Consider the following code that we used to classify the income of adults in the US$.$

inputs $=$ tf$.$keras.layers.Dense$($units$=32,$ activation$=$'relu', input$\_$shape$=[$len$($features$.$columns$)])$

hidden $=$ tf$.$keras.layers.Dense$($units$=32,$ activation$=$'relu'$)$

outputs $=$ tf$.$keras.layers.Dense$($units$=2,$ activation$=$TO BE FILLED$)$

model $=$ tf$.$keras.Sequential$([$inputs$,$ hidden, outputs$])$

loss $=$ 'sparse$\_$categorical$\_$crossentropy'

optimizer $=$ tf$.$keras.optimizers.RMSprop$(0\mathrm{.}001)$

model.compile$($loss$=$loss, optimizer$=$optimizer, metrics$=[\text{'}$accuracy$\text{'}])$

What was the activation function used on the output layer in this deep neural network?

Group of answer choices

softmax

relu

sigmoid

softplus

tanh