tensorflow output layer configuration for classification task

I am trying to teach my first AI with google colab and tensorflow.

.predict method work is not clear to me.

I have following dataset example:

Expected classes are 1 or 4. Dataset contains 50% of each.

My code is:

1. Slice Data:

  features = df.iloc[1:, 1:]
  labels =  df.iloc[1:, 0]

2. Build Model:

model = tf.keras.Sequential([
    tf.keras.layers.Dense(256, activation='tanh', input_shape=(82,)),
    tf.keras.layers.Dense(2, input_shape=(256,), activation='tanh'),
    tf.keras.layers.Dense(1, activation='softmax')
])

model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

3. Train Model:

model.fit(x=features, y=labels, shuffle=True, epochs=1)

Results

result of .predict method is always [[1.]].

But I think that is should be smth like:
[1: 0.4][2: 0.88]

where: 1 and 4 are classifications and 0.4 and 0.88 are probability

>Solution :

You have two issues in the configuration:

1. tf.keras.layers.Dense(1, activation='softmax'), change this to tf.keras.layers.Dense(2, activation='softmax')
2. Since your labels are in integer format, you need to change your loss function to model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

Also, the way your dataset is formatted and by means of softmax, you will never have an output like you described :

But I think that is should be smth like: [1: 0.4][2: 0.88]

The sum of probabilities will always sum up to 1 in your context.