Problem with KerasClassifier #112
Comments
|
Thank for for the report and the kind words! I haven't had a chance to test it (I will do so in a couple of hours), but I think this may be the same bug/issue as #78. If so, it should be fixed by #88, which should happen soon. In the meantime, maybe using |
|
Worked perfectly. Thank you very much for your attention. |
|
As of last night, it should be fixed in master. That said, there are a couple of changes I would make:
model = KerasClassifier(model=build_model, verbose=1)Also, SciKeras can now compile models for you, which allows you to do hyperparameter tuning of optimizer learning rates, for example: def build_model(optimizer): # defaults no longer needed
model = Sequential()
model.add(Flatten(input_shape=(28, 28)))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(10, activation='softmax'))
return model # return an un-compiled model
model = KerasClassifier(
model=build_model,
verbose=1,
loss="sparse_categorical_crossentropy", # you need to specify the loss here
optimizer__learning_rate=0.02, # no optimizer specified, default is rmsprop
)
grid = GridSearchCV(estimator=model, param_grid={"optimizer__learning_rate": [0.1, 0.01, 0.0001]})
result = grid.fit(X, y) |
|
I am still evaluating SciKeras with the |
|
Hi @adriangb, I applied the suggested updates and they worked (Version: 0.2.0). With the Now with |
|
Yes, that is because SciKeras no longer introspects your model's loss function. To make
def build_model():
model = Sequential()
model.add(Flatten(input_shape=(28, 28)))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(10, activation='softmax'))
return model
model = KerasClassifier(
model=build_model,
loss="categorical_crossentropy", # pass categorical_crossentropy
)This applies even if you are compiling your own model: def build_model():
model = Sequential()
model.add(Flatten(input_shape=(28, 28)))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(10, activation='softmax'))
model.compile(
loss="categorical_crossentropy", # loss matches constructor loss
optimizer="sgd", # override default optimizer
metrics=["accuracy"] # add a metric
)
return model
model = KerasClassifier(
model=build_model,
loss="categorical_crossentropy", # pass categorical_crossentropy
) |
|
I'll close this issue since it seems like it's resolved. Let me know if there is any followup. Thanks for the bug report! |
Hi @adriangb,
congratulations for the work.
I am having a problem with a dataset that I am using in my work.
KerasRegressoris working perfectly with me, butKerasClassifierpresented a problem that did not happen withkeras.wrappers.scikit_learn.I took a simple example with the mnist dataset and the problem persisted.
If I change the import
scikerastokeras.wrappers.scikit_learnthe code works perfectly.Shapes of X and y:
X(70000, 28, 28)y(70000,)Thanks a lot for attention.
The text was updated successfully, but these errors were encountered: