Is possible to use sklearn.neighbors.KNeighborsClassifier into a tensorflow Session i.e with Tensor?-CodePudding

I am trying to use the KNN classifier inside a Tensorflow session.

But I am getting the following error:

NotImplementedError: Cannot convert a symbolic Tensor (Const:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported

Outside the session the code works just fine:

import tensorflow as tf
from sklearn.neighbors import KNeighborsClassifier
features= tf.constant([[1., 1.], [2., 2.],[2., 2.],[2., 2.],[2., 2.],[2., 2.]])
label= tf.constant([[1], [2], [2], [2], [2], [2]])
model = KNeighborsClassifier(n_neighbors=3)

# Train the model using the training sets
model.fit(features,label)

teste = tf.constant([[1., 1.], [2., 2.]])
#Predict Output
predicted= model.predict(teste) # 0:Overcast, 2:Mild
print(predicted)

But I need it inside the Session, here is a code with example of the error:

import tensorflow as tf
from sklearn.neighbors import KNeighborsClassifier
@tf.function
def add():
    model = KNeighborsClassifier(n_neighbors=3)


    features= tf.constant([[1., 1.], [2., 2.],[2., 2.],[2., 2.],[2., 2.],[2., 2.]])
    label= tf.constant([[1], [2], [2], [2], [2], [2]])
    # Train the model using the training sets
    model.fit(features,label)
    
    return model

add()

Versions :

tf.version.VERSION
'2.6.0'
sklearn.__version__
1.0.1

CodePudding user response：

This code may help you solve your problem.

import tensorflow as tf
from sklearn.neighbors import KNeighborsClassifier
tf.config.run_functions_eagerly(True)
@tf.function
def add():
    
    model = KNeighborsClassifier(n_neighbors=3)


    features= tf.constant([[1., 1.], [2., 2.],[2., 2.],[2., 2.],[2., 2.],[2., 2.]])
    label= tf.constant([[1], [2], [2], [2], [2], [2]])
    features = features.numpy()
    label = label.numpy()
    # Train the model using the training sets
    model.fit(features,label)
    
    return model

add()

I have run the code in Google Colab. Downgraded NumPy to 1.19.5

Note:

.numpy() changes the tensor to numpy array.
Tensorflow 2 has a config option to run functions "eagerly" which will enable getting Tensor values via .numpy() method. 3rd line of the code [Without that line, the .numpy() will not work since @tf.function decorator prohibits the execution of functions like tensor.numpy() for performance reasons.