Bidirectional RNNs in PyTorch - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to create a bidirectional RNN layer in PyTorch.

PyTorch

import torch.nn as nn

rnn = nn.RNN(input_size=10, hidden_size=20, num_layers=1, bidirectional=[1])

Drag options to blanks, or click blank then click option'

ANone

BTrue

CFalse

D0

Attempts:

3 left

💡 Hint

Common Mistakes

Setting bidirectional to False or 0 will create a unidirectional RNN.

Using None will cause an error because bidirectional expects a boolean.

Setting bidirectional=True creates a bidirectional RNN layer that processes input sequences in both forward and backward directions.

2fill in blank

medium

Complete the code to get the output shape of a bidirectional RNN layer.

PyTorch

output, hidden = rnn(input_seq)
output_shape = output.shape  # (seq_len, batch, [1] * hidden_size)

Drag options to blanks, or click blank then click option'

A1

Bnum_layers

C2

Dinput_size

Attempts:

3 left

💡 Hint

Common Mistakes

Using 1 instead of 2 will give incorrect output shape.

Confusing num_layers with bidirectionality factor.

For a bidirectional RNN, the output's last dimension is 2 * hidden_size because it concatenates forward and backward outputs.

3fill in blank

hard

Fix the error in the code to correctly initialize a bidirectional LSTM.

PyTorch

lstm = nn.LSTM(input_size=15, hidden_size=30, num_layers=2, bidirectional=[1])

Drag options to blanks, or click blank then click option'

ATrue

BFalse

CNone

D2

Attempts:

3 left

💡 Hint

Common Mistakes

Using integer 2 instead of boolean True causes an error.

Setting bidirectional to False disables bidirectionality.

The bidirectional parameter must be a boolean. Setting it to True enables bidirectional LSTM.

4fill in blank

hard

Complete the code to slice the forward and backward outputs from a bidirectional RNN.

PyTorch

forward_out = output[:, :, :[1]]
backward_out = output[:, :, [1]:[2]]

Drag options to blanks, or click blank then click option'

Ahidden_size

B2 * hidden_size

Cnum_layers

Dinput_size

Attempts:

3 left

💡 Hint

Common Mistakes

Using 2 * hidden_size for the forward slice.

Forgetting the end slice for backward output.

Confusing output slicing with hidden state slicing.

Forward output is the first hidden_size features; backward is from hidden_size to 2 * hidden_size.

5fill in blank

hard

Fill all three blanks to extract forward and backward hidden states using slicing.

PyTorch

forward_hn = hn[[1]::[2]]
backward_hn = hn[[3]::[2]]

Drag options to blanks, or click blank then click option'

A0

B1

C2

Dnum_layers

Attempts:

3 left

💡 Hint

Common Mistakes

Swapping start indices (using 1::2 for forward).

Using step=1 or other values instead of 2.

Using contiguous view instead of simple slicing.

forward_hn = hn[0::2] and backward_hn = hn[1::2] separate forward (even indices) and backward (odd indices) across all layers.