Commit c6c9993a authored by Bharath Ramsundar's avatar Bharath Ramsundar
Browse files

Updated graph conv models

parent ff2d2ab0

examples/chembl/chembl_graph_conv.py

+46 −49
Original line number Diff line number Diff line
@@ -6,20 +6,13 @@ from __future__ import division 
from __future__ import unicode_literals



import numpy as np

np.random.seed(123)

import tensorflow as tf

tf.set_random_seed(123)

import deepchem as dc

from keras import backend as K

from chembl_datasets import load_chembl



# Only for debug!

np.random.seed(123)



g = tf.Graph()

sess = tf.Session(graph=g)

K.set_session(sess)



with g.as_default():

  tf.set_random_seed(123)

# Load ChEMBL dataset

chembl_tasks, datasets, transformers = load_chembl(shard_size=2000,

  featurizer="GraphConv", set="5thresh", split="random")

train_dataset, valid_dataset, test_dataset = datasets
@@ -33,23 +26,27 @@ with g.as_default(): 
# Batch size of models

batch_size = 128

graph_model = dc.nn.SequentialGraph(n_feat)

  graph_model.add(dc.nn.GraphConv(128, activation='relu'))

graph_model.add(dc.nn.GraphConv(128, n_feat, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphPool())

  graph_model.add(dc.nn.GraphConv(128, activation='relu'))

graph_model.add(dc.nn.GraphConv(128, 64, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphPool())

# Gather Projection

  graph_model.add(dc.nn.Dense(256, activation='relu'))

graph_model.add(dc.nn.Dense(256, 128, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphGather(batch_size, activation="tanh"))

  # Dense post-processing layer



  with tf.Session() as sess:

model = dc.models.MultitaskGraphRegressor(

      sess, graph_model, len(chembl_tasks), batch_size=batch_size,

      learning_rate=1e-3, learning_rate_decay_time=1000,

      optimizer_type="adam", beta1=.9, beta2=.999)

  graph_model,

  len(chembl_tasks),

  n_feat,

  batch_size=batch_size,

  learning_rate=1e-3,

  learning_rate_decay_time=1000,

  optimizer_type="adam",

  beta1=.9,

  beta2=.999)



# Fit trained model

model.fit(train_dataset, nb_epoch=20)

examples/clintox/clintox_graph_conv.py

+44 −52
Original line number Diff line number Diff line
@@ -7,22 +7,13 @@ from __future__ import division 
from __future__ import unicode_literals



import numpy as np

np.random.seed(123)

import tensorflow as tf

from keras import backend as K



tf.set_random_seed(123)

import deepchem as dc

from clintox_datasets import load_clintox



# Only for debug!

np.random.seed(123)



g = tf.Graph()

sess = tf.Session(graph=g)

K.set_session(sess)



with g.as_default():

# Load clintox dataset

  n_features = 1024

clintox_tasks, clintox_datasets, transformers = load_clintox(

    featurizer='GraphConv', split='random')

train_dataset, valid_dataset, test_dataset = clintox_datasets
@@ -37,26 +28,27 @@ with g.as_default(): 
# Batch size of models

batch_size = 50

graph_model = dc.nn.SequentialGraph(n_feat)

  graph_model.add(dc.nn.GraphConv(64, activation='relu'))

graph_model.add(dc.nn.GraphConv(64, n_feat, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphPool())

  graph_model.add(dc.nn.GraphConv(64, activation='relu'))

graph_model.add(dc.nn.GraphConv(64, 64, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphPool())

# Gather Projection

  graph_model.add(dc.nn.Dense(128, activation='relu'))

graph_model.add(dc.nn.Dense(128, 64, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphGather(batch_size, activation="tanh"))

  # Dense post-processing layer



  with tf.Session() as sess:

    model = dc.models.MultitaskGraphClassifier(sess, graph_model,

model = dc.models.MultitaskGraphClassifier(

    graph_model,

    len(clintox_tasks),

    n_feat,

    batch_size=batch_size,

    learning_rate=1e-3,

    learning_rate_decay_time=1000,

    optimizer_type="adam",

                                               beta1=.9, beta2=.999)

    beta1=.9,

    beta2=.999)



# Fit trained model

model.fit(train_dataset, nb_epoch=10)

examples/delaney/delaney_graph_conv.py

+50 −53
Original line number Diff line number Diff line
@@ -6,21 +6,13 @@ from __future__ import division 
from __future__ import unicode_literals



import numpy as np

np.random.seed(123)

import tensorflow as tf

tf.set_random_seed(123)

import deepchem as dc

from keras import backend as K

from delaney_datasets import load_delaney



# Only for debug!

np.random.seed(123)



g = tf.Graph()

sess = tf.Session(graph=g)

K.set_session(sess)



with g.as_default():

  # Load Tox21 dataset

  tf.set_random_seed(123)

# Load Delaney dataset

delaney_tasks, delaney_datasets, transformers = load_delaney(featurizer='GraphConv',split='index')

train_dataset, valid_dataset, test_dataset = delaney_datasets
@@ -33,23 +25,28 @@ with g.as_default(): 
# Batch size of models

batch_size = 128

graph_model = dc.nn.SequentialGraph(n_feat)

  graph_model.add(dc.nn.GraphConv(128, activation='relu'))

graph_model.add(dc.nn.GraphConv(128, n_feat, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphPool())

  graph_model.add(dc.nn.GraphConv(128, activation='relu'))

graph_model.add(dc.nn.GraphConv(128, 128, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphPool())

# Gather Projection

  graph_model.add(dc.nn.Dense(256, activation='relu'))

graph_model.add(dc.nn.Dense(256, 128, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphGather(batch_size, activation="tanh"))

# Dense post-processing layer



  with tf.Session() as sess:

model = dc.models.MultitaskGraphRegressor(

      sess, graph_model, len(delaney_tasks), batch_size=batch_size,

      learning_rate=1e-3, learning_rate_decay_time=1000,

      optimizer_type="adam", beta1=.9, beta2=.999)

  graph_model,

  len(delaney_tasks),

  n_feat,

  batch_size=batch_size,

  learning_rate=1e-3,

  learning_rate_decay_time=1000,

  optimizer_type="adam",

  beta1=.9,

  beta2=.999)



# Fit trained model

model.fit(train_dataset, nb_epoch=20)

examples/muv/muv_keras.py

deleted100644 → 0
+0 −43
Original line number Diff line number Diff line 
"""

Script that trains Keras multitask models on MUV dataset.

"""

from __future__ import print_function

from __future__ import division

from __future__ import unicode_literals



import os

import numpy as np

import shutil

import deepchem as dc

from muv_datasets import load_muv



# Set some global variables up top

np.random.seed(123)



# Load MUV data

muv_tasks, muv_datasets, transformers = load_muv()

train_dataset, valid_dataset, test_dataset = muv_datasets 

n_features = 1024 





# Build model

metric = dc.metrics.Metric(

    dc.metrics.roc_auc_score, np.mean, mode="classification")



keras_model = dc.models.MultiTaskDNN(

    len(muv_tasks), n_features, "classification",

    dropout=.25, learning_rate=.001, decay=1e-4)

model = dc.models.KerasModel(keras_model, verbosity="high")



# Fit trained model

model.fit(train_dataset)

model.save()



train_scores = model.evaluate(train_dataset, [metric], transformers)

valid_scores = model.evaluate(valid_dataset, [metric], transformers)



print("Train scores")

print(train_scores)



print("Validation scores")

print(valid_scores)

examples/sampl/sampl_graph_conv.py

+50 −56
Original line number Diff line number Diff line
@@ -6,21 +6,13 @@ from __future__ import division 
from __future__ import unicode_literals



import numpy as np

np.random.seed(123)

import tensorflow as tf

tf.set_random_seed(123)

import deepchem as dc

from keras import backend as K

from sampl_datasets import load_sampl



# Only for debug!

np.random.seed(123)



g = tf.Graph()

sess = tf.Session(graph=g)

K.set_session(sess)



with g.as_default():

# Load Tox21 dataset

  tf.set_random_seed(123)

SAMPL_tasks, SAMPL_datasets, transformers = load_sampl(featurizer='GraphConv')

train_dataset, valid_dataset, test_dataset = SAMPL_datasets
@@ -33,23 +25,28 @@ with g.as_default(): 
# Batch size of models

batch_size = 128

graph_model = dc.nn.SequentialGraph(n_feat)

  graph_model.add(dc.nn.GraphConv(128, activation='relu'))

graph_model.add(dc.nn.GraphConv(128, n_feat, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphPool())

  graph_model.add(dc.nn.GraphConv(128, activation='relu'))

graph_model.add(dc.nn.GraphConv(128, 128, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphPool())

# Gather Projection

  graph_model.add(dc.nn.Dense(256, activation='relu'))

graph_model.add(dc.nn.Dense(256, 128, activation='relu'))

graph_model.add(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

graph_model.add(dc.nn.GraphGather(batch_size, activation="tanh"))

# Dense post-processing layer



  with tf.Session() as sess:

model = dc.models.MultitaskGraphRegressor(

      sess, graph_model, len(SAMPL_tasks), batch_size=batch_size,

      learning_rate=1e-3, learning_rate_decay_time=1000,

      optimizer_type="adam", beta1=.9, beta2=.999)

  graph_model,

  len(SAMPL_tasks),

  n_feat,

  batch_size=batch_size,

  learning_rate=1e-3,

  learning_rate_decay_time=1000,

  optimizer_type="adam",

  beta1=.9,

  beta2=.999)



# Fit trained model

model.fit(train_dataset, nb_epoch=20)
@@ -63,6 +60,3 @@ with g.as_default(): 


print("Validation scores")

print(valid_scores)
 
    pred = model.predict(train_dataset, transformers)

    pred2 = model.predict(valid_dataset, transformers)

    pred3 = model.predict(test_dataset, transformers)

CRA Git | Maintained and supported by SUSTech CRA and CCSE