Keras scikit-learn wrapper not compatible with keras functional model #34689
Description
Please make sure that this is a bug. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:bug_template
System information
- TensorFlow version (use command below): TF version: 2.0, 1.15
- Python version: 3.6 , 3.7
Describe the current behavior
when using a keras functional api model via the keras scikit-learn wrapper a crash occurs.
see:
Describe the expected behavior
model.predict() should work on all keras model types besides sequential
Code to reproduce the issue
import numpy as np
import tensorflow as tf
import tensorflow.keras
from tensorflow.keras.wrappers.scikit_learn import KerasClassifier
def build_model():
input = tf.keras.layers.Input(shape=(2,))
pred = tf.keras.layers.Dense(2, activation='softmax')(input)
model = tf.keras.models.Model(inputs=input, outputs=pred)
model.compile(loss='categorical_crossentropy', metrics=['accuracy'])
return model
X = np.array([[1,2],[3,1]])
Y = np.array([[1,0], [0,1]])
model = build_model()
model.fit(X, Y)
print(model.predict(X)) # this works
model_wrapped = KerasClassifier(build_model)
model_wrapped.fit(X, Y)
model_wrapped.predict(X) # this crashesOutput:
Train on 2 samples
2/2 [==============================] - 0s 62ms/sample - loss: 1.1024 - acc: 0.5000
[[0.62487346 0.37512657]
[0.8205698 0.17943017]]
Train on 2 samples
2/2 [==============================] - 0s 64ms/sample - loss: 0.2733 - acc: 1.0000
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-14-48bacae97b80> in <module>()
19 model_wrapped = KerasClassifier(build_model)
20 model_wrapped.fit(X, Y)
---> 21 model_wrapped.predict(X) # this crashes
22
23
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/wrappers/scikit_learn.py in predict(self, x, **kwargs)
239 """
240 kwargs = self.filter_sk_params(Sequential.predict_classes, kwargs)
--> 241 classes = self.model.predict_classes(x, **kwargs)
242 return self.classes_[classes]
243
AttributeError: 'Model' object has no attribute 'predict_classes'