Skip to content

Commit

Permalink
Merge bd8a027 into 229d5dd
Browse files Browse the repository at this point in the history
  • Loading branch information
@timokau
timokau committed Nov 25, 2019
2 parents 229d5dd + bd8a027 commit c0f4c0e
Show file tree
Hide file tree
Showing 114 changed files with 5,972 additions and 2,362 deletions.
41 changes: 24 additions & 17 deletions csrank/callbacks.py
View file
Expand Up @@ -43,9 +43,10 @@ def on_epoch_end(self, epoch, logs=None):
self.best_weights = self.model.get_weights()
if current is None:
warnings.warn(
'Early stopping conditioned on metric `%s` '
'which is not available. Available metrics are: %s' %
(self.monitor, ','.join(list(logs.keys()))), RuntimeWarning
"Early stopping conditioned on metric `%s` "
"which is not available. Available metrics are: %s"
% (self.monitor, ",".join(list(logs.keys()))),
RuntimeWarning,
)
return
if self.monitor_op(current - self.min_delta, self.best):
Expand All @@ -60,16 +61,20 @@ def on_epoch_end(self, epoch, logs=None):

def on_train_end(self, logs=None):
if self.stopped_epoch > 0:
self.logger.info("Setting best weights for final epoch {}".format(self.epoch))
self.logger.info(
"Setting best weights for final epoch {}".format(self.epoch)
)
self.model.set_weights(self.best_weights)

def set_tunable_parameters(self, patience=300, min_delta=2, **point):
self.patience = patience
self.min_delta = min_delta
if len(point) > 0:
self.logger.warning('This callback does not support'
' tunable parameters'
' called: {}'.format(print_dictionary(point)))
self.logger.warning(
"This callback does not support"
" tunable parameters"
" called: {}".format(print_dictionary(point))
)


class weightHistory(Callback):
Expand All @@ -79,12 +84,13 @@ def on_train_begin(self, logs={}):
self.hidden_units_used = []

def on_batch_end(self, batch, logs={}):
hidden = [layer for layer in self.model.layers
if layer.name == 'hidden_1']
hidden = [layer for layer in self.model.layers if layer.name == "hidden_1"]

y = np.array(hidden[0].get_weights()[0])
close = np.isclose(y, 0, atol=1e-3)
self.hidden_units_used.append(len(np.unique(np.where(np.logical_not(close))[1])))
self.hidden_units_used.append(
len(np.unique(np.where(np.logical_not(close))[1]))
)
self.norm.append(np.abs(y).sum())
self.zero_weights.append(close.sum())

Expand Down Expand Up @@ -113,25 +119,26 @@ def set_tunable_parameters(self, epochs_drop=300, drop=0.1, **point):
self.epochs_drop = epochs_drop
self.drop = drop
if len(point) > 0:
self.logger.warning('This callback does not support'
' tunable parameters'
' called: {}'.format(print_dictionary(point)))
self.logger.warning(
"This callback does not support"
" tunable parameters"
" called: {}".format(print_dictionary(point))
)


class DebugOutput(Callback):

def __init__(self, delta=100, **kwargs):
super(DebugOutput, self).__init__(**kwargs)
self.delta = delta

def on_train_end(self, logs=None):
self.logger.debug('Total number of epochs: {}'.format(self.epoch))
self.logger.debug("Total number of epochs: {}".format(self.epoch))

def on_train_begin(self, logs=None):
self.epoch = 0
self.logger = logging.getLogger('DebugOutput')
self.logger = logging.getLogger("DebugOutput")

def on_epoch_end(self, epoch, logs=None):
self.epoch += 1
if self.epoch % self.delta == 0:
self.logger.debug('Epoch {} of the training finished.'.format(self.epoch))
self.logger.debug("Epoch {} of the training finished.".format(self.epoch))
4 changes: 3 additions & 1 deletion csrank/choicefunction/baseline.py
View file
Expand Up @@ -47,7 +47,9 @@ def predict_scores(self, X, Y, **kwargs):
if isinstance(X, dict):
scores = dict()
for ranking_size, x in X.items():
scores[ranking_size] = self._predict_scores_fixed(x, Y[ranking_size], **kwargs)
scores[ranking_size] = self._predict_scores_fixed(
x, Y[ranking_size], **kwargs
)

else:
scores = self._predict_scores_fixed(X, **kwargs)
Expand Down
11 changes: 6 additions & 5 deletions csrank/choicefunction/choice_functions.py
View file
Expand Up @@ -6,11 +6,10 @@
from csrank.metrics_np import f1_measure
from csrank.util import progress_bar

__all__ = ['ChoiceFunctions']
__all__ = ["ChoiceFunctions"]


class ChoiceFunctions(metaclass=ABCMeta):

@property
def learning_problem(self):
return CHOICE_FUNCTION
Expand Down Expand Up @@ -60,9 +59,11 @@ def _tune_threshold(self, X_val, Y_val, thin_thresholds=1, verbose=0):
threshold = p
best = f1
if verbose == 1:
progress_bar(i, len(probabilities), status='Tuning threshold')
progress_bar(i, len(probabilities), status="Tuning threshold")
except KeyboardInterrupt:
self.logger.info("Keyboard interrupted")
self.logger.info('Tuned threshold, obtained {:.2f} which achieved'
' a micro F1-measure of {:.2f}'.format(threshold, best))
self.logger.info(
"Tuned threshold, obtained {:.2f} which achieved"
" a micro F1-measure of {:.2f}".format(threshold, best)
)
return threshold
111 changes: 86 additions & 25 deletions csrank/choicefunction/cmpnet_choice.py
View file
Expand Up @@ -10,10 +10,22 @@


class CmpNetChoiceFunction(CmpNetCore, ChoiceFunctions):
def __init__(self, n_object_features, n_hidden=2, n_units=8, loss_function='binary_crossentropy',
batch_normalization=True, kernel_regularizer=l2(l=1e-4), kernel_initializer='lecun_normal',
activation='relu', optimizer=SGD(lr=1e-4, nesterov=True, momentum=0.9), metrics=['binary_accuracy'],
batch_size=256, random_state=None, **kwargs):
def __init__(
self,
n_object_features,
n_hidden=2,
n_units=8,
loss_function="binary_crossentropy",
batch_normalization=True,
kernel_regularizer=l2(l=1e-4),
kernel_initializer="lecun_normal",
activation="relu",
optimizer=SGD(lr=1e-4, nesterov=True, momentum=0.9),
metrics=["binary_accuracy"],
batch_size=256,
random_state=None,
**kwargs
):
"""
Create an instance of the :class:`CmpNetCore` architecture for learning a choice function.
CmpNet breaks the preferences in form of rankings into pairwise comparisons and learns a pairwise model for
Expand Down Expand Up @@ -67,32 +79,58 @@ def __init__(self, n_object_features, n_hidden=2, n_units=8, loss_function='bina
[1] Leonardo Rigutini, Tiziano Papini, Marco Maggini, and Franco Scarselli. 2011. SortNet: Learning to Rank by a Neural Preference Function. IEEE Trans. Neural Networks 22, 9 (2011), 1368–1380. https://doi.org/10.1109/TNN.2011.2160875
"""
super().__init__(n_object_features=n_object_features, n_hidden=n_hidden, n_units=n_units,
loss_function=loss_function, batch_normalization=batch_normalization,
kernel_regularizer=kernel_regularizer, kernel_initializer=kernel_initializer,
activation=activation, optimizer=optimizer, metrics=metrics, batch_size=batch_size,
random_state=random_state, **kwargs)
super().__init__(
n_object_features=n_object_features,
n_hidden=n_hidden,
n_units=n_units,
loss_function=loss_function,
batch_normalization=batch_normalization,
kernel_regularizer=kernel_regularizer,
kernel_initializer=kernel_initializer,
activation=activation,
optimizer=optimizer,
metrics=metrics,
batch_size=batch_size,
random_state=random_state,
**kwargs
)
self.logger = logging.getLogger(CmpNetChoiceFunction.__name__)
self.logger.info("Initializing network with object features {}".format(self.n_object_features))
self.logger.info(
"Initializing network with object features {}".format(
self.n_object_features
)
)
self.threshold = 0.5

def _convert_instances_(self, X, Y):
self.logger.debug('Creating the Dataset')
self.logger.debug("Creating the Dataset")
x1, x2, garbage, y_double, garbage = generate_complete_pairwise_dataset(X, Y)
del garbage
if x1.shape[0] > self.threshold_instances:
indices = self.random_state.choice(x1.shape[0], self.threshold_instances, replace=False)
indices = self.random_state.choice(
x1.shape[0], self.threshold_instances, replace=False
)
x1 = x1[indices, :]
x2 = x2[indices, :]
y_double = y_double[indices, :]
self.logger.debug('Finished the Dataset instances {}'.format(x1.shape[0]))
self.logger.debug("Finished the Dataset instances {}".format(x1.shape[0]))
return x1, x2, y_double

def construct_model(self):
return super().construct_model()

def fit(self, X, Y, epochs=10, callbacks=None, validation_split=0.1, tune_size=0.1, thin_thresholds=1, verbose=0,
**kwd):
def fit(
self,
X,
Y,
epochs=10,
callbacks=None,
validation_split=0.1,
tune_size=0.1,
thin_thresholds=1,
verbose=0,
**kwd
):
"""
Fit a CmptNet model for learning a choice fucntion on the provided set of queries X and preferences Y of
those objects. The provided queries and corresponding preferences are of a fixed size (numpy arrays). For
Expand Down Expand Up @@ -130,15 +168,25 @@ def fit(self, X, Y, epochs=10, callbacks=None, validation_split=0.1, tune_size=0
Keyword arguments for the fit function
"""
if tune_size > 0:
X_train, X_val, Y_train, Y_val = train_test_split(X, Y, test_size=tune_size, random_state=self.random_state)
X_train, X_val, Y_train, Y_val = train_test_split(
X, Y, test_size=tune_size, random_state=self.random_state
)
try:
super().fit(X_train, Y_train, epochs, callbacks,
validation_split, verbose, **kwd)
super().fit(
X_train,
Y_train,
epochs,
callbacks,
validation_split,
verbose,
**kwd
)
finally:
self.logger.info('Fitting utility function finished. Start tuning threshold.')
self.logger.info(
"Fitting utility function finished. Start tuning threshold."
)
else:
super().fit(X, Y, epochs, callbacks, validation_split, verbose,
**kwd)
super().fit(X, Y, epochs, callbacks, validation_split, verbose, **kwd)
self.threshold = 0.5

def _predict_scores_fixed(self, X, **kwargs):
Expand All @@ -156,7 +204,20 @@ def predict(self, X, **kwargs):
def clear_memory(self, **kwargs):
super().clear_memory(**kwargs)

def set_tunable_parameters(self, n_hidden=32, n_units=2, reg_strength=1e-4, learning_rate=1e-3, batch_size=128,
**point):
super().set_tunable_parameters(n_hidden=n_hidden, n_units=n_units, reg_strength=reg_strength,
learning_rate=learning_rate, batch_size=batch_size, **point)
def set_tunable_parameters(
self,
n_hidden=32,
n_units=2,
reg_strength=1e-4,
learning_rate=1e-3,
batch_size=128,
**point
):
super().set_tunable_parameters(
n_hidden=n_hidden,
n_units=n_units,
reg_strength=reg_strength,
learning_rate=learning_rate,
batch_size=batch_size,
**point
)

0 comments on commit c0f4c0e

Please sign in to comment.