C5W1A1-rnn_cell_backward (optional) almost correct results

Course Q&A Deep Learning Specialization Sequence Models
1

Hi everyone,

My implementation is nearly correct except for the shape of dba.

gradients[“dxt”][1][2] = -1.3872130506020925
gradients[“dxt”].shape = (3, 10)
gradients[“da_prev”][2][3] = -0.15239949377395495
gradients[“da_prev”].shape = (5, 10)
gradients[“dWax”][3][1] = 0.4107728249354584
gradients[“dWax”].shape = (5, 3)
gradients[“dWaa”][1][2] = 1.1503450668497135
gradients[“dWaa”].shape = (5, 5)
gradients[“dba”][4] = 0.2002349138798539
gradients[“dba”].shape = (5,)
Which should be (5,1)

My computation is dba = dtanh.sum(axis=1)

What should I be doing to have the correct dimensions?

2

Very close. As you already pointed out, the clue is the shape of dba. It says that one dimension is reduced… So, “keepdims” should work.

3

Ah…I misread that in the Numpy documentation. Thanks for pointing out that it doesn’t default to True.