def test_gcn_reload():
# load datasets
featurizer = MolGraphConvFeaturizer()
tasks, dataset, transformers, metric = get_dataset('classification',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model_dir = tempfile.mkdtemp()
model = GCNModel(mode='classification',
n_tasks=n_tasks,
number_atom_features=30,
model_dir=model_dir,
batch_size=10,
learning_rate=0.0003)
model.fit(dataset, nb_epoch=70)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
reloaded_model = GCNModel(mode='classification',
n_tasks=n_tasks,
number_atom_features=30,
model_dir=model_dir,
batch_size=10,
learning_rate=0.0003)
reloaded_model.restore()
pred_mols = ["CCCC", "CCCCCO", "CCCCC"]
X_pred = featurizer(pred_mols)
random_dataset = dc.data.NumpyDataset(X_pred)
original_pred = model.predict(random_dataset)
reload_pred = reloaded_model.predict(random_dataset)
assert np.all(original_pred == reload_pred)
def test_attentivefp_regression():
# load datasets
featurizer = MolGraphConvFeaturizer(use_edges=True)
tasks, dataset, transformers, metric = get_dataset(
'regression', featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = AttentiveFPModel(mode='regression', n_tasks=n_tasks, batch_size=10)
# overfit test
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean_absolute_error'] < 0.5
# test on a small MoleculeNet dataset
from deepchem.molnet import load_delaney
tasks, all_dataset, transformers = load_delaney(featurizer=featurizer)
train_set, _, _ = all_dataset
model = AttentiveFPModel(
mode='regression',
n_tasks=len(tasks),
num_layers=1,
num_timesteps=1,
graph_feat_size=2)
model.fit(train_set, nb_epoch=1)
def test_gat_classification():
# load datasets
featurizer = MolGraphConvFeaturizer()
tasks, dataset, transformers, metric = get_dataset('classification',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = GATModel(mode='classification',
n_tasks=n_tasks,
number_atom_features=30,
batch_size=10,
learning_rate=0.001)
# overfit test
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
# test on a small MoleculeNet dataset
from deepchem.molnet import load_bace_classification
tasks, all_dataset, transformers = load_bace_classification(
featurizer=featurizer)
train_set, _, _ = all_dataset
model = dc.models.GATModel(mode='classification',
n_tasks=len(tasks),
graph_attention_layers=[2],
n_attention_heads=1,
residual=False,
predictor_hidden_feats=2)
model.fit(train_set, nb_epoch=1)
def test_mpnn_classification():
# load datasets
featurizer = MolGraphConvFeaturizer(use_edges=True)
tasks, dataset, transformers, metric = get_dataset('classification',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = MPNNModel(mode='classification',
n_tasks=n_tasks,
learning_rate=0.0005)
# overfit test
model.fit(dataset, nb_epoch=200)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
# test on a small MoleculeNet dataset
from deepchem.molnet import load_bace_classification
tasks, all_dataset, transformers = load_bace_classification(
featurizer=featurizer)
train_set, _, _ = all_dataset
model = MPNNModel(mode='classification',
n_tasks=len(tasks),
node_out_feats=2,
edge_hidden_feats=2,
num_step_message_passing=1,
num_step_set2set=1,
num_layer_set2set=1)
model.fit(train_set, nb_epoch=1)
def test_gcn_regression():
# load datasets
featurizer = MolGraphConvFeaturizer()
tasks, dataset, transformers, metric = get_dataset('regression',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = GCNModel(mode='regression',
n_tasks=n_tasks,
number_atom_features=30,
batch_size=10,
learning_rate=0.003)
# overfit test
model.fit(dataset, nb_epoch=300)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean_absolute_error'] < 0.5
# test on a small MoleculeNet dataset
from deepchem.molnet import load_delaney
tasks, all_dataset, transformers = load_delaney(featurizer=featurizer)
train_set, _, _ = all_dataset
model = dc.models.GCNModel(n_tasks=len(tasks),
graph_conv_layers=[2],
residual=False,
predictor_hidden_feats=2)
model.fit(train_set, nb_epoch=1)
def test_attentivefp_classification():
# load datasets
featurizer = MolGraphConvFeaturizer(use_edges=True)
tasks, dataset, transformers, metric = get_dataset(
'classification', featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = AttentiveFPModel(
mode='classification',
n_tasks=n_tasks,
batch_size=10,
learning_rate=0.001)
# overfit test
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
# test on a small MoleculeNet dataset
from deepchem.molnet import load_bace_classification
tasks, all_dataset, transformers = load_bace_classification(
featurizer=featurizer)
train_set, _, _ = all_dataset
model = AttentiveFPModel(
mode='classification',
n_tasks=len(tasks),
num_layers=1,
num_timesteps=1,
graph_feat_size=2)
model.fit(train_set, nb_epoch=1)
def test_jax_model_for_regression():
tasks, dataset, transformers, metric = get_dataset('regression',
featurizer='ECFP')
# sample network
def forward_model(x):
net = hk.nets.MLP([512, 256, 128, 2])
return net(x)
def rms_loss(pred, tar, w):
return jnp.mean(optax.l2_loss(pred, tar))
# Model Initialization
params_init, forward_fn = hk.transform(forward_model)
rng = jax.random.PRNGKey(500)
inputs, _, _, _ = next(iter(dataset.iterbatches(batch_size=256)))
modified_inputs = jnp.array(
[x.astype(np.float32) if x.dtype == np.float64 else x for x in inputs])
params = params_init(rng, modified_inputs)
# Loss Function
criterion = rms_loss
# JaxModel Working
j_m = JaxModel(forward_fn,
params,
criterion,
batch_size=256,
learning_rate=0.001,
log_frequency=2)
_ = j_m.fit(dataset, nb_epochs=25, deterministic=True)
scores = j_m.evaluate(dataset, [metric])
assert scores[metric.name] < 0.5
def test_mpnn_regression():
# load datasets
featurizer = MolGraphConvFeaturizer(use_edges=True)
tasks, dataset, transformers, metric = get_dataset('regression',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = MPNNModel(mode='regression', n_tasks=n_tasks, batch_size=10)
# overfit test
model.fit(dataset, nb_epoch=400)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean_absolute_error'] < 0.5
# test on a small MoleculeNet dataset
from deepchem.molnet import load_delaney
tasks, all_dataset, transformers = load_delaney(featurizer=featurizer)
train_set, _, _ = all_dataset
model = MPNNModel(mode='regression',
n_tasks=len(tasks),
node_out_feats=2,
edge_hidden_feats=2,
num_step_message_passing=1,
num_step_set2set=1,
num_layer_set2set=1)
model.fit(train_set, nb_epoch=1)
def test_pagtn_reload():
# load datasets
featurizer = PagtnMolGraphFeaturizer(max_length=5)
tasks, dataset, transformers, metric = get_dataset(
'classification', featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model_dir = tempfile.mkdtemp()
model = PagtnModel(
mode='classification',
n_tasks=n_tasks,
model_dir=model_dir,
batch_size=16)
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
reloaded_model = PagtnModel(
mode='classification',
n_tasks=n_tasks,
model_dir=model_dir,
batch_size=16)
reloaded_model.restore()
pred_mols = ["CCCC", "CCCCCO", "CCCCC"]
X_pred = featurizer(pred_mols)
random_dataset = dc.data.NumpyDataset(X_pred)
original_pred = model.predict(random_dataset)
reload_pred = reloaded_model.predict(random_dataset)
assert np.all(original_pred == reload_pred)
def test_attentivefp_regression():
# load datasets
featurizer = MolGraphConvFeaturizer(use_edges=True)
tasks, dataset, transformers, metric = get_dataset('regression',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = AttentiveFPModel(mode='regression', n_tasks=n_tasks, batch_size=10)
# overfit test
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean_absolute_error'] < 0.5
def test_gat_regression():
# load datasets
featurizer = MolGraphConvFeaturizer()
tasks, dataset, transformers, metric = get_dataset('regression',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = GATModel(mode='regression', n_tasks=n_tasks, batch_size=10)
# overfit test
# GAT's convergence is a little slow
model.fit(dataset, nb_epoch=300)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean_absolute_error'] < 0.5
def test_gcn_regression():
# load datasets
featurizer = MolGraphConvFeaturizer()
tasks, dataset, transformers, metric = get_dataset('regression',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = GCNModel(mode='regression',
n_tasks=n_tasks,
number_atom_features=30,
batch_size=10)
# overfit test
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean_absolute_error'] < 0.5
def test_attentivefp_classification():
# load datasets
featurizer = MolGraphConvFeaturizer(use_edges=True)
tasks, dataset, transformers, metric = get_dataset('classification',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = AttentiveFPModel(mode='classification',
n_tasks=n_tasks,
batch_size=10,
learning_rate=0.001)
# overfit test
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
def test_gat_classification():
# load datasets
featurizer = MolGraphConvFeaturizer()
tasks, dataset, transformers, metric = get_dataset('regression',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = GATModel(n_tasks=n_tasks,
loss=losses.L2Loss(),
batch_size=4,
learning_rate=0.001)
# overfit test
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
# TODO: check this asseration is correct or not
assert scores['mean_absolute_error'] < 1.0
def test_gat_classification():
# load datasets
featurizer = MolGraphConvFeaturizer()
tasks, dataset, transformers, metric = get_dataset('classification',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = GATModel(mode='classification',
n_tasks=n_tasks,
batch_size=10,
learning_rate=0.001)
# overfit test
# GAT's convergence is a little slow
model.fit(dataset, nb_epoch=150)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
def test_gcn_classification():
# load datasets
featurizer = MolGraphConvFeaturizer()
tasks, dataset, transformers, metric = get_dataset('classification',
featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = GCNModel(mode='classification',
n_tasks=n_tasks,
number_atom_features=30,
batch_size=10,
learning_rate=0.0003)
# overfit test
model.fit(dataset, nb_epoch=70)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
def test_jax_model_for_classification():
tasks, dataset, transformers, metric = get_dataset('classification',
featurizer='ECFP')
# sample network
class Encoder(hk.Module):
def __init__(self, output_size: int = 2):
super().__init__()
self._network = hk.nets.MLP([512, 256, 128, output_size])
def __call__(self, x: jnp.ndarray):
x = self._network(x)
return x, jax.nn.softmax(x)
def bce_loss(pred, tar, w):
tar = jnp.array([
x.astype(np.float32) if x.dtype != np.float32 else x for x in tar
])
return jnp.mean(optax.softmax_cross_entropy(pred[0], tar))
# Model Initilisation
params_init, forward_fn = hk.transform(lambda x: Encoder()(x)) # noqa
rng = jax.random.PRNGKey(500)
inputs, _, _, _ = next(iter(dataset.iterbatches(batch_size=256)))
modified_inputs = jnp.array(
[x.astype(np.float32) if x.dtype == np.float64 else x for x in inputs])
params = params_init(rng, modified_inputs)
# Loss Function
criterion = bce_loss
# JaxModel Working
j_m = JaxModel(forward_fn,
params,
criterion,
output_types=['loss', 'prediction'],
batch_size=256,
learning_rate=0.001,
log_frequency=2)
_ = j_m.fit(dataset, nb_epochs=25, deterministic=True)
scores = j_m.evaluate(dataset, [metric])
assert scores[metric.name] > 0.8
def test_pagtn_regression():
# load datasets
featurizer = PagtnMolGraphFeaturizer(max_length=5)
tasks, dataset, transformers, metric = get_dataset(
'regression', featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = PagtnModel(mode='regression', n_tasks=n_tasks, batch_size=16)
# overfit test
model.fit(dataset, nb_epoch=150)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean_absolute_error'] < 0.65
# test on a small MoleculeNet dataset
from deepchem.molnet import load_delaney
tasks, all_dataset, transformers = load_delaney(featurizer=featurizer)
train_set, _, _ = all_dataset
model = PagtnModel(mode='regression', n_tasks=n_tasks, batch_size=16)
model.fit(train_set, nb_epoch=1)
def test_pagtn_classification():
# load datasets
featurizer = PagtnMolGraphFeaturizer(max_length=5)
tasks, dataset, transformers, metric = get_dataset(
'classification', featurizer=featurizer)
# initialize models
n_tasks = len(tasks)
model = PagtnModel(mode='classification', n_tasks=n_tasks, batch_size=16)
# overfit test
model.fit(dataset, nb_epoch=100)
scores = model.evaluate(dataset, [metric], transformers)
assert scores['mean-roc_auc_score'] >= 0.85
# test on a small MoleculeNet dataset
from deepchem.molnet import load_bace_classification
tasks, all_dataset, transformers = load_bace_classification(
featurizer=featurizer)
train_set, _, _ = all_dataset
model = PagtnModel(mode='classification', n_tasks=len(tasks), batch_size=16)
model.fit(train_set, nb_epoch=1)
