def fit_loop(model, inputs, targets, sample_weights=None, class_weight=None, val_inputs=None, val_targets=None, val_sample_weights=None, batch_size=None, epochs=1, verbose=1, callbacks=None, shuffle=True, callback_metrics=None, initial_epoch=0, steps_per_epoch=None, validation_steps=None): """Fit function for eager execution. Arguments: model: Instance of the model that is being executed in Eager mode. inputs: List of input arrays. targets: List of target arrays. sample_weights: Optional list of sample weight arrays. class_weight: Optional class-weight array to weight the importance of samples in `inputs` based on the class they belong to, as conveyed by `targets`. val_inputs: Input data for validation. val_targets: Target data for validation. val_sample_weights: Sample weight data for validation. batch_size: Integer batch size or None if unknown. epochs: Number of times to iterate over the data verbose: Verbosity mode, 0, 1 or 2 callbacks: List of callbacks to be called during training shuffle: Whether to shuffle the data at the beginning of each epoch callback_metrics: List of strings, the display names of the metrics passed to the callbacks. They should be the concatenation of list the display names of the outputs of `f` and the list of display names of the outputs of `f_val`. initial_epoch: Epoch at which to start training (useful for resuming a previous training run) steps_per_epoch: Total number of steps (batches of samples) before declaring one epoch finished and starting the next epoch. Ignored with the default value of `None`. validation_steps: Number of steps to run validation for (only if doing validation from data tensors). Ignored with default value of `None`. Returns: `History` object. Raises: ValueError: In case of invalid argument values. """ # Convert training inputs to an EagerIterator inputs, steps_per_epoch = training_utils.convert_to_iterator( x=inputs, y=targets, sample_weights=sample_weights, batch_size=batch_size, steps_per_epoch=steps_per_epoch, epochs=epochs, shuffle=shuffle) # Required for eager execution with backend.learning_phase_scope(1): do_validation = False if val_inputs: do_validation = True num_train_samples = None out_labels = None if model._is_compiled: out_labels = model.metrics_names if do_validation: callback_metrics = copy.copy(out_labels) + [ 'val_' + n for n in out_labels ] else: callback_metrics = copy.copy(out_labels) model.history = cbks.History() callbacks = [cbks.BaseLogger()] + (callbacks or []) + [model.history] if verbose: callbacks += [cbks.ProgbarLogger('steps')] callbacks = cbks.CallbackList(callbacks) # it's possible to callback a different model than self # (used by Sequential models) if hasattr(model, 'callback_model') and model.callback_model: callback_model = model.callback_model else: callback_model = model callbacks.set_model(callback_model) callback_params = { 'batch_size': batch_size, 'epochs': epochs, 'steps': steps_per_epoch, 'samples': num_train_samples, 'verbose': verbose, 'do_validation': do_validation, 'metrics': callback_metrics or [], } if validation_steps: callback_params.update({'validation_steps': validation_steps}) callbacks.set_params(callback_params) for cbk in callbacks: if not val_inputs: cbk.validation_data = [] elif isinstance(val_inputs, iterator_ops.EagerIterator): cbk.validation_data = val_inputs elif val_sample_weights: cbk.validation_data = val_inputs + val_targets + val_sample_weights else: cbk.validation_data = val_inputs + val_targets # validation_data must be set before on_train_begin() is called # so that TensorboardCallback can validate its input callbacks.on_train_begin() callback_model.stop_training = False for epoch in range(initial_epoch, epochs): callbacks.on_epoch_begin(epoch) epoch_logs = {} iterator_fit_loop(model, inputs, class_weight, steps_per_epoch=steps_per_epoch, callback_model=callback_model, out_labels=out_labels, epoch_logs=epoch_logs, val_inputs=val_inputs, val_targets=val_targets, val_sample_weights=val_sample_weights, epochs=epochs, verbose=verbose, callbacks=callbacks, callback_metrics=callback_metrics, validation_steps=validation_steps, do_validation=do_validation, batch_size=batch_size) callbacks.on_epoch_end(epoch, epoch_logs) if callback_model.stop_training: break callbacks.on_train_end() return model.history
def fit_loop(model, inputs, targets, sample_weights=None, batch_size=None, epochs=100, verbose=1, callbacks=None, val_inputs=None, val_targets=None, val_sample_weights=None, shuffle=True, callback_metrics=None, initial_epoch=0, steps_per_epoch=None, validation_steps=None): """Abstract fit function for arrays of data. Arguments: model: Keras Model instance. inputs: List of input arrays. targets: List of target arrays. sample_weights: Optional list of sample weight arrays. batch_size: Integer batch size or None if unknown. epochs: Number of times to iterate over the data verbose: Verbosity mode, 0, 1 or 2 callbacks: List of callbacks to be called during training val_inputs: List of input arrays. val_targets: List of target arrays. val_sample_weights: Optional list of sample weight arrays. shuffle: Whether to shuffle the data at the beginning of each epoch callback_metrics: List of strings, the display names of the metrics passed to the callbacks. They should be the concatenation of list the display names of the outputs of `f` and the list of display names of the outputs of `f_val`. initial_epoch: Epoch at which to start training (useful for resuming a previous training run) steps_per_epoch: Total number of steps (batches of samples) before declaring one epoch finished and starting the next epoch. Ignored with the default value of `None`. validation_steps: Number of steps to run validation for (only if doing validation from data tensors). Ignored with the default value of `None`. Returns: `History` object. Raises: ValueError: in case of invalid arguments. """ model._make_train_function() f = model.train_function sample_weights = sample_weights or [] val_sample_weights = val_sample_weights or [] if model.uses_learning_phase and not isinstance(K.learning_phase(), int): ins = inputs + targets + sample_weights + [1] if val_inputs: val_ins = val_inputs + val_targets + val_sample_weights + [1] else: ins = inputs + targets + sample_weights if val_inputs: val_ins = val_inputs + val_targets + val_sample_weights if not val_inputs: val_ins = [] do_validation = False if val_inputs: do_validation = True if (steps_per_epoch is None and verbose and inputs and hasattr(inputs[0], 'shape') and hasattr(val_inputs[0], 'shape')): print('Train on %d samples, validate on %d samples' % (inputs[0].shape[0], val_inputs[0].shape[0])) if validation_steps: do_validation = True if steps_per_epoch is None: raise ValueError('Can only use `validation_steps` ' 'when doing step-wise ' 'training, i.e. `steps_per_epoch` ' 'must be set.') out_labels = model.metrics_names if do_validation: callback_metrics = copy.copy(out_labels) + [ 'val_' + n for n in out_labels ] else: callback_metrics = copy.copy(out_labels) num_train_samples = training_utils.check_num_samples( ins, batch_size, steps_per_epoch, 'steps_per_epoch') if num_train_samples is not None: index_array = np.arange(num_train_samples) model.history = cbks.History() all_callbacks = [ cbks.BaseLogger(stateful_metrics=model.stateful_metric_names) ] if verbose: if steps_per_epoch is not None: count_mode = 'steps' else: count_mode = 'samples' all_callbacks.append( cbks.ProgbarLogger(count_mode, stateful_metrics=model.stateful_metric_names)) all_callbacks += (callbacks or []) + [model.history] callbacks = cbks.CallbackList(all_callbacks) out_labels = out_labels or [] # it's possible to callback a different model than self # (used by Sequential models) if hasattr(model, 'callback_model') and model.callback_model: callback_model = model.callback_model else: callback_model = model callbacks.set_model(callback_model) callbacks.set_params({ 'batch_size': batch_size, 'epochs': epochs, 'steps': steps_per_epoch, 'samples': num_train_samples, 'verbose': verbose, 'do_validation': do_validation, 'metrics': callback_metrics or [], }) callbacks.on_train_begin() callback_model.stop_training = False for cbk in callbacks: cbk.validation_data = val_ins # To prevent a slowdown, we find beforehand the arrays that need conversion. feed = model._feed_inputs + model._feed_targets + model._feed_sample_weights indices_for_conversion_to_dense = [] for i in range(len(feed)): if issparse is not None and issparse( ins[i]) and not K.is_sparse(feed[i]): indices_for_conversion_to_dense.append(i) for epoch in range(initial_epoch, epochs): # Reset stateful metrics for m in model.stateful_metric_functions: m.reset_states() # Update callbacks callbacks.on_epoch_begin(epoch) epoch_logs = {} if steps_per_epoch is not None: for step_index in range(steps_per_epoch): batch_logs = {} batch_logs['batch'] = step_index batch_logs['size'] = 1 callbacks.on_batch_begin(step_index, batch_logs) try: outs = f(ins) except errors.OutOfRangeError: logging.warning( 'Your dataset iterator ran out of data; ' 'interrupting training. Make sure that your dataset ' 'can generate at least `steps_per_epoch * epochs` ' 'batches (in this case, %d batches).' % steps_per_epoch * epochs) break if not isinstance(outs, list): outs = [outs] for l, o in zip(out_labels, outs): batch_logs[l] = o callbacks.on_batch_end(step_index, batch_logs) if callback_model.stop_training: break if do_validation: val_outs = test_loop(model, val_inputs, val_targets, sample_weights=val_sample_weights, batch_size=batch_size, steps=validation_steps, verbose=0) if not isinstance(val_outs, list): val_outs = [val_outs] # Same labels assumed. for l, o in zip(out_labels, val_outs): epoch_logs['val_' + l] = o else: if shuffle == 'batch': index_array = training_utils.batch_shuffle( index_array, batch_size) elif shuffle: np.random.shuffle(index_array) batches = make_batches(num_train_samples, batch_size) for batch_index, (batch_start, batch_end) in enumerate(batches): batch_ids = index_array[batch_start:batch_end] try: if isinstance(ins[-1], int): # Do not slice the training phase flag. ins_batch = slice_arrays(ins[:-1], batch_ids) + [ins[-1]] else: ins_batch = slice_arrays(ins, batch_ids) except TypeError: raise TypeError('TypeError while preparing batch. ' 'If using HDF5 input data, ' 'pass shuffle="batch".') batch_logs = {} batch_logs['batch'] = batch_index batch_logs['size'] = len(batch_ids) callbacks.on_batch_begin(batch_index, batch_logs) for i in indices_for_conversion_to_dense: ins_batch[i] = ins_batch[i].toarray() outs = f(ins_batch) if not isinstance(outs, list): outs = [outs] for l, o in zip(out_labels, outs): batch_logs[l] = o callbacks.on_batch_end(batch_index, batch_logs) if callback_model.stop_training: break if batch_index == len(batches) - 1: # Last batch. if do_validation: val_outs = test_loop(model, val_inputs, val_targets, sample_weights=val_sample_weights, batch_size=batch_size, verbose=0) if not isinstance(val_outs, list): val_outs = [val_outs] # Same labels assumed. for l, o in zip(out_labels, val_outs): epoch_logs['val_' + l] = o callbacks.on_epoch_end(epoch, epoch_logs) if callback_model.stop_training: break callbacks.on_train_end() return model.history
def build_callbacks(self, conf, callbacks_list): ''' The purpose of the method is to set up logging and history. It is based on Keras Callbacks https://github.com/fchollet/keras/blob/fbc9a18f0abc5784607cd4a2a3886558efa3f794/keras/callbacks.py Currently used callbacks include: BaseLogger, CSVLogger, EarlyStopping. Other possible callbacks to add in future: RemoteMonitor, LearningRateScheduler Argument list: - conf: There is a "callbacks" section in conf.yaml file. Relevant parameters are: - list: Parameter specifying additional callbacks, read in the driver script and passed as an argument of type list (see next arg) - metrics: List of quantities monitored during training and validation - mode: one of {auto, min, max}. The decision to overwrite the current save file is made based on either the maximization or the minimization of the monitored quantity. For val_acc, this should be max, for val_loss this should be min, etc. In auto mode, the direction is automatically inferred from the name of the monitored quantity. - monitor: Quantity used for early stopping, has to be from the list of metrics - patience: Number of epochs used to decide on whether to apply early stopping or continue training - callbacks_list: uses callbacks.list configuration parameter, specifies the list of additional callbacks Returns: modified list of callbacks ''' mode = conf['callbacks']['mode'] monitor = conf['callbacks']['monitor'] patience = conf['callbacks']['patience'] csvlog_save_path = conf['paths']['csvlog_save_path'] # CSV callback is on by default if not os.path.exists(csvlog_save_path): os.makedirs(csvlog_save_path) callbacks_list = conf['callbacks']['list'] callbacks = [cbks.BaseLogger()] callbacks += [self.history] callbacks += [ cbks.CSVLogger("{}callbacks-{}.log".format( csvlog_save_path, datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S"))) ] if "earlystop" in callbacks_list: callbacks += [ cbks.EarlyStopping(patience=patience, monitor=monitor, mode=mode) ] if "lr_scheduler" in callbacks_list: pass return cbks.CallbackList(callbacks)
def fit_loop(model, inputs, targets, epochs=100, verbose=1, callbacks=None, val_inputs=None, val_targets=None, callback_metrics=None, initial_epoch=0, steps_per_epoch=None, validation_steps=None): """fit function when using DistributionStrategy for training. Arguments: model: Keras Model instance. inputs: List of input arrays. targets: List of target arrays. epochs: Number of times to iterate over the data verbose: Verbosity mode, 0, 1 or 2 callbacks: List of callbacks to be called during training val_inputs: List of input arrays. val_targets: List of target arrays. callback_metrics: List of strings, the display names of the metrics passed to the callbacks. They should be the concatenation of list the display names of the outputs of `f` and the list of display names of the outputs of `f_val`. initial_epoch: Epoch at which to start training (useful for resuming a previous training run) steps_per_epoch: Total number of steps (batches of samples) before declaring one epoch finished and starting the next epoch. Ignored with the default value of `None`. validation_steps: Number of steps to run validation for (only if doing validation from data tensors). Ignored with the default value of `None`. Returns: `History` object. Raises: ValueError: in case of invalid arguments. """ current_strategy = model._distribution_strategy def _per_device_train_function(model): model._make_train_function() return (model.train_function.inputs, model.train_function.outputs, model.train_function.updates_op, model.train_function.session_kwargs) with current_strategy.scope(): # Create train ops on each of the devices when we call # `_per_device_train_function`. (grouped_inputs, grouped_outputs, grouped_updates, grouped_session_args) = current_strategy.call_for_each_tower( _per_device_train_function, model._grouped_model) # Unwrap all the per device values returned from `call_for_each_tower`. # Unwrapping per device values gives you a list of values that can be # used to construct a new train function that is composed of update ops on # all the devices over which the model is distributed. (all_inputs, all_outputs, all_updates, all_session_args) = distributed_training_utils.unwrap_values( current_strategy, grouped_inputs, grouped_outputs, grouped_updates, grouped_session_args, with_loss_tensor=True) # Dataset inputs and targets are also per devices values that need to be # unwrapped. dataset_inputs = distributed_training_utils.flatten_perdevice_values( current_strategy, inputs) dataset_targets = distributed_training_utils.flatten_perdevice_values( current_strategy, targets) # Create a train function that is composed of all the parameters above. distributed_train_function = K.Function(all_inputs, all_outputs, updates=all_updates, name='distributed_train_function', **all_session_args) # We need to set sample_weights to None since there are sample weight # placeholders that are created with default values. sample_weights = [ None for _ in range(len(model.outputs) * current_strategy.num_towers) ] if model.uses_learning_phase and not isinstance(K.learning_phase(), int): ins = dataset_inputs + dataset_targets + sample_weights + [1] else: ins = dataset_inputs + dataset_targets do_validation = False if validation_steps: do_validation = True if steps_per_epoch is None: raise ValueError('Can only use `validation_steps` ' 'when doing step-wise ' 'training, i.e. `steps_per_epoch` ' 'must be set.') out_labels = model.metrics_names if do_validation: callback_metrics = copy.copy(out_labels) + [ 'val_' + n for n in out_labels ] else: callback_metrics = copy.copy(out_labels) model.history = cbks.History() all_callbacks = [ cbks.BaseLogger(stateful_metrics=model.stateful_metric_names) ] if verbose: # We assume that `steps_per_epoch` is always set since we have to use # Datasets. count_mode = 'steps' all_callbacks.append( cbks.ProgbarLogger(count_mode, stateful_metrics=model.stateful_metric_names)) all_callbacks += (callbacks or []) + [model.history] callbacks = cbks.CallbackList(all_callbacks) out_labels = out_labels or [] # We set the callback model to an instance of the `DistributedModel` that we # create in the `compile` call. The `DistributedModel` is initialized with # the first replicated model. We need to set the callback model to a # DistributedModel to allow us to override saving and loading weights when # we checkpoint the model during training. callback_model = model._replicated_model callbacks.set_model(callback_model) callbacks.set_params({ 'epochs': epochs, 'steps': steps_per_epoch, 'samples': None, 'verbose': verbose, 'do_validation': do_validation, 'metrics': callback_metrics or [], }) callbacks.on_train_begin() callback_model.stop_training = False out_labels = out_labels or [] # Copy the weights from the original model to each of the replicated models. orig_model_weights = model.get_weights() with current_strategy.scope(): distributed_model = current_strategy.unwrap(model._grouped_model)[0] distributed_training_utils.set_weights(current_strategy, distributed_model, orig_model_weights) for epoch in range(initial_epoch, epochs): callbacks.on_epoch_begin(epoch) if steps_per_epoch is not None: epoch_logs = {} for step_index in range(steps_per_epoch): batch_logs = {} batch_logs['batch'] = step_index batch_logs['size'] = 1 callbacks.on_batch_begin(step_index, batch_logs) try: outs = distributed_train_function(ins) except errors.OutOfRangeError: logging.warning( 'Your dataset iterator ran out of data; ' 'interrupting training. Make sure that your dataset ' 'can generate at least `steps_per_epoch * epochs` ' 'batches (in this case, %d batches).' % steps_per_epoch * epochs) break if not isinstance(outs, list): outs = [outs] # TODO(anjalisridhar): Temporary workaround for aggregating metrics # across towers. Replace with the new metrics module eventually. merged_output = [] # The first output is the total loss. merged_output.append(outs[0]) current_index = 1 num_devices = len(current_strategy._devices) # Each label in `out_labels` corresponds to one set of metrics. The # number of metric values corresponds to the number of devices. We # currently take the mean of the values. for _ in out_labels[1:]: m = np.mean(outs[current_index:current_index + num_devices]) merged_output.append(m) current_index += num_devices for l, o in zip(out_labels, outs): batch_logs[l] = o callbacks.on_batch_end(step_index, batch_logs) if callback_model.stop_training: break if do_validation: val_outs = test_loop(model, val_inputs, val_targets, steps=validation_steps, verbose=0) if not isinstance(val_outs, list): val_outs = [val_outs] # Same labels assumed. for l, o in zip(out_labels, val_outs): epoch_logs['val_' + l] = o callbacks.on_epoch_end(epoch, epoch_logs) if callback_model.stop_training: break callbacks.on_train_end() # Copy the weights back from the replicated model to the original model. with current_strategy.scope(): updated_weights = current_strategy.unwrap( model._grouped_model)[0].get_weights() model.set_weights(updated_weights) return model.history
def fit_generator(model, generator, steps_per_epoch=None, epochs=1, verbose=1, callbacks=None, validation_data=None, validation_steps=None, class_weight=None, max_queue_size=10, workers=1, use_multiprocessing=False, shuffle=True, initial_epoch=0): """See docstring for `Model.fit_generator`.""" wait_time = 0.01 # in seconds epoch = initial_epoch do_validation = bool(validation_data) is_sequence = isinstance(generator, Sequence) if not is_sequence and use_multiprocessing and workers > 1: logging.warning( UserWarning('Using a generator with `use_multiprocessing=True`' ' and multiple workers may duplicate your data.' ' Please consider using the`keras.utils.Sequence' ' class.')) if steps_per_epoch is None: if is_sequence: steps_per_epoch = len(generator) else: raise ValueError('`steps_per_epoch=None` is only valid for a' ' generator based on the `keras.utils.Sequence`' ' class. Please specify `steps_per_epoch` or use' ' the `keras.utils.Sequence` class.') # python 2 has 'next', 3 has '__next__' # avoid any explicit version checks val_gen = ( hasattr(validation_data, 'next') or hasattr(validation_data, '__next__') or isinstance(validation_data, Sequence)) if (val_gen and not isinstance(validation_data, Sequence) and not validation_steps): raise ValueError('`validation_steps=None` is only valid for a' ' generator based on the `keras.utils.Sequence`' ' class. Please specify `validation_steps` or use' ' the `keras.utils.Sequence` class.') # Prepare display labels. out_labels = model.metrics_names callback_metrics = out_labels + ['val_%s' % n for n in out_labels] # prepare callbacks model.history = cbks.History() callbacks = [cbks.BaseLogger()] + (callbacks or []) + [model.history] if verbose: callbacks += [cbks.ProgbarLogger(count_mode='steps')] callbacks = cbks.CallbackList(callbacks) # it's possible to callback a different model than self: if hasattr(model, 'callback_model') and model.callback_model: callback_model = model.callback_model else: callback_model = model callbacks.set_model(callback_model) callback_params = { 'epochs': epochs, 'steps': steps_per_epoch, 'verbose': verbose, 'do_validation': do_validation, 'metrics': callback_metrics, } if do_validation: # need to create the test_function before start of the first epoch # because TensorBoard callback on_epoch_begin adds summary to the # list of fetches of the test_function model._make_test_function() # determine the number of validation batches given a generator if validation_steps: callback_params.update({'validation_steps': validation_steps}) elif isinstance(validation_data, Sequence): callback_params.update({'validation_steps': len(validation_data)}) callbacks.set_params(callback_params) enqueuer = None val_enqueuer = None try: if do_validation and not val_gen: # Prepare data for validation if len(validation_data) == 2: val_x, val_y = validation_data # pylint: disable=unpacking-non-sequence val_sample_weight = None elif len(validation_data) == 3: val_x, val_y, val_sample_weight = validation_data # pylint: disable=unpacking-non-sequence else: raise ValueError( '`validation_data` should be a tuple ' '`(val_x, val_y, val_sample_weight)` ' 'or `(val_x, val_y)`. Found: ' + str(validation_data)) val_x, val_y, val_sample_weights = model._standardize_user_data( val_x, val_y, val_sample_weight) val_data = val_x + val_y + val_sample_weights if model.uses_learning_phase and not isinstance(K.learning_phase(), int): val_data += [0.] for cbk in callbacks: cbk.validation_data = val_data if workers > 0: if is_sequence: enqueuer = OrderedEnqueuer( generator, use_multiprocessing=use_multiprocessing, shuffle=shuffle) else: enqueuer = GeneratorEnqueuer( generator, use_multiprocessing=use_multiprocessing, wait_time=wait_time) enqueuer.start(workers=workers, max_queue_size=max_queue_size) output_generator = enqueuer.get() else: if is_sequence: output_generator = iter(generator) else: output_generator = generator callback_model.stop_training = False # validation_data must be set before on_train_begin() is called # so that TensorboardCallback can validate its input callbacks.on_train_begin() # Construct epoch logs. epoch_logs = {} while epoch < epochs: for m in model.stateful_metric_functions: m.reset_states() callbacks.on_epoch_begin(epoch) steps_done = 0 batch_index = 0 while steps_done < steps_per_epoch: generator_output = next(output_generator) if not hasattr(generator_output, '__len__'): raise ValueError('Output of generator should be ' 'a tuple `(x, y, sample_weight)` ' 'or `(x, y)`. Found: ' + str(generator_output)) if len(generator_output) == 2: x, y = generator_output sample_weight = None elif len(generator_output) == 3: x, y, sample_weight = generator_output else: raise ValueError('Output of generator should be ' 'a tuple `(x, y, sample_weight)` ' 'or `(x, y)`. Found: ' + str(generator_output)) # build batch logs batch_logs = {} if isinstance(x, list): batch_size = x[0].shape[0] elif isinstance(x, dict): batch_size = list(x.values())[0].shape[0] else: batch_size = x.shape[0] batch_logs['batch'] = batch_index batch_logs['size'] = batch_size callbacks.on_batch_begin(batch_index, batch_logs) outs = model.train_on_batch( x, y, sample_weight=sample_weight, class_weight=class_weight) if not isinstance(outs, list): outs = [outs] for l, o in zip(out_labels, outs): batch_logs[l] = o callbacks.on_batch_end(batch_index, batch_logs) batch_index += 1 steps_done += 1 # Epoch finished. if steps_done >= steps_per_epoch and do_validation: if val_gen: val_outs = evaluate_generator( model, validation_data, validation_steps, workers=workers, use_multiprocessing=use_multiprocessing, max_queue_size=max_queue_size) else: # No need for try/except because # data has already been validated. val_outs = model.evaluate( val_x, val_y, batch_size=batch_size, sample_weight=val_sample_weights, verbose=0) if not isinstance(val_outs, list): val_outs = [val_outs] # Same labels assumed. for l, o in zip(out_labels, val_outs): epoch_logs['val_' + l] = o if callback_model.stop_training: break callbacks.on_epoch_end(epoch, epoch_logs) epoch += 1 if callback_model.stop_training: break finally: try: if enqueuer is not None: enqueuer.stop() finally: if val_enqueuer is not None: val_enqueuer.stop() callbacks.on_train_end() return model.history
def train_model(name, g_train, d_train, sampler, generator, samples_per_epoch, epochs, z_dim=100, verbose=1, callbacks=[], saver=None): """ Main training loop. Modified version of fit_generator. """ self = {} epoch = 0 counter = 0 out_labels = ['g_loss', 'd_loss', 'd_loss_fake', 'd_loss_legit', 'time'] # self.metrics_names callback_metrics = out_labels + ['val_' + n for n in out_labels] # prepare callbacks history = cbks.History() callbacks = [cbks.BaseLogger()] + callbacks + [history] if verbose: callbacks += [cbks.ProgbarLogger()] callbacks = cbks.CallbackList(callbacks) callback_params = { 'epochs': epochs, 'samples': samples_per_epoch, 'verbose': verbose, 'metrics': callback_metrics, } callbacks.set_params(callback_params) callbacks.on_train_begin() while epoch < epochs: callbacks.on_epoch_begin(epoch) samples_seen = 0 batch_index = 0 while samples_seen < samples_per_epoch: z, x = next(generator) # build batch logs batch_logs = {} if type(x) is list: batch_size = len(x[0]) elif type(x) is dict: batch_size = len(list(x.values())[0]) else: batch_size = len(x) batch_logs['batch'] = batch_index batch_logs['size'] = batch_size callbacks.on_batch_begin(batch_index, batch_logs) t1 = time.time() d_losses = d_train(x, z, counter) z, x = next(generator) g_loss, samples, xs = g_train(x, z, counter) outs = (g_loss, ) + d_losses + (time.time() - t1, ) counter += 1 # save samples if batch_index % 100 == 0: join_image = np.zeros_like( np.concatenate([samples[:64], xs[:64]], axis=0)) for j, (i1, i2) in enumerate(zip(samples[:64], xs[:64])): join_image[j * 2] = i1 join_image[j * 2 + 1] = i2 save_images( join_image, [8 * 2, 8], './outputs/samples_%s/train_%s_%s.png' % (name, epoch, batch_index)) samples, xs = sampler(z, x) join_image = np.zeros_like( np.concatenate([samples[:64], xs[:64]], axis=0)) for j, (i1, i2) in enumerate(zip(samples[:64], xs[:64])): join_image[j * 2] = i1 join_image[j * 2 + 1] = i2 save_images( join_image, [8 * 2, 8], './outputs/samples_%s/test_%s_%s.png' % (name, epoch, batch_index)) for l, o in zip(out_labels, outs): batch_logs[l] = o callbacks.on_batch_end(batch_index, batch_logs) # construct epoch logs epoch_logs = {} batch_index += 1 samples_seen += batch_size if saver is not None: saver(epoch) callbacks.on_epoch_end(epoch, epoch_logs) epoch += 1 # _stop.set() callbacks.on_train_end()