def report(self, batch_in_epoch=None):
if batch_in_epoch:
batch_text = ' Batch ' + '{0:03}'.format(batch_in_epoch)
else:
batch_text = ''
if self.latest_val_acc is None:
val_acc_text = ''
else:
val_acc_text = 'ValidAcc (prev epoch)=' + '{:.1f}'.format(
100. * self.latest_val_acc) + '%'
Print.printflush(
'Epoch %s%s: TrainAcc (approx)=%s%%, %s' %
('{:,}'.format(self.latest_epoch), batch_text, '{:.1f}'.format(
100. * self.approx_train_acc_in_latest_epoch), val_acc_text),
end='\r')
if self.bokeh_output == 'notebook':
# NOTE: there are currently bugs relared to memory leaks & JavaScript
# ref: http://stackoverflow.com/questions/34442414/safely-use-push-notebook-in-bokeh-before-plot-is-shown
# ref: https://github.com/bokeh/bokeh/issues/1732
# ref: https://github.com/bokeh/bokeh/pull/3370
self.fig_train_losses_data_source.push_notebook()
self.fig_val_losses_data_source.push_notebook()
elif self.bokeh_output == 'server':
self.bokeh_session.store_objects(self.fig_train_losses_data_source,
self.fig_val_losses_data_source)
def report(self, batch_in_epoch=None):
if batch_in_epoch:
batch_text = ' Batch ' + '{0:03}'.format(batch_in_epoch)
else:
batch_text = ''
if self.latest_val_acc is None:
val_acc_text = ''
else:
val_acc_text = 'ValidAcc (prev epoch)=' + '{:.1f}'.format(100. * self.latest_val_acc) + '%'
Print.printflush(
'Epoch %s%s: TrainAcc (approx)=%s%%, %s'
% ('{:,}'.format(self.latest_epoch),
batch_text,
'{:.1f}'.format(100. * self.approx_train_acc_in_latest_epoch),
val_acc_text), end='\r')
if self.bokeh_output == 'notebook':
# NOTE: there are currently bugs relared to memory leaks & JavaScript
# ref: http://stackoverflow.com/questions/34442414/safely-use-push-notebook-in-bokeh-before-plot-is-shown
# ref: https://github.com/bokeh/bokeh/issues/1732
# ref: https://github.com/bokeh/bokeh/pull/3370
self.fig_train_losses_data_source.push_notebook()
self.fig_val_losses_data_source.push_notebook()
elif self.bokeh_output == 'server':
self.bokeh_session.store_objects(
self.fig_train_losses_data_source,
self.fig_val_losses_data_source)
def on_train_end(self, logs={}):
if self.reporting_freq:
self.report()
Print.printflush('\nFFNN Training Finished! (%s Batches in total)\n' %
'{:,}'.format(self.latest_batch))
if self.latest_val_acc is None:
Print.printflush(
'Training Accuracy (approx) = %s%%\n' %
'{:.1f}'.format(100. * self.approx_train_acc_in_latest_epoch))
else:
Print.printflush(
'Best trained FFNN (with lowest Validation Loss) is from epoch #%s'
% '{:,}'.format(self.best_model_epoch))
Print.printflush(
'Training Accuracy (approx) = %s%%, Validation Accuracy = %s%%\n'
% ('{:.1f}'.format(100. * self.best_model_train_acc),
'{:.1f}'.format(100. * self.latest_val_acc)))
def on_train_end(self, logs={}):
if self.reporting_freq:
self.report()
Print.printflush(
'\nFFNN Training Finished! (%s Batches in total)\n'
% '{:,}'.format(self.latest_batch))
if self.latest_val_acc is None:
Print.printflush(
'Training Accuracy (approx) = %s%%\n'
% '{:.1f}'.format(100. * self.approx_train_acc_in_latest_epoch))
else:
Print.printflush(
'Best trained FFNN (with lowest Validation Loss) is from epoch #%s'
% '{:,}'.format(self.best_model_epoch))
Print.printflush(
'Training Accuracy (approx) = %s%%, Validation Accuracy = %s%%\n'
% ('{:.1f}'.format(100. * self.best_model_train_acc),
'{:.1f}'.format(100. * self.latest_val_acc)))
def on_train_begin(self, logs={}):
Print.printflush('\nFFNN Training Progress')
Print.printflush('______________________')
def __init__(self, reporting_freq=False, plot_title='Neural Network Learning Curves', bokeh_output='notebook'):
super(callbacks.Callback, self).__init__()
self.latest_epoch = -1
self.latest_batch = -1
self.batches = []
self.train_losses = []
self.approx_train_acc_in_latest_epoch = 0.
self.val_losses = []
self.latest_val_acc = None
self.min_val_loss = numpy.inf
self.best_model = None
self.best_model_epoch = None
self.best_model_train_acc = None
self.best_model_val_acc = None
self.reporting_freq = reporting_freq
self.bokeh_output = bokeh_output
if bokeh_output == 'notebook':
bokeh.plotting.output_notebook(hide_banner=True)
elif bokeh_output == 'server':
Print.printflush(
'\nConnecting to Bokeh Server for live Learning Curves plotting...\n')
bokeh.plotting.output_server('')
self.bokeh_session = bokeh.plotting.cursession()
try:
self.fig = \
bokeh.plotting.figure(
title=plot_title,
x_axis_label='# of Training Data Batches',
y_axis_label='Loss',
plot_height=680,
plot_width=880)
self.fig_train_losses_data_source = \
bokeh.models.ColumnDataSource(
data=dict(
batches=self.batches,
train_losses=self.train_losses))
self.fig.line(
x='batches',
y='train_losses',
source=self.fig_train_losses_data_source,
name='TrainLoss',
legend='Training Loss')
self.fig_val_losses_data_source = \
bokeh.models.ColumnDataSource(
data=dict(
batches=self.batches,
val_losses=self.val_losses))
self.fig.circle(
x='batches',
y='val_losses',
source=self.fig_val_losses_data_source,
name='ValidLoss',
legend='Validation Loss',
color='red')
bokeh.plotting.show(self.fig)
except:
Print.printflush(
'\nBokeh Notebook/Server Connection *FAILED!*')
Print.printflush(
'Please make sure Bokeh package is already installed in Python, and')
Print.printflush(
'please open a new Command-Line Terminal window\n (separate from this Terminal window)')
Print.printflush(
' and run the following command firs to launch Bokeh Server:')
Print.printflush(
' bokeh-server --backend=memory\n')
os._exit(0)
def on_train_begin(self, logs={}):
Print.printflush('\nFFNN Training Progress')
Print.printflush('______________________')
def __init__(self,
reporting_freq=False,
plot_title='Neural Network Learning Curves',
bokeh_output='notebook'):
super(callbacks.Callback, self).__init__()
self.latest_epoch = -1
self.latest_batch = -1
self.batches = []
self.train_losses = []
self.approx_train_acc_in_latest_epoch = 0.
self.val_losses = []
self.latest_val_acc = None
self.min_val_loss = numpy.inf
self.best_model = None
self.best_model_epoch = None
self.best_model_train_acc = None
self.best_model_val_acc = None
self.reporting_freq = reporting_freq
self.bokeh_output = bokeh_output
if bokeh_output == 'notebook':
bokeh.plotting.output_notebook(hide_banner=True)
elif bokeh_output == 'server':
Print.printflush(
'\nConnecting to Bokeh Server for live Learning Curves plotting...\n'
)
bokeh.plotting.output_server('')
self.bokeh_session = bokeh.plotting.cursession()
try:
self.fig = \
bokeh.plotting.figure(
title=plot_title,
x_axis_label='# of Training Data Batches',
y_axis_label='Loss',
plot_height=680,
plot_width=880)
self.fig_train_losses_data_source = \
bokeh.models.ColumnDataSource(
data=dict(
batches=self.batches,
train_losses=self.train_losses))
self.fig.line(x='batches',
y='train_losses',
source=self.fig_train_losses_data_source,
name='TrainLoss',
legend='Training Loss')
self.fig_val_losses_data_source = \
bokeh.models.ColumnDataSource(
data=dict(
batches=self.batches,
val_losses=self.val_losses))
self.fig.circle(x='batches',
y='val_losses',
source=self.fig_val_losses_data_source,
name='ValidLoss',
legend='Validation Loss',
color='red')
bokeh.plotting.show(self.fig)
except:
Print.printflush('\nBokeh Notebook/Server Connection *FAILED!*')
Print.printflush(
'Please make sure Bokeh package is already installed in Python, and'
)
Print.printflush(
'please open a new Command-Line Terminal window\n (separate from this Terminal window)'
)
Print.printflush(
' and run the following command firs to launch Bokeh Server:'
)
Print.printflush(' bokeh-server --backend=memory\n')
os._exit(0)
