valid_dataset)")

print(len(valid_dataset))



#

#X, y, w, ids = X[w_flat != 0], y[w_flat != 0], w[w_flat != 0], ids[w_flat != 0]

#print("Shape after removing zeros")

#print("X.shape")

#print(X.shape)

#X_train, X_valid = X[:num_train], X[num_train:num_train+num_valid]

#y_train, y_valid = y[:num_train], y[num_train:num_train+num_valid]

#w_train, w_valid = w[:num_train], w[num_train:num_train+num_valid]

#ids_train, ids_valid = ids[:num_train], ids[num_train:num_train+num_valid]

#

#

#if os.path.exists(train_dir):

#  shutil.rmtree(train_dir)

#train_dataset = Dataset.from_numpy(train_dir, MUV_tasks, X_train, y_train,

#                                   w_train, ids_train)

#print("len(train_dataset)")

#print(len(train_dataset))

#if os.path.exists(valid_dir):

#  shutil.rmtree(valid_dir)

#valid_dataset = Dataset.from_numpy(valid_dir, MUV_tasks, X_valid, y_valid,

#                                   w_valid, ids_valid)

#print("len(valid_dataset)")

#print(len(valid_dataset))

#

#y = full_dataset.get_labels()

#train_dataset = Dataset(data_dir=train_dir, samples=train_samples, 

#                        featurizers=featurizers, tasks=MUV_tasks,
@@ -195,19 +213,19 @@ classification_metric = Metric(metrics.roc_auc_score, np.mean, 
                               verbosity=verbosity,

                               mode="classification")

params_dict = { 

    "batch_size": 128,

    #"batch_size": 5120,

    "nb_epoch": 50,

    "batch_size": 64,

    "nb_epoch": 10,

    "data_shape": train_dataset.get_data_shape(),

    "layer_sizes": [1000],

    "weight_init_stddevs": [1.],

    "weight_init_stddevs": [.1],

    "bias_init_consts": [1.],

    "dropouts": [.25],

    "num_classification_tasks": len(MUV_tasks),

    "num_classes": 2,

    "penalty": .0,

    "optimizer": "adam",

    "learning_rate": .0003,

    "optimizer": "momentum",

    "learning_rate": .001,

    "momentum": .9,

}   



model = TensorflowModel(MUV_tasks, MUV_task_types, params_dict, model_dir,
@@ -219,19 +237,6 @@ model = TensorflowModel(MUV_tasks, MUV_task_types, params_dict, model_dir, 
model.fit(train_dataset)

model.save()



############# DEBUG

#import sklearn

#y_train_pred_proba = np.squeeze(model.predict_proba(train_dataset))

#y_train = train_dataset.get_labels()

#w_train = train_dataset.get_weights()

#print("y_train.shape, y_train_pred_proba.shape")

#print(y_train.shape, y_train_pred_proba.shape)

#print("sklearn.metrics.roc_auc_score(to_one_hot(y_train), y_train_pred_proba)")

#print(sklearn.metrics.roc_auc_score(to_one_hot(y_train), y_train_pred_proba))

#print("sklearn.metrics.roc_auc_score(to_one_hot(y_train), y_train_pred_proba, sample_weight=w_train)")

#print(sklearn.metrics.roc_auc_score(to_one_hot(y_train), y_train_pred_proba, sample_weight=w_train))

############# DEBUG



train_evaluator = Evaluator(model, train_dataset, transformers, verbosity=verbosity)

train_scores = train_evaluator.compute_model_performance([classification_metric])

CRA Git | Maintained and supported by SUSTech CRA and CCSE