def on_train_begin(self, logs={}):
self.start_time = time.time()
self.last_batch_time = time.time()
deepkit.epoch(0, self.params['epochs'])
deepkit.set_info('parameters', get_total_params(self.model))
# self.job_backend.upload_keras_graph(self.model)
if self.model.optimizer and hasattr(self.model.optimizer,
'get_config'):
config = self.model.optimizer.get_config()
# deepkit.set_info('optimizer', type(self.model.optimizer).__name__)
for i, v in config.items():
deepkit.set_info('optimizer.' + str(i), v)
# compatibility with keras 1.x
if 'epochs' not in self.params and 'nb_epoch' in self.params:
self.params['epochs'] = self.params['nb_epoch']
if 'samples' not in self.params and 'nb_sample' in self.params:
self.params['samples'] = self.params['nb_sample']
xaxis = {
# 'range': [1, self.params['epochs']],
# 'title': u'Epoch ⇢'
}
yaxis = {'tickformat': '%', 'hoverformat': '%', 'rangemode': 'tozero'}
traces = ['training', 'validation']
if hasattr(self.model,
'output_layers') and len(self.model.output_layers) > 1:
traces = []
for output in self.model.output_layers:
traces.append('train_' + output.name)
traces.append('val_' + output.name)
self.accuracy_metric = deepkit.create_metric('accuracy',
traces=traces,
xaxis=xaxis,
yaxis=yaxis)
self.loss_metric = deepkit.create_loss_metric('loss', xaxis=xaxis)
self.learning_rate_metric = deepkit.create_metric(
'learning rate', traces=['start', 'end'], xaxis=xaxis)
deepkit.epoch(0, self.params['epochs'])
if hasattr(self.model,
'output_layers') and len(self.model.output_layers) > 1:
loss_traces = []
for output in self.model.output_layers:
loss_traces.append('train_' + output.name)
loss_traces.append('val_' + output.name)
self.all_losses = deepkit.create_metric('loss_all',
xaxis=xaxis,
traces=loss_traces)
def on_batch_begin(self, batch, logs={}):
if 'nb_batches' not in self.current:
batch_size = logs['size']
if 'samples' in self.params:
nb_batches = math.ceil(self.params['samples'] / batch_size) # normal nb batches
else:
nb_batches = self.params['steps']
self.current['nb_batches'] = nb_batches
self.current['batch_size'] = batch_size
deepkit.set_info('Batch size', batch_size)
import random
import deepkit
context = deepkit.context()
context.add_file(__file__)
test = deepkit.create_metric('test')
for i in range(100):
deepkit.set_info(i, random.random())
total = 1_000_000;
for i in range(1_000_000):
test.send(i, random.random())
deepkit.epoch(i, total)
print("Bye.")