10.3. Deep Recurrent Neural Networks¶ Open the notebook in SageMaker Studio Lab
Up until now, we have focused on defining networks consisting of a sequence input, a single hidden RNN layer, and an output layer. Despite having just one hidden layer between the input at any time step and the corresponding output, there is a sense in which these networks are deep. Inputs from the first time step can influence the outputs at the final time step \(T\) (often 100s or 1000s of steps later). These inputs pass through \(T\) applications of the recurrent layer before reaching the final output. However, we often also wish to retain the ability to express complex relationships between the inputs at a given time step and the outputs at that same time step. Thus we often construct RNNs that are deep not only in the time direction but also in the input-to-output direction. This is precisely the notion of depth that we have already encountered in our development of MLPs and deep CNNs.
The standard method for building this sort of deep RNN is strikingly simple: we stack the RNNs on top of each other. Given a sequence of length \(T\), the first RNN produces a sequence of outputs, also of length \(T\). These, in turn, constitute the inputs to the next RNN layer. In this short section, we illustrate this design pattern and present a simple example for how to code up such stacked RNNs. Below, in Fig. 10.3.1, we illustrate a deep RNN with \(L\) hidden layers. Each hidden state operates on a sequential input and produces a sequential output. Moreover, any RNN cell (white box in Fig. 10.3.1) at each time step depends on both the same layer’s value at the previous time step and the previous layer’s value at the same time step.
Fig. 10.3.1 Architecture of a deep RNN.¶
Formally, suppose that we have a minibatch input \(\mathbf{X}_t \in \mathbb{R}^{n \times d}\) (number of examples \(=n\); number of inputs in each example \(=d\)) at time step \(t\). At the same time step, let the hidden state of the \(l^\textrm{th}\) hidden layer (\(l=1,\ldots,L\)) be \(\mathbf{H}_t^{(l)} \in \mathbb{R}^{n \times h}\) (number of hidden units \(=h\)) and the output layer variable be \(\mathbf{O}_t \in \mathbb{R}^{n \times q}\) (number of outputs: \(q\)). Setting \(\mathbf{H}_t^{(0)} = \mathbf{X}_t\), the hidden state of the \(l^\textrm{th}\) hidden layer that uses the activation function \(\phi_l\) is calculated as follows:
where the weights \(\mathbf{W}_{\textrm{xh}}^{(l)} \in \mathbb{R}^{h \times h}\) and \(\mathbf{W}_{\textrm{hh}}^{(l)} \in \mathbb{R}^{h \times h}\), together with the bias \(\mathbf{b}_\textrm{h}^{(l)} \in \mathbb{R}^{1 \times h}\), are the model parameters of the \(l^\textrm{th}\) hidden layer.
At the end, the calculation of the output layer is only based on the hidden state of the final \(L^\textrm{th}\) hidden layer:
where the weight \(\mathbf{W}_{\textrm{hq}} \in \mathbb{R}^{h \times q}\) and the bias \(\mathbf{b}_\textrm{q} \in \mathbb{R}^{1 \times q}\) are the model parameters of the output layer.
Just as with MLPs, the number of hidden layers \(L\) and the number of hidden units \(h\) are hyperparameters that we can tune. Common RNN layer widths (\(h\)) are in the range \((64, 2056)\), and common depths (\(L\)) are in the range \((1, 8)\). In addition, we can easily get a deep-gated RNN by replacing the hidden state computation in (10.3.1) with that from an LSTM or a GRU.
import torch
from torch import nn
from d2l import torch as d2l
from mxnet import np, npx
from mxnet.gluon import rnn
from d2l import mxnet as d2l
npx.set_np()
import jax
from flax import linen as nn
from jax import numpy as jnp
from d2l import jax as d2l
import tensorflow as tf
from d2l import tensorflow as d2l
10.3.1. Implementation from Scratch¶
To implement a multilayer RNN from scratch, we can treat each layer as
an RNNScratch instance with its own learnable parameters.
class StackedRNNScratch(d2l.Module):
def __init__(self, num_inputs, num_hiddens, num_layers, sigma=0.01):
super().__init__()
self.save_hyperparameters()
self.rnns = nn.Sequential(*[d2l.RNNScratch(
num_inputs if i==0 else num_hiddens, num_hiddens, sigma)
for i in range(num_layers)])
class StackedRNNScratch(d2l.Module):
def __init__(self, num_inputs, num_hiddens, num_layers, sigma=0.01):
super().__init__()
self.save_hyperparameters()
self.rnns = [d2l.RNNScratch(num_inputs if i==0 else num_hiddens,
num_hiddens, sigma)
for i in range(num_layers)]
class StackedRNNScratch(d2l.Module):
num_inputs: int
num_hiddens: int
num_layers: int
sigma: float = 0.01
def setup(self):
self.rnns = [d2l.RNNScratch(self.num_inputs if i==0 else self.num_hiddens,
self.num_hiddens, self.sigma)
for i in range(self.num_layers)]
class StackedRNNScratch(d2l.Module):
def __init__(self, num_inputs, num_hiddens, num_layers, sigma=0.01):
super().__init__()
self.save_hyperparameters()
self.rnns = [d2l.RNNScratch(num_inputs if i==0 else num_hiddens,
num_hiddens, sigma)
for i in range(num_layers)]
The multilayer forward computation simply performs forward computation layer by layer.
@d2l.add_to_class(StackedRNNScratch)
def forward(self, inputs, Hs=None):
outputs = inputs
if Hs is None: Hs = [None] * self.num_layers
for i in range(self.num_layers):
outputs, Hs[i] = self.rnns[i](outputs, Hs[i])
outputs = torch.stack(outputs, 0)
return outputs, Hs
@d2l.add_to_class(StackedRNNScratch)
def forward(self, inputs, Hs=None):
outputs = inputs
if Hs is None: Hs = [None] * self.num_layers
for i in range(self.num_layers):
outputs, Hs[i] = self.rnns[i](outputs, Hs[i])
outputs = np.stack(outputs, 0)
return outputs, Hs
@d2l.add_to_class(StackedRNNScratch)
def forward(self, inputs, Hs=None):
outputs = inputs
if Hs is None: Hs = [None] * self.num_layers
for i in range(self.num_layers):
outputs, Hs[i] = self.rnns[i](outputs, Hs[i])
outputs = jnp.stack(outputs, 0)
return outputs, Hs
@d2l.add_to_class(StackedRNNScratch)
def forward(self, inputs, Hs=None):
outputs = inputs
if Hs is None: Hs = [None] * self.num_layers
for i in range(self.num_layers):
outputs, Hs[i] = self.rnns[i](outputs, Hs[i])
outputs = tf.stack(outputs, 0)
return outputs, Hs
As an example, we train a deep GRU model on The Time Machine dataset (same as in Section 9.5). To keep things simple we set the number of layers to 2.
data = d2l.TimeMachine(batch_size=1024, num_steps=32)
rnn_block = StackedRNNScratch(num_inputs=len(data.vocab),
num_hiddens=32, num_layers=2)
model = d2l.RNNLMScratch(rnn_block, vocab_size=len(data.vocab), lr=2)
trainer = d2l.Trainer(max_epochs=100, gradient_clip_val=1, num_gpus=1)
trainer.fit(model, data)
data = d2l.TimeMachine(batch_size=1024, num_steps=32)
rnn_block = StackedRNNScratch(num_inputs=len(data.vocab),
num_hiddens=32, num_layers=2)
model = d2l.RNNLMScratch(rnn_block, vocab_size=len(data.vocab), lr=2)
trainer = d2l.Trainer(max_epochs=100, gradient_clip_val=1, num_gpus=1)
trainer.fit(model, data)
data = d2l.TimeMachine(batch_size=1024, num_steps=32)
rnn_block = StackedRNNScratch(num_inputs=len(data.vocab),
num_hiddens=32, num_layers=2)
model = d2l.RNNLMScratch(rnn_block, vocab_size=len(data.vocab), lr=2)
trainer = d2l.Trainer(max_epochs=100, gradient_clip_val=1, num_gpus=1)
trainer.fit(model, data)
data = d2l.TimeMachine(batch_size=1024, num_steps=32)
with d2l.try_gpu():
rnn_block = StackedRNNScratch(num_inputs=len(data.vocab),
num_hiddens=32, num_layers=2)
model = d2l.RNNLMScratch(rnn_block, vocab_size=len(data.vocab), lr=2)
trainer = d2l.Trainer(max_epochs=100, gradient_clip_val=1)
trainer.fit(model, data)
10.3.2. Concise Implementation¶
Fortunately many of the logistical details required to implement multiple layers of an RNN are readily available in high-level APIs. Our concise implementation will use such built-in functionalities. The code generalizes the one we used previously in Section 10.2, letting us specify the number of layers explicitly rather than picking the default of only one layer.
class GRU(d2l.RNN): #@save
"""The multilayer GRU model."""
def __init__(self, num_inputs, num_hiddens, num_layers, dropout=0):
d2l.Module.__init__(self)
self.save_hyperparameters()
self.rnn = nn.GRU(num_inputs, num_hiddens, num_layers,
dropout=dropout)
Fortunately many of the logistical details required to implement multiple layers of an RNN are readily available in high-level APIs. Our concise implementation will use such built-in functionalities. The code generalizes the one we used previously in Section 10.2, letting us specify the number of layers explicitly rather than picking the default of only one layer.
class GRU(d2l.RNN): #@save
"""The multilayer GRU model."""
def __init__(self, num_hiddens, num_layers, dropout=0):
d2l.Module.__init__(self)
self.save_hyperparameters()
self.rnn = rnn.GRU(num_hiddens, num_layers, dropout=dropout)
Flax takes a minimalistic approach while implementing RNNs. Defining the
number of layers in an RNN or combining it with dropout is not available
out of the box. Our concise implementation will use all built-in
functionalities and add num_layers and dropout features on top.
The code generalizes the one we used previously in Section 10.2,
allowing specification of the number of layers explicitly rather than
picking the default of a single layer.
class GRU(d2l.RNN): #@save
"""The multilayer GRU model."""
num_hiddens: int
num_layers: int
dropout: float = 0
@nn.compact
def __call__(self, X, state=None, training=False):
outputs = X
new_state = []
if state is None:
batch_size = X.shape[1]
state = [nn.GRUCell.initialize_carry(jax.random.PRNGKey(0),
(batch_size,), self.num_hiddens)] * self.num_layers
GRU = nn.scan(nn.GRUCell, variable_broadcast="params",
in_axes=0, out_axes=0, split_rngs={"params": False})
# Introduce a dropout layer after every GRU layer except last
for i in range(self.num_layers - 1):
layer_i_state, X = GRU()(state[i], outputs)
new_state.append(layer_i_state)
X = nn.Dropout(self.dropout, deterministic=not training)(X)
# Final GRU layer without dropout
out_state, X = GRU()(state[-1], X)
new_state.append(out_state)
return X, jnp.array(new_state)
Fortunately many of the logistical details required to implement multiple layers of an RNN are readily available in high-level APIs. Our concise implementation will use such built-in functionalities. The code generalizes the one we used previously in Section 10.2, letting us specify the number of layers explicitly rather than picking the default of only one layer.
class GRU(d2l.RNN): #@save
"""The multilayer GRU model."""
def __init__(self, num_hiddens, num_layers, dropout=0):
d2l.Module.__init__(self)
self.save_hyperparameters()
gru_cells = [tf.keras.layers.GRUCell(num_hiddens, dropout=dropout)
for _ in range(num_layers)]
self.rnn = tf.keras.layers.RNN(gru_cells, return_sequences=True,
return_state=True, time_major=True)
def forward(self, X, state=None):
outputs, *state = self.rnn(X, state)
return outputs, state
The architectural decisions such as choosing hyperparameters are very
similar to those of Section 10.2. We pick the same number of
inputs and outputs as we have distinct tokens, i.e., vocab_size. The
number of hidden units is still 32. The only difference is that we now
select a nontrivial number of hidden layers by specifying the value of
num_layers.
gru = GRU(num_inputs=len(data.vocab), num_hiddens=32, num_layers=2)
model = d2l.RNNLM(gru, vocab_size=len(data.vocab), lr=2)
trainer.fit(model, data)
model.predict('it has', 20, data.vocab, d2l.try_gpu())
'it has for and the time th'
gru = GRU(num_hiddens=32, num_layers=2)
model = d2l.RNNLM(gru, vocab_size=len(data.vocab), lr=2)
# Running takes > 1h (pending fix from MXNet)
# trainer.fit(model, data)
# model.predict('it has', 20, data.vocab, d2l.try_gpu())
gru = GRU(num_hiddens=32, num_layers=2)
model = d2l.RNNLM(gru, vocab_size=len(data.vocab), lr=2)
trainer.fit(model, data)
model.predict('it has', 20, data.vocab, trainer.state.params)
'it has is the prough said '
gru = GRU(num_hiddens=32, num_layers=2)
with d2l.try_gpu():
model = d2l.RNNLM(gru, vocab_size=len(data.vocab), lr=2)
trainer.fit(model, data)
model.predict('it has', 20, data.vocab)
'it has the time traveller '
10.3.3. Summary¶
In deep RNNs, the hidden state information is passed to the next time step of the current layer and the current time step of the next layer. There exist many different flavors of deep RNNs, such as LSTMs, GRUs, or vanilla RNNs. Conveniently, these models are all available as parts of the high-level APIs of deep learning frameworks. Initialization of models requires care. Overall, deep RNNs require considerable amount of work (such as learning rate and clipping) to ensure proper convergence.
10.3.4. Exercises¶
Replace the GRU by an LSTM and compare the accuracy and training speed.
Increase the training data to include multiple books. How low can you go on the perplexity scale?
Would you want to combine sources of different authors when modeling text? Why is this a good idea? What could go wrong?