Unverified Commit 29d01b5e authored by Bharath Ramsundar's avatar Bharath Ramsundar Committed by GitHub
Browse files

Merge pull request #2228 from hsjang001205/DAG_reload 

Fix directed acyclic graph network bug
parents 2b792b4f e94b9db0

deepchem/models/layers.py

+48 −18
Original line number Diff line number Diff line
@@ -2935,22 +2935,37 @@ class DAGLayer(tf.keras.layers.Layer): 
    self.W_list = []

    self.b_list = []

    self.dropouts = []

    init = initializers.get(self.init)

    prev_layer_size = self.n_inputs

    for layer_size in self.layer_sizes:

      self.W_list.append(init([prev_layer_size, layer_size]))

      self.b_list.append(backend.zeros(shape=[

          layer_size,

      ]))

      self.W_list.append(

          self.add_weight(

              name='kernel',

              shape=(prev_layer_size, layer_size),

              initializer=self.init,

              trainable=True))

      self.b_list.append(

          self.add_weight(

              name='bias',

              shape=(layer_size,),

              initializer='zeros',

              trainable=True))

      if self.dropout is not None and self.dropout > 0.0:

        self.dropouts.append(Dropout(rate=self.dropout))

      else:

        self.dropouts.append(None)

      prev_layer_size = layer_size

    self.W_list.append(init([prev_layer_size, self.n_outputs]))

    self.b_list.append(backend.zeros(shape=[

        self.n_outputs,

    ]))

    self.W_list.append(

        self.add_weight(

            name='kernel',

            shape=(prev_layer_size, self.n_outputs),

            initializer=self.init,

            trainable=True))

    self.b_list.append(

        self.add_weight(

            name='bias',

            shape=(self.n_outputs,),

            initializer='zeros',

            trainable=True))

    if self.dropout is not None and self.dropout > 0.0:

      self.dropouts.append(Dropout(rate=self.dropout))

    else:
@@ -3068,22 +3083,37 @@ class DAGGather(tf.keras.layers.Layer): 
    self.W_list = []

    self.b_list = []

    self.dropouts = []

    init = initializers.get(self.init)

    prev_layer_size = self.n_graph_feat

    for layer_size in self.layer_sizes:

      self.W_list.append(init([prev_layer_size, layer_size]))

      self.b_list.append(backend.zeros(shape=[

          layer_size,

      ]))

      self.W_list.append(

          self.add_weight(

              name='kernel',

              shape=(prev_layer_size, layer_size),

              initializer=self.init,

              trainable=True))

      self.b_list.append(

          self.add_weight(

              name='bias',

              shape=(layer_size,),

              initializer='zeros',

              trainable=True))

      if self.dropout is not None and self.dropout > 0.0:

        self.dropouts.append(Dropout(rate=self.dropout))

      else:

        self.dropouts.append(None)

      prev_layer_size = layer_size

    self.W_list.append(init([prev_layer_size, self.n_outputs]))

    self.b_list.append(backend.zeros(shape=[

        self.n_outputs,

    ]))

    self.W_list.append(

        self.add_weight(

            name='kernel',

            shape=(prev_layer_size, self.n_outputs),

            initializer=self.init,

            trainable=True))

    self.b_list.append(

        self.add_weight(

            name='bias',

            shape=(self.n_outputs,),

            initializer='zeros',

            trainable=True))

    if self.dropout is not None and self.dropout > 0.0:

      self.dropouts.append(Dropout(rate=self.dropout))

    else:

deepchem/models/tests/test_reload.py

+68 −66
Original line number Diff line number Diff line
@@ -522,72 +522,74 @@ def test_progressivemultitaskregressor_reload(): 
  assert scores[regression_metric.name] < 0.1





## TODO: THIS IS FAILING!

#def test_DAG_regression_reload():

#  """Test DAG regressor reloads."""

#  np.random.seed(123)

#  tf.random.set_seed(123)

#  n_tasks = 1

#  #current_dir = os.path.dirname(os.path.abspath(__file__))

#

#  # Load mini log-solubility dataset.

#  featurizer = dc.feat.ConvMolFeaturizer()

#  tasks = ["outcome"]

#  mols = ["C", "CO", "CC"]

#  n_samples = len(mols)

#  X = featurizer(mols)

#  y = np.random.rand(n_samples, n_tasks)

#  dataset = dc.data.NumpyDataset(X, y)

#

#  regression_metric = dc.metrics.Metric(

#      dc.metrics.pearson_r2_score, task_averager=np.mean)

#

#  n_feat = 75

#  batch_size = 10

#  transformer = dc.trans.DAGTransformer(max_atoms=50)

#  dataset = transformer.transform(dataset)

#

#  model_dir = tempfile.mkdtemp()

#  model = dc.models.DAGModel(

#      n_tasks,

#      max_atoms=50,

#      n_atom_feat=n_feat,

#      batch_size=batch_size,

#      learning_rate=0.001,

#      use_queue=False,

#      mode="regression",

#      model_dir=model_dir)

#

#  # Fit trained model

#  model.fit(dataset, nb_epoch=1200)

#

#  # Eval model on train

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

#  assert scores[regression_metric.name] > .8

#

#  reloaded_model = dc.models.DAGModel(

#      n_tasks,

#      max_atoms=50,

#      n_atom_feat=n_feat,

#      batch_size=batch_size,

#      learning_rate=0.001,

#      use_queue=False,

#      mode="regression",

#      model_dir=model_dir)

#  reloaded_model.restore()

#

#  # Check predictions match on random sample

#  predmols = ["CCCC", "CCCCCO", "CCCCC"]

#  Xpred = featurizer(predmols)

#  predset = dc.data.NumpyDataset(Xpred)

#  predset = transformer.transform(predset)

#  origpred = model.predict(predset)

#  reloadpred = reloaded_model.predict(predset)

#  assert np.all(origpred == reloadpred)

#

#  # Eval model on train

#  scores = reloaded_model.evaluate(dataset, [classification_metric])

#  assert scores[classification_metric.name] > .9

def test_DAG_regression_reload():

  """Test DAG regressor reloads."""

  np.random.seed(123)

  tf.random.set_seed(123)

  n_tasks = 1

  #current_dir = os.path.dirname(os.path.abspath(__file__))



  # Load mini log-solubility dataset.

  featurizer = dc.feat.ConvMolFeaturizer()

  tasks = ["outcome"]

  mols = ["CC", "CCO", "CC", "CCC", "CCCCO", "CO", "CC", "CCCCC", "CCC", "CCCO"]

  n_samples = len(mols)

  X = featurizer(mols)

  y = np.random.rand(n_samples, n_tasks)

  dataset = dc.data.NumpyDataset(X, y)



  regression_metric = dc.metrics.Metric(

      dc.metrics.pearson_r2_score, task_averager=np.mean)



  n_feat = 75

  batch_size = 10

  transformer = dc.trans.DAGTransformer(max_atoms=50)

  dataset = transformer.transform(dataset)



  model_dir = tempfile.mkdtemp()

  model = dc.models.DAGModel(

      n_tasks,

      max_atoms=50,

      n_atom_feat=n_feat,

      batch_size=batch_size,

      learning_rate=0.001,

      use_queue=False,

      mode="regression",

      model_dir=model_dir)



  # Fit trained model

  model.fit(dataset, nb_epoch=100)



  # Eval model on train

  scores = model.evaluate(dataset, [regression_metric])

  assert scores[regression_metric.name] > .1



  reloaded_model = dc.models.DAGModel(

      n_tasks,

      max_atoms=50,

      n_atom_feat=n_feat,

      batch_size=batch_size,

      learning_rate=0.001,

      use_queue=False,

      mode="regression",

      model_dir=model_dir)



  reloaded_model.restore()



  # Check predictions match on random sample

  predmols = ["CCCC", "CCCCCO", "CCCCC"]

  Xpred = featurizer(predmols)

  predset = dc.data.NumpyDataset(Xpred)

  predset = transformer.transform(predset)

  origpred = model.predict(predset)

  reloadpred = reloaded_model.predict(predset)



  assert np.all(origpred == reloadpred)



  # Eval model on train

  scores = reloaded_model.evaluate(dataset, [regression_metric])

  assert scores[regression_metric.name] > .1





## TODO: THIS IS FAILING!

#def test_weave_classification_reload_alt():

CRA Git | Maintained and supported by SUSTech CRA and CCSE