def evaluate(self, x, y):
#Define evaluate sequence
'''
For predict for Keras see line 56 in https://github.com/gradescope/fsdl-text-recognizer-project/blob/master/lab6_sln/text_recognizer/models/base.py
'''
sequence = DatasetSequence(x, y, batch_size=12)
preds = self.algorithm.predict(sequence.x)
report = metrics.classification_report(sequence.y, preds)
return report
def fit(self, dataset, batch_size=32, epochs=10, callbacks=[]):
self.algorithm.compile(loss=self.loss(),
optimizer=self.optimizer(),
metrics=self.metrics())
train_sequence = DatasetSequence(dataset.x_train,
dataset.y_train,
batch_size,
augment_fn=self.batch_augment_fn,
format_fn=self.batch_format_fn)
test_sequence = DatasetSequence(dataset.x_test,
dataset.y_test,
batch_size,
augment_fn=self.batch_augment_fn,
format_fn=self.batch_format_fn)
self.algorithm.fit_generator(train_sequence,
epochs=epochs,
callbacks=callbacks,
validation_data=test_sequence,
use_multiprocessing=True,
workers=1,
shuffle=True)
def fit(self,
dataset,
batch_size=32,
epochs=10,
callbacks=[],
lr: Optional[float] = 0.001,
beta_1: Optional[float] = 0.9,
beta_2: Optional[float] = 0.999,
epsilon=None,
decay=0.0):
self.algorithm.compile(loss=self.loss(),
optimizer=self.adam_optimizer(lr=lr,
beta_1=beta_1,
beta_2=beta_2,
epsilon=epsilon,
decay=decay),
metrics=self.metrics())
train_sequence = DatasetSequence(dataset.x_train,
dataset.y_train,
batch_size,
augment_fn=self.batch_augment_fn,
format_fn=self.batch_format_fn)
test_sequence = DatasetSequence(dataset.x_test,
dataset.y_test,
batch_size,
augment_fn=self.batch_augment_fn,
format_fn=self.batch_format_fn)
self.algorithm.fit_generator(train_sequence,
epochs=epochs,
callbacks=callbacks,
validation_data=test_sequence,
use_multiprocessing=True,
workers=1,
shuffle=True)
def fit(self, dataset, batch_size=None, epochs=None, callbacks=[]):
#Define fit sequence
'''
Fit generator for keras. See line 44 https://github.com/gradescope/fsdl-text-recognizer-project/blob/master/lab6_sln/text_recognizer/models/base.py
Arguments for fit generator
train_sequence,
epochs = epochs,
callbacks = callbacks,
validation_data = test_sequence,
use_multiprocessing = False,
workers = 1,
shuffle = True
'''
#Updated for sklearn
train_sequence = DatasetSequence(dataset.x_train, dataset.y_train,
batch_size)
self.algorithm.fit(train_sequence.x, train_sequence.y)
def evaluate(self, x, y):
sequence = DatasetSequence(
x, y, batch_size=16) # Use a small batch size to use less memory
preds = self.algorithm.predict_generator(sequence)
return np.mean(np.argmax(preds, -1) == np.argmax(y, -1))
def evaluate(self, x, y):
sequence = DatasetSequence(x, y, batch_size=12)
preds = self.algorithm.predict(sequence.x)
report = metrics.classification_report(sequence.y, preds)
return report