Commit c5b049a0 authored by Bharath Ramsundar's avatar Bharath Ramsundar
Browse files

Merge pull request #182 from rbharath/multitask_debug 

Correct multitask data loading issue.
parents af059dd0 588901e2

deepchem/datasets/__init__.py

+12 −3
Original line number Diff line number Diff line
@@ -199,7 +199,17 @@ class Dataset(object): 
        yield (X_batch, y_batch, w_batch, ids_batch)



  @staticmethod

  def from_numpy(data_dir, tasks, X, y, w, ids):

  def from_numpy(data_dir, X, y, w=None, ids=None, tasks=None):

    n_samples = len(X)

    # The -1 indicates that y will be reshaped to have length -1

    y = np.reshape(y, (n_samples, -1))

    n_tasks = y.shape[1]

    if ids is None:

      ids = np.arange(n_samples)

    if w is None:

      w = np.ones_like(y)

    if tasks is None:

      tasks = np.arange(n_tasks)

    raw_data = (ids, X, y, w)

    return Dataset(data_dir=data_dir, tasks=tasks, raw_data=raw_data)
@@ -374,8 +384,7 @@ def _df_to_numpy(df, feature_types, tasks): 
  # perform common train/test split across all tasks

  n_samples = df.shape[0]

  n_tasks = len(tasks)

  y = np.array([df[task].values for task in tasks])

  y = np.reshape(y, (n_samples, n_tasks))

  y = np.hstack([np.reshape(np.array(df[task].values), (n_samples, 1)) for task in tasks])

  w = np.ones((n_samples, n_tasks))

  missing = np.zeros_like(y).astype(int)

  tensors = []

deepchem/datasets/tests/test_load.py

0 → 100644
+126 −0
Original line number Diff line number Diff line 
"""

Testing singletask/multitask data loading capabilities.

"""

from __future__ import print_function

from __future__ import division

from __future__ import unicode_literals



__author__ = "Bharath Ramsundar"

__copyright__ = "Copyright 2016, Stanford University"

__license__ = "LGPL"



import os

import shutil

import tempfile

import numpy as np

from deepchem.models.tests import TestAPI

from deepchem.utils.save import load_from_disk

from deepchem.featurizers.fingerprints import CircularFingerprint

from deepchem.featurizers.featurize import DataFeaturizer

from deepchem.datasets import Dataset



## task0: 1,1,0,-,0,-,1,-,-,1



class TestLoad(TestAPI):

  """

  Test singletask/multitask data loading.

  """



  def test_singletask_matches_multitask_load(self):

    """Check that singletask load and multitask load of dataset are same."""

    # Only for debug!

    np.random.seed(123)



    # Set some global variables up top

    reload = True

    verbosity = "high"



    base_dir = tempfile.mkdtemp()



    current_dir = os.path.dirname(os.path.realpath(__file__))

    #Make directories to store the raw and featurized datasets.

    feature_dir = os.path.join(base_dir, "features")

    samples_dir = os.path.join(base_dir, "samples")

    full_dir = os.path.join(base_dir, "full_dataset")

    train_dir = os.path.join(base_dir, "train_dataset")

    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")



    # Load dataset

    print("About to load dataset.")

    dataset_file = os.path.join(

        current_dir, "../../models/tests/multitask_example.csv")

    dataset = load_from_disk(dataset_file)

    print("Columns of dataset: %s" % str(dataset.columns.values))

    print("Number of examples in dataset: %s" % str(dataset.shape[0]))



    # Featurize tox21 dataset

    print("About to featurize dataset.")

    featurizers = [CircularFingerprint(size=1024)]

    all_tasks = ["task%d"%i for i in range(17)] 

    # For debugging purposes

    n_tasks = 17 

    tasks = all_tasks[0:n_tasks]

    valid_scores = {}



    ####### Do multitask load

    if os.path.exists(feature_dir):

      shutil.rmtree(feature_dir)

    featurizer = DataFeaturizer(tasks=tasks,

                                smiles_field="smiles",

                                compound_featurizers=featurizers,

                                verbosity=verbosity)

    featurized_samples = featurizer.featurize(

        dataset_file, feature_dir,

        samples_dir, shard_size=8192,

        reload=reload)

    if os.path.exists(full_dir):

      shutil.rmtree(full_dir)

    full_dataset = Dataset(data_dir=full_dir, samples=featurized_samples, 

                            featurizers=featurizers, tasks=tasks,

                            verbosity=verbosity, reload=reload)



    # Do train/valid split.

    X_multi, y_multi, w_multi, ids_multi = full_dataset.to_numpy()





    ####### Do singletask load

    X_tasks, y_tasks, w_tasks, ids_tasks = [], [], [], []

    for task in tasks:

      print("Processing task %s" % task)

      if os.path.exists(feature_dir):

        shutil.rmtree(feature_dir)

      featurizer = DataFeaturizer(tasks=[task],

                                  smiles_field="smiles",

                                  compound_featurizers=featurizers,

                                  verbosity=verbosity)

      featurized_samples = featurizer.featurize(

          dataset_file, feature_dir,

          samples_dir, shard_size=8192,

          reload=reload)

      if os.path.exists(full_dir):

        shutil.rmtree(full_dir)

      full_dataset = Dataset(data_dir=full_dir, samples=featurized_samples, 

                              featurizers=featurizers, tasks=[task],

                              verbosity=verbosity, reload=reload)



      X_task, y_task, w_task, ids_task = full_dataset.to_numpy()

      X_tasks.append(X_task)

      y_tasks.append(y_task)

      w_tasks.append(w_task)

      ids_tasks.append(ids_task)



    ################## Do comparison

    for ind, task in enumerate(tasks):

      y_multi_task = y_multi[:, ind]

      w_multi_task = w_multi[:, ind]



      #X_task = X_tasks[ind]

      y_task = y_tasks[ind]

      w_task = w_tasks[ind]

      ids_task = ids_tasks[ind]



      np.testing.assert_allclose(y_multi_task.flatten(), y_task.flatten())

      np.testing.assert_allclose(w_multi_task.flatten(), w_task.flatten())

    shutil.rmtree(base_dir)

deepchem/featurizers/featurize.py

+0 −1
Original line number Diff line number Diff line
@@ -350,7 +350,6 @@ class FeaturizedSamples(object): 
    self.compounds_df = compounds_df

    self.num_samples = len(compounds_df)





  def _get_compounds_filename(self):

    """

    Get standard location for file listing compounds in this dataframe.

deepchem/hyperparameters/tests/test_hyperparam_opt.py

+6 −4
Original line number Diff line number Diff line
@@ -94,8 +94,9 @@ class TestHyperparamOptAPI(TestAPI): 
    y_train = np.random.randint(2, size=(n_train, n_tasks))

    w_train = np.ones_like(y_train)

    ids_train = ["C"] * n_train

    train_dataset = Dataset.from_numpy(self.train_dir, tasks,

                                       X_train, y_train, w_train, ids_train)

    train_dataset = Dataset.from_numpy(self.train_dir,

                                       X_train, y_train, w_train, ids_train,

                                       tasks)



    # Define validation dataset

    n_valid = 10
@@ -103,8 +104,9 @@ class TestHyperparamOptAPI(TestAPI): 
    y_valid = np.random.randint(2, size=(n_valid, n_tasks))

    w_valid = np.ones_like(y_valid)

    ids_valid = ["C"] * n_valid

    valid_dataset = Dataset.from_numpy(self.valid_dir, tasks,

                                       X_valid, y_valid, w_valid, ids_valid)

    valid_dataset = Dataset.from_numpy(self.valid_dir,

                                       X_valid, y_valid, w_valid, ids_valid,

                                       tasks)

    params_dict = {

        "batch_size": [32],

        "data_shape": [train_dataset.get_data_shape()],

deepchem/metrics/__init__.py

+9 −3
Original line number Diff line number Diff line
@@ -123,7 +123,7 @@ class Metric(object): 
    assert mode in ["classification", "regression"]

    self.mode = mode



  def compute_metric(self, y_true, y_pred, w, n_classes=2):

  def compute_metric(self, y_true, y_pred, w=None, n_classes=2):

    """Compute a performance metric for each task.



    Args:
@@ -134,12 +134,18 @@ class Metric(object): 
    Returns:

      A numpy array containing metric values for each task.

    """

    assert y_true.shape[0] == y_pred.shape[0] == w.shape[0]

    if len(y_true.shape) > 1:

      n_samples, n_tasks = y_true.shape[0], y_true.shape[1] 

    else:

      n_samples, n_tasks = y_true.shape[0], 1

    if self.mode == "classification":

      y_pred = np.reshape(y_pred, (n_samples, n_tasks, n_classes))

    else:

      y_pred = np.reshape(y_pred, (n_samples, n_tasks))

    y_true = np.reshape(y_true, (n_samples, n_tasks))

    if w is None:

      w = np.ones_like(y_true)

    assert y_true.shape[0] == y_pred.shape[0] == w.shape[0]

    computed_metrics = []

    for task in xrange(n_tasks):

      y_task = y_true[:, task]
@@ -183,8 +189,8 @@ class Metric(object): 
    # If there are no nonzero examples, metric is ill-defined.

    if not y_true.size:

      return np.nan

    y_true = np.reshape(y_true, (n_samples,))



    y_true = np.reshape(y_true, (n_samples,))

    if self.mode == "classification":

      n_classes = y_pred.shape[-1]

      # TODO(rbharath): This has been a major source of bugs. Is there a more

CRA Git | Maintained and supported by SUSTech CRA and CCSE