Questions

1. Prove the linear relation of logistic regression by assuming that our Gaussian distributions share the same covariance matrix.
2. Change the learning rate of the optimizer and see how the loss value is increased/decreased in iterations.
3. Try to find the mapping function so that you can convert our non-linearly separable samples into linearly separable data points.
4. What will happen if the bias vector is not added to the input data?
5. What will happen if the loss function is changed? Change it to each of the following:
   • Mean squared error (tf.losses.meanSquaredError)
   • Absolute error (tf.losses.absoluteDifference)
   • Weighted loss (tf.losses.computeWeightedLoss)
6. Let's try to implement multiclass logistic regression that supports three-class predictions.
   • Hint: Combine two logistic regression models to do binary classification twice.
7. Save and load the logistic regression...