Commit bcb91dc9 authored by miaecle's avatar miaecle
Browse files

adding choice of splitting function

parent 59417eba

examples/benchmark.py

+32 −22
Original line number Diff line number Diff line
@@ -51,8 +51,9 @@ from toxcast.toxcast_datasets import load_toxcast 
from sider.sider_datasets import load_sider



def benchmark_loading_datasets(base_dir_o, hyper_parameters, 

                               dataset_name='all', model='tf', reload = True,

                               verbosity='high', out_path='/tmp'):

                               dataset_name='all', model='tf', split=None,

                               reload = True, verbosity='high', 

                               out_path='.'):

  """

  Loading dataset for benchmark test
@@ -71,7 +72,10 @@ def benchmark_loading_datasets(base_dir_o, hyper_parameters, 
      choice of which model to use, should be: rf, tf, tf_robust, logreg,

      graphconv



  out_path : string, optional(default='/tmp')

  model : string,  optional (default=None)

      choice of splitter function, None = using the default splitter



  out_path : string, optional(default='.')

      path of result file

      

  """
@@ -105,6 +109,11 @@ def benchmark_loading_datasets(base_dir_o, hyper_parameters, 
    

    time_start = time.time()

    #loading datasets

    if split is not None:

      print('Splitting function: %s' % split)  

      tasks,datasets,transformers = loading_functions[dname](

          featurizer=featurizer, split=split)

    else:

      tasks,datasets,transformers = loading_functions[dname](

          featurizer=featurizer)

    train_dataset, valid_dataset, test_dataset = datasets
@@ -375,36 +384,37 @@ if __name__ == '__main__': 
  os.makedirs(base_dir_o)

  

  #Datasets and models used in the benchmark test, all=all the datasets

  dataset_name = 'tox21'

  model = 'tf'

  dataset_name = 'all'

  models = ['tf', 'tf_robust', 'logreg', 'graphconv']



  #input hyperparameters

  #tf: dropouts, learning rate, layer_sizes, weight initial stddev,penalty,

  #    batch_size

  hps = {}

  hps = {}

  hps['tf'] = [{'layer_sizes': [500], 'weight_init_stddevs': [0.02], 

                'bias_init_consts': [1.], 'dropouts': [0.5], 'penalty': 0, 

  hps['tf'] = [{'layer_sizes': [1500], 'weight_init_stddevs': [0.02], 

                'bias_init_consts': [1.], 'dropouts': [0.5], 'penalty': 0.1, 

                'penalty_type': 'l2', 'batch_size': 50, 'nb_epoch': 10, 

                'learning_rate': 0.001}]



  hps['tf_robust'] = [{'layer_sizes': [500], 'weight_init_stddevs': [0.02], 

  hps['tf_robust'] = [{'layer_sizes': [1500], 'weight_init_stddevs': [0.02], 

                       'bias_init_consts': [1.], 'dropouts': [0.5], 

                       'bypass_layer_sizes': [100], 

                       'bypass_layer_sizes': [200], 

                       'bypass_weight_init_stddevs': [0.02],

                       'bypass_bias_init_consts': [1.], 

                       'bypass_dropouts': [0.5], 'penalty': 0,

                       'bypass_dropouts': [0.5], 'penalty': 0.1,

                       'penalty_type': 'l2', 'batch_size': 50, 

                       'nb_epoch': 10, 'learning_rate': 0.001}]

                       'nb_epoch': 10, 'learning_rate': 0.0005}]

             

  hps['logreg'] = [{'penalty': 0, 'penalty_type': 'l2', 'batch_size': 50, 

                    'nb_epoch': 10, 'learning_rate': 0.001}]

  hps['logreg'] = [{'penalty': 0.1, 'penalty_type': 'l2', 'batch_size': 50, 

                    'nb_epoch': 10, 'learning_rate': 0.005}]

                

  hps['graphconv'] = [{'batch_size': 50, 'nb_epoch': 10, 

                       'learning_rate': 0.001, 'n_filters': 64, 

                       'n_fully_connected_nodes': 128}]

                       'learning_rate': 0.0005, 'n_filters': 64, 

                       'n_fully_connected_nodes': 128, 'seed': 123}]



  hps['rf'] = [{'n_estimators': 500}]

         

  for model in models:

    benchmark_loading_datasets(base_dir_o, hps, dataset_name=dataset_name,

                             model=model, reload=reload, verbosity='high')
 
                               model=model, split='random', verbosity='high', out_path='.')

examples/muv/muv_datasets.py

+5 −2
Original line number Diff line number Diff line
@@ -10,7 +10,7 @@ import numpy as np 
import shutil

import deepchem as dc



def load_muv(featurizer='ECFP'):

def load_muv(featurizer='ECFP', split='index'):

  """Load MUV datasets. Does not do train/test split"""

  # Load MUV dataset

  print("About to load MUV dataset.")
@@ -42,7 +42,10 @@ def load_muv(featurizer='ECFP'): 
  for transformer in transformers:

    dataset = transformer.transform(dataset)



  splitter = dc.splits.IndexSplitter()

  splitters = {'index': dc.splits.IndexSplitter(),

               'random': dc.splits.RandomSplitter(),

               'scaffold': dc.splits.ScaffoldSplitter()}

  splitter = splitters[split]

  train, valid, test = splitter.train_valid_test_split(

	dataset, compute_feature_statistics=False)

  return MUV_tasks, (train, valid, test), transformers

examples/nci/nci_datasets.py

+6 −2
Original line number Diff line number Diff line
@@ -13,7 +13,8 @@ import numpy as np 
import shutil

import deepchem as dc



def load_nci(featurizer='ECFP', shard_size=1000, num_shards_per_batch=4):

def load_nci(featurizer='ECFP', shard_size=1000, 

             num_shards_per_batch=4, split='random'):



  current_dir = os.path.dirname(os.path.realpath(__file__))
@@ -62,7 +63,10 @@ def load_nci(featurizer='ECFP', shard_size=1000, num_shards_per_batch=4): 
  for transformer in transformers:

    dataset = transformer.transform(dataset)

  

  splitter = dc.splits.RandomSplitter()

  splitters = {'index': dc.splits.IndexSplitter(),

               'random': dc.splits.RandomSplitter(),

               'scaffold': dc.splits.ScaffoldSplitter()}

  splitter = splitters[split]

  print("Performing new split.")

  train, valid, test = splitter.train_valid_test_split(dataset,

	compute_feature_statistics=False)

examples/pcba/pcba_datasets.py

+6 −3
Original line number Diff line number Diff line
@@ -10,7 +10,7 @@ import numpy as np 
import shutil

import deepchem as dc



def load_pcba(featurizer='ECFP'):

def load_pcba(featurizer='ECFP', split='random'):

  """Load PCBA datasets. Does not do train/test split"""

  

  current_dir = os.path.dirname(os.path.realpath(__file__))
@@ -63,7 +63,10 @@ def load_pcba(featurizer='ECFP'): 
  for transformer in transformers:

    dataset = transformer.transform(dataset)

  

  splitter = dc.splits.RandomSplitter()

  splitters = {'index': dc.splits.IndexSplitter(),

               'random': dc.splits.RandomSplitter(),

               'scaffold': dc.splits.ScaffoldSplitter()}

  splitter = splitters[split]

  print("Performing new split.")

  train, valid, test = splitter.train_valid_test_split(

	dataset, compute_feature_statistics=False)

examples/sider/sider_datasets.py

+5 −2
Original line number Diff line number Diff line
@@ -10,7 +10,7 @@ import numpy as np 
import shutil

import deepchem as dc



def load_sider(featurizer='ECFP'):

def load_sider(featurizer='ECFP', split='index'):

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



	  # Load SIDER dataset
@@ -46,7 +46,10 @@ def load_sider(featurizer='ECFP'): 
  for transformer in transformers:

    dataset = transformer.transform(dataset)



  splitter = dc.splits.IndexSplitter()

  splitters = {'index': dc.splits.IndexSplitter(),

               'random': dc.splits.RandomSplitter(),

               'scaffold': dc.splits.ScaffoldSplitter()}

  splitter = splitters[split]

  train, valid, test = splitter.train_valid_test_split(dataset,

      compute_feature_statistics=False)

CRA Git | Maintained and supported by SUSTech CRA and CCSE