Update gdb7 benchmark to use FitTransformRegressor

  gdb7_tasks = ["u0_atom"]

  if featurizer is None:

    featurizer = dc.feat.CoulombMatrixEig(23)

  else:

    raise ValueError('Only support Coulomb Matrix featurizer')

  loader = dc.data.SDFLoader(tasks=gdb7_tasks, smiles_field="smiles", 

                             mol_field="mol", featurizer=featurizer)

  dataset = loader.featurize(dataset_file)

"""

Script that trains Tensorflow singletask models on GDB7 dataset.

"""

from __future__ import print_function

from __future__ import division

from __future__ import unicode_literals

import os

import deepchem as dc

import numpy as np

import shutil

from sklearn.kernel_ridge import KernelRidge

np.random.seed(123)

base_dir = "/tmp/gdb7_sklearn"

data_dir = os.path.join(base_dir, "dataset")

model_dir = os.path.join(base_dir, "model")

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

test_dir = os.path.join(base_dir, "test")

if os.path.exists(base_dir):

  shutil.rmtree(base_dir)

os.makedirs(base_dir)

max_num_atoms = 23

featurizers = dc.feat.CoulombMatrix(max_num_atoms)

input_file = "gdb7.sdf"

tasks = ["u0_atom"]

smiles_field = "smiles"

mol_field = "mol"

featurizer = dc.data.SDFLoader(tasks, smiles_field=smiles_field, mol_field=mol_field, featurizer=featurizers)

dataset = featurizer.featurize(input_file, data_dir)

random_splitter = dc.splits.RandomSplitter()

train_dataset, test_dataset = random_splitter.train_test_split(dataset, train_dir, test_dir)

transformers = [dc.trans.NormalizationTransformer(transform_y=True, dataset=train_dataset)]

fit_transformers = [dc.trans.CoulombRandomizationFitTransformer(), dc.trans.NormalizationFitTransformer()]

for transformer in transformers:

    train_dataset = transformer.transform(train_dataset)

for transformer in transformers:

    test_dataset = transformer.transform(test_dataset)

regression_metric = dc.metrics.Metric(dc.metrics.mean_absolute_error, mode="regression")

model = dc.models.tensorflow_models.fcnet.TensorflowMultiTaskFitTransformRegressor(n_tasks=len(tasks), n_features=23, logdir=model_dir,

                                    learning_rate=.001, momentum=.8, batch_size=512,

                                    weight_init_stddevs=[1/np.sqrt(2000),1/np.sqrt(800),1/np.sqrt(800),1/np.sqrt(1000)],

                                    bias_init_consts=[0.,0.,0.,0.], layer_sizes=[2000,800,800,1000], 

                                    dropouts=[0.1,0.1,0.1,0.1], fit_transformers=fit_transformers)

# Fit trained model

model.fit(train_dataset, nb_epoch=10)

model.save()

train_evaluator = dc.utils.evaluate.Evaluator(model, train_dataset, transformers)

train_scores = train_evaluator.compute_model_performance([regression_metric])

print("Train scores [kcal/mol]")

print(train_scores)

test_evaluator = dc.utils.evaluate.Evaluator(model, test_dataset, transformers)

test_scores = test_evaluator.compute_model_performance([regression_metric])

print("Validation scores [kcal/mol]")

print(test_scores)

np.random.seed(123)

gdb7_tasks, datasets, transformers = load_gdb7()

gdb7_tasks, datasets, transformers = load_gdb7(featurizer=dc.feat.CoulombMatrix(23))

train_dataset, valid_dataset, test_dataset = datasets

fit_transformers = [dc.trans.CoulombRandomizationFitTransformer(), dc.trans.NormalizationFitTransformer()]

regression_metric = dc.metrics.Metric(dc.metrics.mean_absolute_error, 

                                      mode="regression")

model = dc.models.TensorflowMultiTaskRegressor(

model = dc.models.TensorflowMultiTaskFitTransformRegressor(

    n_tasks=len(gdb7_tasks), n_features=23,

    learning_rate=.0002, momentum=.8, batch_size=512,

    weight_init_stddevs=[1/np.sqrt(2000),1/np.sqrt(800),1/np.sqrt(800),1/np.sqrt(1000)],

    bias_init_consts=[0.,0.,0.,0.], layer_sizes=[2000,800,800,1000], 

    dropouts=[0.1,0.1,0.1,0.1], seed=123)

    dropouts=[0.1,0.1,0.1,0.1], fit_transformers=fit_transformers, seed=123)

# Fit trained model

model.fit(train_dataset, nb_epoch=50)