def fit(self,
dataset,
nb_epoch=10,
batch_size=50,
pad_batches=False,
**kwargs):
"""
Fits a model on data in a Dataset object.
"""
# TODO(rbharath/enf): We need a structured way to deal with potential GPU
# memory overflows.
for epoch in range(nb_epoch):
log("Starting epoch %s" % str(epoch + 1), self.verbosity)
losses = []
for (X_batch, y_batch, w_batch,
ids_batch) in dataset.iterbatches(batch_size,
pad_batches=pad_batches):
if self.fit_transformers:
X_batch, y_batch, w_batch = self.transform_on_batch(
X_batch, y_batch, w_batch)
if pad_batches:
X_batch, y_batch, w_batch, ids_batch = pad_batch(
batch_size, X_batch, y_batch, w_batch, ids_batch)
losses.append(self.fit_on_batch(X_batch, y_batch, w_batch))
log(
"Avg loss for epoch %d: %f" %
(epoch + 1, np.array(losses).mean()), self.verbosity)
def predict_proba_on_batch(self, support, test_batch):
"""Make predictions on batch of data."""
n_samples = len(test_batch)
padded_test_batch = NumpyDataset(
*pad_batch(self.test_batch_size, test_batch.X, test_batch.y,
test_batch.w, test_batch.ids))
feed_dict = self.construct_feed_dict(padded_test_batch, support)
# Get scores
pred, scores = self.sess.run([self.pred_op, self.scores_op],
feed_dict=feed_dict)
y_pred_batch = to_one_hot(np.round(pred))
return y_pred_batch
def fit(self, dataset):
"""
Fits a model on data in a Dataset object.
"""
# TODO(rbharath/enf): We need a structured way to deal with potential GPU
# memory overflows.
batch_size = self.model_params["batch_size"]
if "pad_batches" in self.model_params:
pad_batches = self.model_params["pad_batches"]
else:
pad_batches = False
for epoch in range(self.model_params["nb_epoch"]):
log("Starting epoch %s" % str(epoch + 1), self.verbosity)
losses = []
for (X_batch, y_batch, w_batch, ids_batch) in dataset.iterbatches(batch_size, pad_batches=pad_batches):
if self.fit_transformers:
X_batch, y_batch, w_batch = self.transform_on_batch(X_batch, y_batch, w_batch)
if pad_batches:
X_batch, y_batch, w_batch, ids_batch = pad_batch(batch_size, X_batch, y_batch, w_batch, ids_batch)
losses.append(self.fit_on_batch(X_batch, y_batch, w_batch))
log("Avg loss for epoch %d: %f" % (epoch + 1, np.array(losses).mean()), self.verbosity)
def test_pad_batches(self):
"""Test that pad_batch pads batches correctly."""
batch_size = 100
num_features = 10
num_tasks = 5
# Test cases where n_samples < 2*n_samples < batch_size
n_samples = 29
X_b = np.zeros((n_samples, num_features))
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples,))
X_out, y_out, w_out, ids_out = pad_batch(
batch_size, X_b, y_b, w_b, ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(ids_out) == batch_size
# Test cases where n_samples < batch_size
n_samples = 79
X_b = np.zeros((n_samples, num_features))
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples,))
X_out, y_out, w_out, ids_out = pad_batch(
batch_size, X_b, y_b, w_b, ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(ids_out) == batch_size
# Test case where n_samples == batch_size
n_samples = 100
X_b = np.zeros((n_samples, num_features))
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples,))
X_out, y_out, w_out, ids_out = pad_batch(
batch_size, X_b, y_b, w_b, ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(ids_out) == batch_size
# Test case for object featurization.
n_samples = 2
X_b = np.array([{"a": 1}, {"b": 2}])
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples,))
X_out, y_out, w_out, ids_out = pad_batch(
batch_size, X_b, y_b, w_b, ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(ids_out) == batch_size
# Test case for more complicated object featurization
n_samples = 2
X_b = np.array([(1, {"a": 1}), (2, {"b": 2})])
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples,))
X_out, y_out, w_out, ids_out = pad_batch(
batch_size, X_b, y_b, w_b, ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(ids_out) == batch_size
# Test case with multidimensional data
n_samples = 50
num_atoms = 15
d = 3
X_b = np.zeros((n_samples, num_atoms, d))
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples,))
X_out, y_out, w_out, ids_out = pad_batch(
batch_size, X_b, y_b, w_b, ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(ids_out) == batch_size
def test_pad_batches(self):
"""Test that pad_batch pads batches correctly."""
batch_size = 100
num_features = 10
num_tasks = 5
# Test cases where n_samples < 2*n_samples < batch_size
n_samples = 29
X_b = np.zeros((n_samples, num_features))
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples, ))
X_out, y_out, w_out, ids_out = pad_batch(batch_size, X_b, y_b, w_b,
ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(
ids_out) == batch_size
# Test cases where n_samples < batch_size
n_samples = 79
X_b = np.zeros((n_samples, num_features))
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples, ))
X_out, y_out, w_out, ids_out = pad_batch(batch_size, X_b, y_b, w_b,
ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(
ids_out) == batch_size
# Test case where n_samples == batch_size
n_samples = 100
X_b = np.zeros((n_samples, num_features))
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples, ))
X_out, y_out, w_out, ids_out = pad_batch(batch_size, X_b, y_b, w_b,
ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(
ids_out) == batch_size
# Test case for object featurization.
n_samples = 2
X_b = np.array([{"a": 1}, {"b": 2}])
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples, ))
X_out, y_out, w_out, ids_out = pad_batch(batch_size, X_b, y_b, w_b,
ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(
ids_out) == batch_size
# Test case for more complicated object featurization
n_samples = 2
X_b = np.array([(1, {"a": 1}), (2, {"b": 2})])
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples, ))
X_out, y_out, w_out, ids_out = pad_batch(batch_size, X_b, y_b, w_b,
ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(
ids_out) == batch_size
# Test case with multidimensional data
n_samples = 50
num_atoms = 15
d = 3
X_b = np.zeros((n_samples, num_atoms, d))
y_b = np.zeros((n_samples, num_tasks))
w_b = np.zeros((n_samples, num_tasks))
ids_b = np.zeros((n_samples, ))
X_out, y_out, w_out, ids_out = pad_batch(batch_size, X_b, y_b, w_b,
ids_b)
assert len(X_out) == len(y_out) == len(w_out) == len(
ids_out) == batch_size