Show progress bar for each epoch during batchwise training in Keras
Question:
When I load the whole dataset in memory and train the network in Keras using following code:
model.fit(X, y, nb_epoch=40, batch_size=32, validation_split=0.2, verbose=1)
This generates a progress bar per epoch with metrics like ETA, accuracy, loss, etc
When I train the network in batches, I’m using the following code
for e in range(40):
for X, y in data.next_batch():
model.fit(X, y, nb_epoch=1, batch_size=data.batch_size, verbose=1)
This will generate a progress bar for each batch instead of each epoch. Is it possible to generate a progress bar for each epoch during batchwise training?
Answers:
model.fit(X, y, nb_epoch=40, batch_size=32, validation_split=0.2, verbose=1)
In the above change to verbose=2, as it is mentioned in the documentation:
verbose: 0 for no logging to stdout, 1 for progress bar logging, 2 for one log line per epoch
It’ll show your output as:
Epoch 1/100
0s - loss: 0.2506 - acc: 0.5750 - val_loss: 0.2501 - val_acc: 0.3750
Epoch 2/100
0s - loss: 0.2487 - acc: 0.6250 - val_loss: 0.2498 - val_acc: 0.6250
Epoch 3/100
0s - loss: 0.2495 - acc: 0.5750 - val_loss: 0.2496 - val_acc: 0.6250
.....
.....
If you want to show a progress bar for completion of epochs, keep verbose=0 (which shuts out logging to stdout) and implement in the following manner:
from time import sleep
import sys
epochs = 10
for e in range(epochs):
sys.stdout.write('r')
for X, y in data.next_batch():
model.fit(X, y, nb_epoch=1, batch_size=data.batch_size, verbose=0)
# print loss and accuracy
# the exact output you're looking for:
sys.stdout.write("[%-60s] %d%%" % ('='*(60*(e+1)/10), (100*(e+1)/10)))
sys.stdout.flush()
sys.stdout.write(", epoch %d"% (e+1))
sys.stdout.flush()
The output will be as follows:
[============================================================] 100%, epoch 10
If you want to show loss after every n batches, you can use:
out_batch = NBatchLogger(display=1000)
model.fit([X_train_aux,X_train_main],Y_train,batch_size=128,callbacks=[out_batch])
Though, I haven’t ever tried it before. The above example was taken from this keras github issue: Show Loss Every N Batches #2850
You can also follow a demo of NBatchLogger here:
class NBatchLogger(Callback):
def __init__(self, display):
self.seen = 0
self.display = display
def on_batch_end(self, batch, logs={}):
self.seen += logs.get('size', 0)
if self.seen % self.display == 0:
metrics_log = ''
for k in self.params['metrics']:
if k in logs:
val = logs[k]
if abs(val) > 1e-3:
metrics_log += ' - %s: %.4f' % (k, val)
else:
metrics_log += ' - %s: %.4e' % (k, val)
print('{}/{} ... {}'.format(self.seen,
self.params['samples'],
metrics_log))
You can also use progbar for progress, but it’ll print progress batchwise
from keras.utils import generic_utils
progbar = generic_utils.Progbar(X_train.shape[0])
for X_batch, Y_batch in datagen.flow(X_train, Y_train):
loss, acc = model_test.train([X_batch]*2, Y_batch, accuracy=True)
progbar.add(X_batch.shape[0], values=[("train loss", loss), ("acc", acc)])
you can set verbose=0 and set callbacks that will update progress at the end of each fitting,
clf.fit(X, y, nb_epoch=1, batch_size=data.batch_size, verbose=0, callbacks=[some_callback])
https://keras.io/callbacks/#example-model-checkpoints
or set callback https://keras.io/callbacks/#remotemonitor
tqdm (version >= 4.41.0) has also just added built-in support for keras so you could do:
from tqdm.keras import TqdmCallback
...
model.fit(..., verbose=0, callbacks=[TqdmCallback(verbose=2)])
This turns off keras‘ progress (verbose=0), and uses tqdm instead. For the callback, verbose=2 means separate progressbars for epochs and batches. 1 means clear batch bars when done. 0 means only show epochs (never show batch bars).
When I load the whole dataset in memory and train the network in Keras using following code:
model.fit(X, y, nb_epoch=40, batch_size=32, validation_split=0.2, verbose=1)
This generates a progress bar per epoch with metrics like ETA, accuracy, loss, etc
When I train the network in batches, I’m using the following code
for e in range(40):
for X, y in data.next_batch():
model.fit(X, y, nb_epoch=1, batch_size=data.batch_size, verbose=1)
This will generate a progress bar for each batch instead of each epoch. Is it possible to generate a progress bar for each epoch during batchwise training?
model.fit(X, y, nb_epoch=40, batch_size=32, validation_split=0.2, verbose=1)
In the above change to verbose=2, as it is mentioned in the documentation:
verbose: 0 for no logging to stdout, 1 for progress bar logging, 2 for one log line per epoch
It’ll show your output as:
Epoch 1/100
0s - loss: 0.2506 - acc: 0.5750 - val_loss: 0.2501 - val_acc: 0.3750
Epoch 2/100
0s - loss: 0.2487 - acc: 0.6250 - val_loss: 0.2498 - val_acc: 0.6250
Epoch 3/100
0s - loss: 0.2495 - acc: 0.5750 - val_loss: 0.2496 - val_acc: 0.6250
.....
.....
If you want to show a progress bar for completion of epochs, keep verbose=0 (which shuts out logging to stdout) and implement in the following manner:
from time import sleep
import sys
epochs = 10
for e in range(epochs):
sys.stdout.write('r')
for X, y in data.next_batch():
model.fit(X, y, nb_epoch=1, batch_size=data.batch_size, verbose=0)
# print loss and accuracy
# the exact output you're looking for:
sys.stdout.write("[%-60s] %d%%" % ('='*(60*(e+1)/10), (100*(e+1)/10)))
sys.stdout.flush()
sys.stdout.write(", epoch %d"% (e+1))
sys.stdout.flush()
The output will be as follows:
[============================================================] 100%, epoch 10
If you want to show loss after every n batches, you can use:
out_batch = NBatchLogger(display=1000)
model.fit([X_train_aux,X_train_main],Y_train,batch_size=128,callbacks=[out_batch])
Though, I haven’t ever tried it before. The above example was taken from this keras github issue: Show Loss Every N Batches #2850
You can also follow a demo of NBatchLogger here:
class NBatchLogger(Callback):
def __init__(self, display):
self.seen = 0
self.display = display
def on_batch_end(self, batch, logs={}):
self.seen += logs.get('size', 0)
if self.seen % self.display == 0:
metrics_log = ''
for k in self.params['metrics']:
if k in logs:
val = logs[k]
if abs(val) > 1e-3:
metrics_log += ' - %s: %.4f' % (k, val)
else:
metrics_log += ' - %s: %.4e' % (k, val)
print('{}/{} ... {}'.format(self.seen,
self.params['samples'],
metrics_log))
You can also use progbar for progress, but it’ll print progress batchwise
from keras.utils import generic_utils
progbar = generic_utils.Progbar(X_train.shape[0])
for X_batch, Y_batch in datagen.flow(X_train, Y_train):
loss, acc = model_test.train([X_batch]*2, Y_batch, accuracy=True)
progbar.add(X_batch.shape[0], values=[("train loss", loss), ("acc", acc)])
you can set verbose=0 and set callbacks that will update progress at the end of each fitting,
clf.fit(X, y, nb_epoch=1, batch_size=data.batch_size, verbose=0, callbacks=[some_callback])
https://keras.io/callbacks/#example-model-checkpoints
or set callback https://keras.io/callbacks/#remotemonitor
tqdm (version >= 4.41.0) has also just added built-in support for keras so you could do:
from tqdm.keras import TqdmCallback
...
model.fit(..., verbose=0, callbacks=[TqdmCallback(verbose=2)])
This turns off keras‘ progress (verbose=0), and uses tqdm instead. For the callback, verbose=2 means separate progressbars for epochs and batches. 1 means clear batch bars when done. 0 means only show epochs (never show batch bars).