How to predict classes for binary classification using Sub classing method using TensorFlow?

Issue

I am doing a binary classification using sub classing using Tensorflow. My code is:

class ChurnClassifier(Model):
    def __init__(self):
        super(ChurnClassifier, self).__init__()
        self.layer1 = layers.Dense(20, input_dim = 20, activation = 'relu')
        self.layer2 = layers.Dense(41, activation = 'relu')
        self.layer3 = layers.Dense(83, activation = 'relu')
        self.layer4 = layers.Dense(2,  activation = 'sigmoid')
        
    def call(self, inputs):
        x = self.layer1(inputs)
        x = self.layer2(x)
        x = self.layer3(x)
        return self.layer4(x)
        
ChurnClassifier = ChurnClassifier()
ChurnClassifier.compile(optimizer = 'adam',
                        loss=tf.keras.losses.CategoricalCrossentropy(),
                       metrics = ['accuracy'])

Now I fitted the model:

history = ChurnClassifier.fit(X_train_nur, Y_train_nur, 
          epochs=20, 
          batch_size=512,
          validation_data=(X_val_nur, Y_val_nur),
          shuffle=True)

Now, I want to predict the class either 0 or 1 so I used the code – prediction = ChurnClassifier.predict(X_val_nur)

Now I want to see how many are 0 and 1 to calculate the TN, FN, TP, FP. so I crated a Dataframe for prediction. Code-

pred_y = pd.DataFrame(prediction , columns=['pred_y']) 

But I am getting the following DataFrame-

enter image description here

My Sample X_train:

array([[2.02124594e+08, 3.63743942e+04, 2.12000000e+02, ...,
        4.30000000e+01, 0.00000000e+00, 1.00000000e+00],
       [4.93794595e+08, 6.66593354e+02, 4.22000000e+02, ...,
        2.60000000e+01, 0.00000000e+00, 1.00000000e+00],
       [7.28506124e+08, 1.17953696e+04, 1.14000000e+03, ...,
        2.50000000e+01, 0.00000000e+00, 1.00000000e+00],
       ...,
       [4.63797916e+08, 1.19273275e+03, 4.10000000e+02, ...,
        9.00000000e+00, 0.00000000e+00, 1.00000000e+00],
       [4.04285400e+08, 1.87350825e+04, 3.01000000e+02, ...,
        1.60000000e+01, 0.00000000e+00, 1.00000000e+00],
       [5.08433538e+08, 3.19289528e+03, 4.18000000e+02, ...,
        9.00000000e+00, 0.00000000e+00, 1.00000000e+00]])

My Sample y_train- array([0, 0, 0, ..., 0, 0, 0], dtype=int64)

Y_train_nur only contain 0 and 1

What’s the issue??

Thanks in advance!

Solution

For Binary Classification, Last layer in model has to contain one neuron and model has to be compiled with

loss = tf.keras.losses.BinaryCrossentropy(from_logits=True),

Revised code would be

class ChurnClassifier(Model):
    def __init__(self):
        super(ChurnClassifier, self).__init__()
        self.layer1 = layers.Dense(20, input_dim = 20, activation = 'relu')
        self.layer2 = layers.Dense(41, activation = 'relu')
        self.layer3 = layers.Dense(83, activation = 'relu')
        self.layer4 = layers.Dense(1)
        
    def call(self, inputs):
        x = self.layer1(inputs)
        x = self.layer2(x)
        x = self.layer3(x)
        return self.layer4(x)
        
ChurnClassifier = ChurnClassifier()
ChurnClassifier.compile(optimizer = 'adam',
                        loss =tf.keras.losses.BinaryCrossentropy(from_logits=True),
                       metrics = ['accuracy'])

history = ChurnClassifier.fit(X_train_nur, Y_train_nur, 
          epochs=20, 
          batch_size=512,
          validation_data=(X_val_nur, Y_val_nur),
          shuffle=True)

Answered By – Tensorflow Support

This Answer collected from stackoverflow, is licensed under cc by-sa 2.5 , cc by-sa 3.0 and cc by-sa 4.0

Leave a Reply

(*) Required, Your email will not be published