Commit 0c092b03 authored by Bharath Ramsundar's avatar Bharath Ramsundar Committed by GitHub
Browse files

Merge pull request #277 from rbharath/support_models 

Updated support models for new generator API
parents 87851610 07b5932a

examples/low_data/tox_attn_one_fold.py

+17 −32
Original line number Diff line number Diff line
@@ -14,18 +14,18 @@ from datasets import load_tox21_convmol 
# Number of folds for split 

K = 4 

# Depth of attention module

max_depth = 8

max_depth = 3

# number positive/negative ligands

n_pos = 1 

n_pos = 10

n_neg = 10

# Set batch sizes for network

test_batch_size = 128

support_batch_size = n_pos + n_neg

n_train_trials = 5000

nb_epochs = 1

n_train_trials = 2000

n_eval_trials = 20

n_steps_per_trial = 1

# Sample supports without replacement (all pos/neg should be different)

replace = False

learning_rate = 1e-4

log_every_n_samples = 50

# Number of features on conv-mols

n_feat = 71
@@ -46,29 +46,17 @@ test_dataset = fold_datasets[-1] 
support_model = dc.nn.SequentialSupportGraph(n_feat)



# Add layers

# Need to add batch-norm separately to test/support due to differing

# shapes.

# Adding 1st layer 

# output will be (n_atoms, 64)

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

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add(dc.nn.GraphPool())

# Addding 2nd layer

# output will be (n_atoms, 64)

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

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

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

support_model.add(dc.nn.GraphPool())

# Adding 3rd layer

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

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add(dc.nn.GraphPool())

support_model.add(dc.nn.Dense(128, activation='tanh'))



support_model.add_test(dc.nn.GraphGather(test_batch_size, activation='tanh'))

support_model.add_support(dc.nn.GraphGather(support_batch_size, activation='tanh'))



# Gather into molecules

support_model.add_test(dc.nn.GraphGather(test_batch_size))

support_model.add_support(dc.nn.GraphGather(support_batch_size))

# Apply an attention lstm layer

support_model.join(dc.nn.AttnLSTMEmbedding(

    test_batch_size, support_batch_size, max_depth))
@@ -76,22 +64,19 @@ support_model.join(dc.nn.AttnLSTMEmbedding( 
with tf.Session() as sess:

  model = dc.models.SupportGraphClassifier(

    sess, support_model, test_batch_size=test_batch_size,

    support_batch_size=support_batch_size,

    learning_rate=1e-3, verbosity="high")

    support_batch_size=support_batch_size, learning_rate=learning_rate,

    verbosity="high")



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

  print("FIT")

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

  model.fit(train_dataset, n_trials=n_train_trials,

            n_steps_per_trial=n_steps_per_trial, n_pos=n_pos,

            n_neg=n_neg, replace=False)

  model.save()



  model.fit(train_dataset, nb_epochs=nb_epochs, 

            n_episodes_per_epoch=n_train_trials,

            n_pos=n_pos, n_neg=n_neg, log_every_n_samples=log_every_n_samples)

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

  print("EVAL")

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

  scores = model.evaluate(

      test_dataset, metric, n_pos=n_pos, n_neg=n_neg, replace=replace,

      n_trials=n_eval_trials)

      test_dataset, metric, n_pos, n_neg, n_trials=n_eval_trials)

  print("Scores on evaluation dataset")

  print(scores)

examples/low_data/tox_res_one_fold.py

+22 −35
Original line number Diff line number Diff line 
"""

Train low-data attn models on random forests. Test last fold only.

Train low-data res models on Tox21. Test last fold only.

"""

from __future__ import print_function

from __future__ import division
@@ -12,21 +12,20 @@ import tensorflow as tf 
from datasets import load_tox21_convmol



# Number of folds for split 

K = 12 

K = 4 

# Depth of attention module

max_depth = 4

max_depth = 3

# num positive/negative ligands

n_pos = 1 

n_pos = 10

n_neg = 10

# Set batch sizes for network

test_batch_size = 128

support_batch_size = n_pos + n_neg

n_train_trials = 3000 

n_eval_trials = 5

n_steps_per_trial = 1

nb_epochs = 1

n_train_trials = 2000

n_eval_trials = 20

learning_rate = 1e-4

log_every_n_samples = 50

# Sample supports without replacement (all pos/neg should be different)

replace = False

# Number of features on conv-mols

n_feat = 71
@@ -47,49 +46,37 @@ test_dataset = fold_datasets[-1] 
support_model = dc.nn.SequentialSupportGraph(n_feat)



# Add layers

# 1st conv layer + batchnorm

# output will be (n_atoms, 64)

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

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

# 2nd conv layer + batchnorm

# output will be (n_atoms, 64)

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

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

# 3nd conv layer + batchnorm

# output will be (n_atoms, 64)

support_model.add(dc.nn.GraphPool())

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

support_model.add(dc.nn.GraphPool())

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

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))



support_model.add(dc.nn.GraphPool())

support_model.add_test(dc.nn.GraphGather(test_batch_size))

support_model.add_support(dc.nn.GraphGather(support_batch_size))

support_model.add(dc.nn.Dense(128, activation='tanh'))



support_model.add_test(dc.nn.GraphGather(test_batch_size, activation='tanh'))

support_model.add_support(dc.nn.GraphGather(support_batch_size, activation='tanh'))



# Apply a residual lstm layer

support_model.join(dc.nn.ResiLSTMEmbedding(

    test_batch_size, support_batch_size, max_depth))





with tf.Session() as sess:

  model = dc.models.SupportGraphClassifier(

    sess, support_model, test_batch_size=test_batch_size,

    support_batch_size=support_batch_size,

    learning_rate=1e-3, verbosity="high")

    support_batch_size=support_batch_size, learning_rate=learning_rate,

    verbosity="high")



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

  print("FIT")

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

  model.old_fit(test_dataset, n_trials=n_train_trials,

            n_steps_per_trial=n_steps_per_trial, n_pos=n_pos,

            n_neg=n_neg, log_every_n_samples=log_every_n_samples, replace=False)

  model.save()



  model.fit(train_dataset, nb_epochs=nb_epochs,

            n_episodes_per_epoch=n_train_trials,

            n_pos=n_pos, n_neg=n_neg, log_every_n_samples=log_every_n_samples)

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

  print("EVAL")

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

  scores = model.evaluate(

      test_dataset, metric, n_pos=n_pos, n_neg=n_neg, replace=replace,

      n_trials=n_eval_trials)

      test_dataset, metric, n_pos, n_neg, n_trials=n_eval_trials)

  print("Scores on evaluation dataset")

  print(scores)

examples/low_data/tox_rf_one_fold.py

+5 −8
Original line number Diff line number Diff line
@@ -14,12 +14,10 @@ from sklearn.ensemble import RandomForestClassifier 
# 4-fold splits

K = 4

# num positive/negative ligands

n_pos = 1

n_pos = 5

n_neg = 10

# 10 trials on test-set

n_trials = 10

# Sample supports without replacement (all pos/neg should be different)

replace = False

n_trials = 20



tox21_tasks, dataset, transformers = load_tox21_ecfp()
@@ -35,15 +33,14 @@ test_dataset = fold_datasets[-1] 


# Get supports on test-set

support_generator = dc.data.SupportGenerator(

    test_dataset, range(len(test_dataset.get_task_names())), n_pos, n_neg,

    n_trials, replace)

    test_dataset, n_pos, n_neg, n_trials)



# Compute accuracies

task_scores = {task: [] for task in range(len(test_dataset.get_task_names()))}

for (task, support) in support_generator:

  # Train model on support

  sklearn_model = RandomForestClassifier(

      class_weight="balanced", n_estimators=50)

      class_weight="balanced", n_estimators=100)

  model = dc.models.SklearnModel(sklearn_model)

  model.fit(support)
@@ -53,7 +50,7 @@ for (task, support) in support_generator: 
  y_pred = model.predict_proba(task_dataset)

  score = metric.compute_metric(

      task_dataset.y, y_pred, task_dataset.w)

  #print("Score on task %s is %s" % (str(task), str(score)))

  print("Score on task %s is %s" % (str(task), str(score)))

  task_scores[task].append(score)



# Join information for all tasks.

examples/low_data/tox_siamese_one_fold.py

+19 −38
Original line number Diff line number Diff line
@@ -14,16 +14,17 @@ from datasets import load_tox21_convmol 
# Number of folds for split 

K = 4 

# num positive/negative ligands

n_pos = 3

n_pos = 10

n_neg = 10

# Set batch sizes for network

test_batch_size = 100

test_batch_size = 128

support_batch_size = n_pos + n_neg

n_train_trials = 3000

nb_epochs = 1

n_train_trials = 2000

n_eval_trials = 20 

n_steps_per_trial = 1

# Sample supports without replacement (all pos/neg should be different)

replace = False

learning_rate = 1e-4

log_every_n_samples = 50

# Number of features on conv-mols

n_feat = 71
@@ -44,53 +45,33 @@ test_dataset = fold_datasets[-1] 
support_model = dc.nn.SequentialSupportGraph(n_feat)



# Add layers



# Adding 1st layer

# output will be (n_atoms, 64)

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

# Need to add batch-norm to test/support due to differing shapes.

# output will be (n_atoms, 64)

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

# Addding 2nd layer

# output will be (n_atoms, 64)

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

support_model.add(dc.nn.GraphPool())

# Adding 3rd layer

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

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

# Adding 4th layer

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

support_model.add(dc.nn.GraphPool())

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

support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

## Adding 5th layer

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

#support_model.add_support(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

#support_model.add_test(dc.nn.BatchNormalization(epsilon=1e-5, mode=1))

support_model.add(dc.nn.GraphPool())

support_model.add(dc.nn.Dense(128, activation='tanh'))



# Gather atoms into batches

support_model.add_test(dc.nn.GraphGather(test_batch_size))

support_model.add_support(dc.nn.GraphGather(support_batch_size))

support_model.add_test(dc.nn.GraphGather(test_batch_size, activation='tanh'))

support_model.add_support(dc.nn.GraphGather(support_batch_size, activation='tanh'))



with tf.Session() as sess:

  model = dc.models.SupportGraphClassifier(

    sess, support_model, test_batch_size=test_batch_size,

    support_batch_size=support_batch_size, learning_rate=3e-3, verbosity="high")

    support_batch_size=support_batch_size, learning_rate=learning_rate,

    verbosity="high")



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

  print("FIT")

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

  model.fit(train_dataset, n_trials=n_train_trials,

            n_steps_per_trial=n_steps_per_trial, n_pos=n_pos,

            n_neg=n_neg, replace=False)

  model.save()



  model.fit(train_dataset, nb_epochs=nb_epochs,

            n_episodes_per_epoch=n_train_trials,

            n_pos=n_pos, n_neg=n_neg, log_every_n_samples=log_every_n_samples)

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

  print("EVAL")

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

  scores = model.evaluate(

      test_dataset, metric, n_pos=n_pos, n_neg=n_neg, replace=replace,

      n_trials=n_eval_trials)

  print("Scores on held-out dataset")

      test_dataset, metric, n_pos, n_neg, n_trials=n_eval_trials)

  print("Scores on evaluation dataset")

  print(scores)

CRA Git | Maintained and supported by SUSTech CRA and CCSE