import tensorflow as tf x = tf.random.normal([1, 3]) w = tf.ones([3, 1]) b = tf.ones([1]) y = tf.constant([1]) with tf.GradientTape() as tape: tape.watch([w, b]) prob = tf.sigmoid(x @ w + b) loss = tf.reduce_mean(tf.losses.MSE(y, prob)) grads = tape.gradient(loss, [w, b])
grads[0]
grads[1]