Commit 241c90a3 authored by Bharath Ramsundar's avatar Bharath Ramsundar
Browse files

Now saving train/test data separately. Allows for eval of model on train and on test separately.

parent 5d54bc74

deep_chem/models/deep.py

+5 −7
Original line number Diff line number Diff line
@@ -10,7 +10,7 @@ from keras.optimizers import SGD 
from deep_chem.utils.preprocess import to_one_hot





def fit_multitask_mlp(per_task_data, task_types, **training_params):

def fit_multitask_mlp(train_data, task_types, **training_params):

  """

  Perform stochastic gradient descent optimization for a keras multitask MLP.

  Returns AUCs, R^2 scores, and RMS values.
@@ -26,13 +26,12 @@ def fit_multitask_mlp(per_task_data, task_types, **training_params): 
  models = {}

  # Follows convention from process_datasets that the data for multitask models

  # is grouped under key "all"

  (_, X_train, y_train, W_train), (test, X_test, y_test, W_test) = (

      per_task_data["all"])

  (_, X_train, y_train, W_train) = train_data["all"]

  models["all"] = train_multitask_model(X_train, y_train, W_train, task_types,

                                **training_params)

  return models



def fit_singletask_mlp(per_task_data, task_types, **training_params):

def fit_singletask_mlp(train_data, task_types, **training_params):

  """

  Perform stochastic gradient descent optimization for a keras MLP.
@@ -46,11 +45,10 @@ def fit_singletask_mlp(per_task_data, task_types, **training_params): 
    Aggregates keyword parameters to pass to train_multitask_model

  """

  models = {}

  for index, target in enumerate(sorted(per_task_data.keys())):

  for index, target in enumerate(sorted(train_data.keys())):

    print "Training model %d" % index

    print "Target %s" % target

    (train_ids, X_train, y_train, W_train), (test, X_test, y_test, W_test) = (

        per_task_data[target])

    (train_ids, X_train, y_train, W_train) = train_data[target]

    print "%d compounds in Train" % len(train_ids)

    print "%d compounds in Test" % len(test)

    models[target] = train_multitask_model(X_train, y_train, W_train,

deep_chem/models/standard.py

+3 −4
Original line number Diff line number Diff line
@@ -13,7 +13,7 @@ from sklearn.linear_model import ElasticNetCV 
from sklearn.linear_model import LassoLarsCV

from sklearn.svm import SVR



def fit_singletask_models(per_task_data, modeltype, task_types):

def fit_singletask_models(train_data, modeltype, task_types):

  """Fits singletask linear regression models to potency.



  Parameters
@@ -34,10 +34,9 @@ def fit_singletask_models(per_task_data, modeltype, task_types): 
    None or "log". Only for regression outputs.

  """

  models = {}

  for index, target in enumerate(sorted(per_task_data.keys())):

  for index, target in enumerate(sorted(train_data.keys())):

    print "Building model %d" % index

    (_, X_train, y_train, W_train), (test, X_test, y_test, W_test) = (

        per_task_data[target])

    (_, X_train, y_train, W_train) = train_data[target]

    if modeltype == "rf_regressor":

      model = RandomForestRegressor(n_estimators=500, n_jobs=-1,

          warm_start=True, max_features="sqrt")

deep_chem/scripts/modeler.py

+19 −15
Original line number Diff line number Diff line
@@ -10,9 +10,6 @@ from deep_chem.utils.featurize import extract_data 
from deep_chem.utils.featurize import generate_targets

from deep_chem.utils.featurize import generate_features

from deep_chem.utils.featurize import generate_vs_utils_features

from deep_chem.models.deep import fit_singletask_mlp

from deep_chem.models.deep import fit_multitask_mlp

from deep_chem.models.deep3d import fit_3D_convolution

from deep_chem.models.standard import fit_singletask_models

from deep_chem.utils.load import get_target_names

from deep_chem.utils.load import process_datasets
@@ -63,7 +60,7 @@ def parse_args(input_args=None): 
  featurize_cmd.set_defaults(func=featurize_input)



  # TRANSFORM FLAGS

  train_test_cmd = subparsers.add_parser("train-test",

  train_test_cmd = subparsers.add_parser("train-test-split",

                      help="Apply standard data transforms to raw features generated by featurize,\n"

                           "then split data into train/test and store data as (X,y) matrices.")

  train_test_cmd.add_argument("--input-transforms", nargs="+", default=[],
@@ -88,8 +85,10 @@ def parse_args(input_args=None): 
  train_test_cmd.add_argument("--mode", default="singletask",

                      choices=["singletask", "multitask"],

                      help="Type of model being built.")

  train_test_cmd.add_argument("--out", type=str, required=1,

                     help="Location to save transformed mode.")

  train_test_cmd.add_argument("--train-out", type=str, required=1,

                     help="Location to save train set.")

  train_test_cmd.add_argument("--test-out", type=str, required=1,

                     help="Location to save test set.")

  train_test_cmd.set_defaults(func=train_test_input)



  # TRAIN FLAGS
@@ -192,12 +191,14 @@ def train_test_input(args): 
  """Saves transformed model."""

  targets = get_target_names(args.paths)

  output_transforms = {target: args.output_transforms for target in targets}

  per_task_data = process_datasets(args.paths,

  train_dict, test_dict= process_datasets(args.paths,

      args.input_transforms, output_transforms, feature_types=args.feature_types, 

      splittype=args.splittype, weight_positives=args.weight_positives,

      mode=args.mode)

  with gzip.open(args.out, "wb") as f:

    pickle.dump(per_task_data, f)

  with gzip.open(args.train_out, "wb") as f:

    pickle.dump(train_dict, f)

  with gzip.open(args.test_out, "wb") as f:

    pickle.dump(test_dict, f)



def fit_model(args):

  """Builds model from featurized data."""
@@ -205,24 +206,27 @@ def fit_model(args): 
  task_types = {target: args.task_type for target in targets}



  with gzip.open(args.saved_data) as f:

    per_task_data = pickle.load(f)

    train_dict = pickle.load(f)



  if args.model == "singletask_deep_network":

    models = fit_singletask_mlp(per_task_data, task_types, n_hidden=args.n_hidden,

    from deep_chem.models.deep import fit_singletask_mlp

    models = fit_singletask_mlp(train_dict, task_types, n_hidden=args.n_hidden,

      learning_rate=args.learning_rate, dropout=args.dropout,

      nb_epoch=args.n_epochs, decay=args.decay, batch_size=args.batch_size,

      validation_split=args.validation_split)

  elif args.model == "multitask_deep_network":

    models = fit_multitask_mlp(per_task_data, task_types,

    from deep_chem.models.deep import fit_multitask_mlp

    models = fit_multitask_mlp(train_dict, task_types,

      n_hidden=args.n_hidden, learning_rate = args.learning_rate,

      dropout = args.dropout, batch_size=args.batch_size,

      nb_epoch=args.n_epochs, decay=args.decay,

      validation_split=args.validation_split)

  elif args.model == "3D_cnn":

    from deep_chem.models.deep3d import fit_3D_convolution

    models = fit_3D_convolution(train_data, test_data, task_types,

        nb_epoch=args.n_epochs, batch_size=args.batch_size)

  else:

    models = fit_singletask_models(per_task_data, args.model, task_types)

    models = fit_singletask_models(train_dict, args.model, task_types)

  if args.model in ["singletask_deep_network", "multitask_deep_network", "3D_cnn"]:

    modeltype = "keras"

  else:
@@ -235,9 +239,9 @@ def eval_trained_model(args): 
  task_types = {target: args.task_type for target in targets}



  with gzip.open(args.saved_data) as f:

    per_task_data = pickle.load(f)

    test_dict = pickle.load(f)



  results, aucs, r2s, rms = compute_model_performance(per_task_data, task_types, model, args.modeltype,

  results, aucs, r2s, rms = compute_model_performance(test_dict, task_types, model, args.modeltype,

    args.compute_aucs, args.compute_r2s, args.compute_rms) 

  if args.csv_out is not None:

    results_to_csv(results, args.csv_out, task_type=args.task_type)

deep_chem/utils/evaluate.py

+3 −3
Original line number Diff line number Diff line
@@ -16,14 +16,14 @@ from sklearn.metrics import r2_score 
from rdkit import Chem

from rdkit.Chem.Descriptors import ExactMolWt



def compute_model_performance(per_task_data, task_types, models, modeltype,

def compute_model_performance(test_data, task_types, models, modeltype,

    aucs=True, r2s=False, rms=False):

  """Computes statistics for model performance on test set."""

  all_results, auc_vals, r2_vals, rms_vals = {}, {}, {}, {}

  for index, target in enumerate(sorted(per_task_data.keys())):

  for index, target in enumerate(sorted(test_data.keys())):

    print "Evaluating model %d" % index

    print "Target %s" % target

    (train_ids, Xtrain, ytrain, wtrain), (test_ids, Xtest, ytest, wtest) = per_task_data[target]

    (test_ids, Xtest, ytest, wtest) = test_data[target]

    model = models[target]

    results = eval_model(test_ids, Xtest, ytest, wtest, model, {target: task_types[target]}, 

                         modeltype=modeltype)

deep_chem/utils/load.py

+6 −7
Original line number Diff line number Diff line
@@ -38,7 +38,7 @@ def process_datasets(paths, input_transforms, output_transforms, 
  """

  dataset = load_and_transform_dataset(paths, input_transforms, output_transforms,

      feature_types=feature_types, weight_positives=weight_positives)

  arrays = {}

  train_dict, test_dict = {}, {}

  if mode == "singletask":

    singletask = multitask_to_singletask(dataset)

    for target in singletask:
@@ -46,17 +46,16 @@ def process_datasets(paths, input_transforms, output_transforms, 
      if len(data) == 0:

        continue

      train, test = split_dataset(dataset, splittype)

      train_data, test_data = to_arrays(train, test)

      arrays[target] = (train_data, test_data)

      train_dict[target], test_dict[target] = to_arrays(train, test)

  elif mode == "multitask":

    train, test = split_dataset(dataset, splittype)

    train_data, test_data = to_arrays(train, test)

    arrays["all"] = (train_data, test_data)

    train_dict["all"], test_dict["all"] = train_data, test_data

  else:

    raise ValueError("Unsupported mode for process_datasets.")

  print "Shape of Xtest"

  print np.shape(arrays['CANVAS-BACE'][0][1])

  return arrays

  print "Shape of Xtrain"

  print np.shape(train_dict['CANVAS-BACE'][1])

  return train_dict, test_dict 





def load_molecules(paths, feature_types=["fingerprints"]):

CRA Git | Maintained and supported by SUSTech CRA and CCSE