def test_forward_pass_with_missing_data():
it = Undivided(default=np.random.randn(3, 2, 4))(simple_net.handler)
with pytest.raises(KeyError):
for _ in run_network(simple_net, it):
pass
for _ in run_network(simple_net, it, all_inputs=False):
pass
def test_forward_pass_with_missing_data():
it = Undivided(default=np.random.randn(3, 2, 4))(simple_net.handler)
with pytest.raises(KeyError):
for _ in run_network(simple_net, it):
pass
for _ in run_network(simple_net, it, all_inputs=False):
pass
def train(self, net, training_data_iter, **named_data_iters):
"""
Train a network using a data iterator and further named data
iterators.
"""
if self.verbose:
if self.logging_function == print:
self.logging_function('\n\n' + 10 * '- ' + "Before Training" +
10 * ' -')
else:
self.logging_function(10 * '- ' + "Before Training" + 10 * ' -')
assert set(training_data_iter.data_shapes.keys()) == set(
net.buffer.Input.outputs.keys()), \
"The data names provided by the training data iterator {} do not "\
"map to the network input names {}".format(
training_data_iter.data_shapes.keys(),
net.buffer.Input.outputs.keys())
self.stepper.start(net)
named_data_iters['training_data_iter'] = training_data_iter
self._start_hooks(net, named_data_iters)
if self._emit_hooks(net, 'update') or self._emit_hooks(net, 'epoch'):
return
should_stop = False
while not should_stop:
self.current_epoch_nr += 1
sys.stdout.flush()
train_scores = {s.__name__: [] for s in self.train_scorers}
train_scores.update({n: [] for n in net.get_loss_values()})
if self.verbose:
if self.logging_function == print:
self.logging_function('\n\n' + 12 * '- ' + "Epoch" +
str(self.current_epoch_nr) + 12 * ' -')
else:
self.logging_function(12 * '- ' + "Epoch" +
str(self.current_epoch_nr) + 12 * ' -')
iterator = training_data_iter(handler=net.handler)
for _ in run_network(net, iterator):
self.current_update_nr += 1
self.stepper.run()
gather_losses_and_scores(net, self.train_scorers, train_scores)
net.apply_weight_modifiers()
if self._emit_hooks(net, 'update'):
should_stop = True
break
self._add_log('rolling_training',
aggregate_losses_and_scores(train_scores, net,
self.train_scorers))
should_stop |= self._emit_hooks(net, 'epoch')
