CODEWITHSUNDEEP
pythontensorflowkerasscikit-learn
IzzyGiessen
IzzyGiessen

Reputation: 31

"ValueError: activation is not a legal parameter" with Keras classifier

I've been playing around with Tensorflow and Keras and I finally got the following error while trying hyper parameter tuning: "ValueError: activation is not a legal parameter"

The point is that I want to try different activation functions in my model to see which one works best. I have the following code:

import pandas as pd 
import tensorflow as tf 
from keras.wrappers.scikit_learn import KerasClassifier
from sklearn.model_selection import GridSearchCV
import numpy as np

ds = pd.read_csv(
    "https://storage.googleapis.com/download.tensorflow.org/data/abalone_train.csv",
    names=["Length", "Diameter", "Height", "Whole weight", "Shucked weight",
           "Viscera weight", "Shell weight", "Age"])
print(ds)

x_train = ds.copy()
y_train = x_train.pop('Age')
x_train = np.array(x_train)

def create_model(layers, activations):
    model = tf.keras.Sequential()
    for i, nodes in enumerate(layers):
        if i == 0:
            model.add(tf.keras.layers.Dense(nodes, input_dim=x_train.shape[1]))
            model.add(layers.Activation(activations))
            model.add(Dropout(0.3))
        else:
            model.add(tf.keras.layers.Dense(nodes))
            model.add(layers.Activation(activations))
            model.add(Dropout(0.3))

    model.add(tf.keras.layers.Dense(units=1, kernel_initializer='glorot_uniform'))
    model.add(layers.Activation('sigmoid'))
    model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
    return model

model = KerasClassifier(build_fn=create_model, verbose=0)

layers = [[20], [40,20], [45, 30, 15]]
activations = ['sigmoid', 'relu']
param_grid = dict(layers=layers, activation=activations, batch_size = [128, 256], epochs=[30])
grid = GridSearchCV(estimator=model, param_grid=param_grid, cv=5)

grid_result = grid.fit(x_train, y_train)

print(grid_result.best_score_,grid_reslult.best_params_)

pred_y = grid.predict(x_test)
y_pred = (pred_y > 0.5)

cm=confusion_matrix(y_pred, y_test)
score=accuracy_score(y_pred, y_test) 

model.fit(x_train, y_train, epochs=30, callbacks=[cp_callback])
#steps_per_epoch
model.evaluate(x_test, y_test, verbose=2)

probability_model = tf.keras.Sequential([
    model,
    tf.keras.layers.Softmax()
]) 

probability_model(x_test[:100])

Upvotes: 3

Views: 3642

Answers (2)

George
George

Reputation: 5691

If you see here, you must specify activations as :

from tensorflow.keras import activations layers.Activation(activations.relu)

Right now, you have:

activations = ['sigmoid', 'relu']

So , that's why the value error.

You should change your code to sth like this:

model.add(tf.keras.layers.Dense(nodes, activation=activations[i], input_dim=x_train.shape[1]))

So, remove the Activation layer: model.add(layers.Activation(activations)) and instead use the activation inside each layer.

Example:

def create_model(layers, activations):
    model = tf.keras.Sequential()
    for i in range(2):
        if i == 0:
            model.add(tf.keras.layers.Dense(2, activation=activations[i], input_dim=x_train.shape[1]))
            model.add(tf.keras.layers.Dropout(0.3))
        else:
            model.add(tf.keras.layers.Dense(2, activation=activations[i]))
            model.add(tf.keras.layers.Dropout(0.3))

    model.add(tf.keras.layers.Dense(units=1, activation='sigmoid', kernel_initializer='glorot_uniform'))
    model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
    return model

Upvotes: 1

jxcvi
jxcvi

Reputation: 19

layers.Activation() expects a function or a string, such as 'sigmoid' but you are currently passing an array activations to it. Use your index i (or a different index) to access the activation function like activations[i].

You can also pass the activation as string directly to the Dense layer like so:

 model.add(tf.keras.layers.Dense(nodes, activation=activations[i], input_dim=x_train.shape[1])))

Upvotes: 0

Related Questions