I want to know feature importance for the same reason i want to give my model as input to permutation_importance( ) function, which wants model to give output as 0 or 1 only.
My dataset outcome is binary, I want Output of ANN to be either 0 or 1, but i am getting probabilities in the range 0 to 1.
I tried to customise step activation function for output layer , but not much luck.
This is my ANN
model = keras.models.Sequential() model.add(keras.layers.Dense(32,input_dim=a,activation='relu')) ## input layer(8 neurons) --> hidden layer(32) model.add(keras.layers.Dropout(0.2)) model.add(keras.layers.Dense(64,activation="relu")) ## second hidden layer model.add(keras.layers.Dropout(0.2)) model.add(keras.layers.Dense(2,activation="softmax")) ## output layer
model.compile(loss="binary_crossentropy",optimizer='adam',metrics=["accuracy"])
Kindly suggest how can i get output as 0 or 1 for this ANN
Khalid Youssef and Vadym Slyusar
Thanks for the response
I am trying to add a layer for classification with customised activation function as mentioned below ,
but getting error that " TypeError: To be compatible with tf.eager.defun, Python functions must return zero or more Tensors; in compilation of , found return value of type , which is not a Tensor. "
Kindly suggest for to proceed
@tf.function def custom_activation(x): aa = tf.constant(1.0) bb = tf.constant(0.0) def grad(dy): return dy if (h > 0.5): return aa, grad else: return bb, grad
model = keras.models.Sequential() model.add(keras.layers.Dense(32,input_dim=a,activation='relu')) ## input layer(8 neurons) --> hidden layer(32) model.add(keras.layers.Dropout(0.2)) model.add(keras.layers.Dense(64,activation="relu")) ## second hidden layer model.add(keras.layers.Dropout(0.2)) model.add(keras.layers.Dense(1,activation="softmax")) ## output layer model.add(keras.layers.Dense(1,activation="custom_activation")) ### Compile the model model.compile(loss="binary_crossentropy",optimizer='adam',metrics=["accuracy"])
model.fit(train_feats,train_labels,epochs=500)
Thanks in advance
Dear Himani Deshpande!
You made a mistake on the output layer: You should have one neuron at the output, with the sigmoid activation function. Softmax is for probabilistic problems. You should also convert the output signal as a single column with values of 0 or 1.
Y_train - your output signal should be in the form of a single column, inside which contains 0 or 1.
I corrected your code.
- Badges
- Science topic