Skip to content

Recurrent Neural Networks (RNN)

View all lessons
Explore and motivate the need for representation via embeddings.
Goku Mohandas
Goku Mohandas
· · ·
Repository · Notebook

Subscribe to our newsletter

📬  Receive new lessons straight to your inbox (once a month) and join 40K+ developers in learning how to responsibly deliver value with ML.

Overview

So far we've processed inputs as whole (ex. applying filters across the entire input to extract features) but we can also process our inputs sequentially. For example we can think of each token in our text as an event in time (timestep). We can process each timestep, one at a time, and predict the class after the last timestep (token) has been processed. This is very powerful because the model now has a meaningful way to account for the sequential order of tokens in our sequence and predict accordingly.

vanilla RNN

$$ \text{RNN forward pass for a single time step } X_t $$:

\[ h_t = tanh(W_{hh}h_{t-1} + W_{xh}X_t+b_h) \]

Variable Description
\(N\) batch size
\(E\) embeddings dimension
\(H\) # of hidden units
\(W_{hh}\) RNN weights \(\in \mathbb{R}^{HXH}\)
\(h_{t-1}\) previous timestep's hidden state \(\in in \mathbb{R}^{NXH}\)
\(W_{xh}\) input weights \(\in \mathbb{R}^{EXH}\)
\(X_t\) input at time step \(t \in \mathbb{R}^{NXE}\)
\(b_h\) hidden units bias \(\in \mathbb{R}^{HX1}\)
\(h_t\) output from RNN for timestep \(t\)

  • Objective:
    • Process sequential data by accounting for the current input and also what has been learned from previous inputs.
  • Advantages:
    • Account for order and previous inputs in a meaningful way.
    • Conditioned generation for generating sequences.
  • Disadvantages:
    • Each time step's prediction depends on the previous prediction so it's difficult to parallelize RNN operations.
    • Processing long sequences can yield memory and computation issues.
    • Interpretability is difficult but there are few techniques that use the activations from RNNs to see what parts of the inputs are processed.
  • Miscellaneous:
    • Architectural tweaks to make RNNs faster and interpretable is an ongoing area of research.

Set up

Let's set our seed and device for our main task. 

1
2
3
4
5
import numpy as np
import pandas as pd
import random
import torch
import torch.nn as nn

1
SEED = 1234

1
2
3
4
5
6
7
def set_seeds(seed=1234):
    """Set seeds for reproducibility."""
    np.random.seed(seed)
    random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)
    torch.cuda.manual_seed_all(seed) # multi-GPU

1
2
# Set seeds for reproducibility
set_seeds(seed=SEED)

1
2
3
4
5
6
7
8
# Set device
cuda = True
device = torch.device("cuda" if (
    torch.cuda.is_available() and cuda) else "cpu")
torch.set_default_tensor_type("torch.FloatTensor")
if device.type == "cuda":
    torch.set_default_tensor_type("torch.cuda.FloatTensor")
print (device)


cuda

Load data

We will download the AG News dataset, which consists of 120K text samples from 4 unique classes (Business, Sci/Tech, Sports, World) 

1
2
3
4
5
# Load data
url = "https://raw.githubusercontent.com/GokuMohandas/Made-With-ML/main/datasets/news.csv"
df = pd.read_csv(url, header=0) # load
df = df.sample(frac=1).reset_index(drop=True) # shuffle
df.head()

title category
0 Sharon Accepts Plan to Reduce Gaza Army Operation... World
1 Internet Key Battleground in Wildlife Crime Fight Sci/Tech
2 July Durable Good Orders Rise 1.7 Percent Business
3 Growing Signs of a Slowing on Wall Street Business
4 The New Faces of Reality TV World

Preprocessing

We're going to clean up our input data first by doing operations such as lower text, removing stop (filler) words, filters using regular expressions, etc. 

1
2
3
4
import nltk
from nltk.corpus import stopwords
from nltk.stem import PorterStemmer
import re

1
2
3
4
nltk.download("stopwords")
STOPWORDS = stopwords.words("english")
print (STOPWORDS[:5])
porter = PorterStemmer()


[nltk_data] Downloading package stopwords to /root/nltk_data...
[nltk_data]   Package stopwords is already up-to-date!
['i', 'me', 'my', 'myself', 'we']


 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
def preprocess(text, stopwords=STOPWORDS):
    """Conditional preprocessing on our text unique to our task."""
    # Lower
    text = text.lower()

    # Remove stopwords
    pattern = re.compile(r"\b(" + r"|".join(stopwords) + r")\b\s*")
    text = pattern.sub("", text)

    # Remove words in parenthesis
    text = re.sub(r"\([^)]*\)", "", text)

    # Spacing and filters
    text = re.sub(r"([-;;.,!?<=>])", r" \1 ", text)
    text = re.sub("[^A-Za-z0-9]+", " ", text) # remove non alphanumeric chars
    text = re.sub(" +", " ", text)  # remove multiple spaces
    text = text.strip()

    return text

1
2
3
# Sample
text = "Great week for the NYSE!"
preprocess(text=text)


great week nyse

1
2
3
4
# Apply to dataframe
preprocessed_df = df.copy()
preprocessed_df.title = preprocessed_df.title.apply(preprocess)
print (f"{df.title.values[0]}\n\n{preprocessed_df.title.values[0]}")

Sharon Accepts Plan to Reduce Gaza Army Operation, Haaretz Says

sharon accepts plan reduce gaza army operation haaretz says

Warning

If you have preprocessing steps like standardization, etc. that are calculated, you need to separate the training and test set first before applying those operations. This is because we cannot apply any knowledge gained from the test set accidentally (data leak) during preprocessing/training. However for global preprocessing steps like the function above where we aren't learning anything from the data itself, we can perform before splitting the data.

Split data

1
2
import collections
from sklearn.model_selection import train_test_split

1
2
3
TRAIN_SIZE = 0.7
VAL_SIZE = 0.15
TEST_SIZE = 0.15

1
2
3
4
5
def train_val_test_split(X, y, train_size):
    """Split dataset into data splits."""
    X_train, X_, y_train, y_ = train_test_split(X, y, train_size=TRAIN_SIZE, stratify=y)
    X_val, X_test, y_val, y_test = train_test_split(X_, y_, train_size=0.5, stratify=y_)
    return X_train, X_val, X_test, y_train, y_val, y_test

1
2
3
# Data
X = preprocessed_df["title"].values
y = preprocessed_df["category"].values

1
2
3
4
5
6
7
# Create data splits
X_train, X_val, X_test, y_train, y_val, y_test = train_val_test_split(
    X=X, y=y, train_size=TRAIN_SIZE)
print (f"X_train: {X_train.shape}, y_train: {y_train.shape}")
print (f"X_val: {X_val.shape}, y_val: {y_val.shape}")
print (f"X_test: {X_test.shape}, y_test: {y_test.shape}")
print (f"Sample point: {X_train[0]} → {y_train[0]}")


X_train: (84000,), y_train: (84000,)
X_val: (18000,), y_val: (18000,)
X_test: (18000,), y_test: (18000,)
Sample point: china battles north korea nuclear talks → World

Label encoding

Next we'll define a LabelEncoder to encode our text labels into unique indices 

1
import itertools

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
class LabelEncoder(object):
    """Label encoder for tag labels."""
    def __init__(self, class_to_index={}):
        self.class_to_index = class_to_index or {}  # mutable defaults ;)
        self.index_to_class = {v: k for k, v in self.class_to_index.items()}
        self.classes = list(self.class_to_index.keys())

    def __len__(self):
        return len(self.class_to_index)

    def __str__(self):
        return f"<LabelEncoder(num_classes={len(self)})>"

    def fit(self, y):
        classes = np.unique(y)
        for i, class_ in enumerate(classes):
            self.class_to_index[class_] = i
        self.index_to_class = {v: k for k, v in self.class_to_index.items()}
        self.classes = list(self.class_to_index.keys())
        return self

    def encode(self, y):
        encoded = np.zeros((len(y)), dtype=int)
        for i, item in enumerate(y):
            encoded[i] = self.class_to_index[item]
        return encoded

    def decode(self, y):
        classes = []
        for i, item in enumerate(y):
            classes.append(self.index_to_class[item])
        return classes

    def save(self, fp):
        with open(fp, "w") as fp:
            contents = {'class_to_index': self.class_to_index}
            json.dump(contents, fp, indent=4, sort_keys=False)

    @classmethod
    def load(cls, fp):
        with open(fp, "r") as fp:
            kwargs = json.load(fp=fp)
        return cls(**kwargs)

1
2
3
4
5
# Encode
label_encoder = LabelEncoder()
label_encoder.fit(y_train)
NUM_CLASSES = len(label_encoder)
label_encoder.class_to_index


{'Business': 0, 'Sci/Tech': 1, 'Sports': 2, 'World': 3}

1
2
3
4
5
6
# Convert labels to tokens
print (f"y_train[0]: {y_train[0]}")
y_train = label_encoder.encode(y_train)
y_val = label_encoder.encode(y_val)
y_test = label_encoder.encode(y_test)
print (f"y_train[0]: {y_train[0]}")

y_train[0]: World
y_train[0]: 3

1
2
3
4
# Class weights
counts = np.bincount(y_train)
class_weights = {i: 1.0/count for i, count in enumerate(counts)}
print (f"counts: {counts}\nweights: {class_weights}")

counts: [21000 21000 21000 21000]
weights: {0: 4.761904761904762e-05, 1: 4.761904761904762e-05, 2: 4.761904761904762e-05, 3: 4.761904761904762e-05}

Tokenizer

We'll define a Tokenizer to convert our text input data into token indices.

1
2
3
import json
from collections import Counter
from more_itertools import take

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
class Tokenizer(object):
    def __init__(self, char_level, num_tokens=None,
                 pad_token="<PAD>", oov_token="<UNK>",
                 token_to_index=None):
        self.char_level = char_level
        self.separator = "" if self.char_level else " "
        if num_tokens: num_tokens -= 2 # pad + unk tokens
        self.num_tokens = num_tokens
        self.pad_token = pad_token
        self.oov_token = oov_token
        if not token_to_index:
            token_to_index = {pad_token: 0, oov_token: 1}
        self.token_to_index = token_to_index
        self.index_to_token = {v: k for k, v in self.token_to_index.items()}

    def __len__(self):
        return len(self.token_to_index)

    def __str__(self):
        return f"<Tokenizer(num_tokens={len(self)})>"

    def fit_on_texts(self, texts):
        if not self.char_level:
            texts = [text.split(" ") for text in texts]
        all_tokens = [token for text in texts for token in text]
        counts = Counter(all_tokens).most_common(self.num_tokens)
        self.min_token_freq = counts[-1][1]
        for token, count in counts:
            index = len(self)
            self.token_to_index[token] = index
            self.index_to_token[index] = token
        return self

    def texts_to_sequences(self, texts):
        sequences = []
        for text in texts:
            if not self.char_level:
                text = text.split(" ")
            sequence = []
            for token in text:
                sequence.append(self.token_to_index.get(
                    token, self.token_to_index[self.oov_token]))
            sequences.append(np.asarray(sequence))
        return sequences

    def sequences_to_texts(self, sequences):
        texts = []
        for sequence in sequences:
            text = []
            for index in sequence:
                text.append(self.index_to_token.get(index, self.oov_token))
            texts.append(self.separator.join([token for token in text]))
        return texts

    def save(self, fp):
        with open(fp, "w") as fp:
            contents = {
                "char_level": self.char_level,
                "oov_token": self.oov_token,
                "token_to_index": self.token_to_index
            }
            json.dump(contents, fp, indent=4, sort_keys=False)

    @classmethod
    def load(cls, fp):
        with open(fp, "r") as fp:
            kwargs = json.load(fp=fp)
        return cls(**kwargs)

Warning

It's important that we only fit using our train data split because during inference, our model will not always know every token so it's important to replicate that scenario with our validation and test splits as well.

1
2
3
4
5
# Tokenize
tokenizer = Tokenizer(char_level=False, num_tokens=5000)
tokenizer.fit_on_texts(texts=X_train)
VOCAB_SIZE = len(tokenizer)
print (tokenizer)

<Tokenizer(num_tokens=5000)>


1
2
3
# Sample of tokens
print (take(5, tokenizer.token_to_index.items()))
print (f"least freq token's freq: {tokenizer.min_token_freq}") # use this to adjust num_tokens

[('<PAD>', 0), ('<UNK>', 1), ('39', 2), ('b', 3), ('gt', 4)]
least freq token's freq: 14

1
2
3
4
5
6
7
8
# Convert texts to sequences of indices
X_train = tokenizer.texts_to_sequences(X_train)
X_val = tokenizer.texts_to_sequences(X_val)
X_test = tokenizer.texts_to_sequences(X_test)
preprocessed_text = tokenizer.sequences_to_texts([X_train[0]])[0]
print ("Text to indices:\n"
    f"  (preprocessed) → {preprocessed_text}\n"
    f"  (tokenized) → {X_train[0]}")

Text to indices:
  (preprocessed) → china battles north korea nuclear talks
  (tokenized) → [  16 1491  285  142  114   24]

Padding

We'll need to do 2D padding to our tokenized text. 

1
2
3
4
5
6
7
def pad_sequences(sequences, max_seq_len=0):
    """Pad sequences to max length in sequence."""
    max_seq_len = max(max_seq_len, max(len(sequence) for sequence in sequences))
    padded_sequences = np.zeros((len(sequences), max_seq_len))
    for i, sequence in enumerate(sequences):
        padded_sequences[i][:len(sequence)] = sequence
    return padded_sequences

1
2
3
4
# 2D sequences
padded = pad_sequences(X_train[0:3])
print (padded.shape)
print (padded)


(3, 6)
[[1.600e+01 1.491e+03 2.850e+02 1.420e+02 1.140e+02 2.400e+01]
 [1.445e+03 2.300e+01 6.560e+02 2.197e+03 1.000e+00 0.000e+00]
 [1.200e+02 1.400e+01 1.955e+03 1.005e+03 1.529e+03 4.014e+03]]

Datasets

We're going to create Datasets and DataLoaders to be able to efficiently create batches with our data splits.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
class Dataset(torch.utils.data.Dataset):
    def __init__(self, X, y):
        self.X = X
        self.y = y

    def __len__(self):
        return len(self.y)

    def __str__(self):
        return f"<Dataset(N={len(self)})>"

    def __getitem__(self, index):
        X = self.X[index]
        y = self.y[index]
        return [X, len(X), y]

    def collate_fn(self, batch):
        """Processing on a batch."""
        # Get inputs
        batch = np.array(batch)
        X = batch[:, 0]
        seq_lens = batch[:, 1]
        y = batch[:, 2]

        # Pad inputs
        X = pad_sequences(sequences=X)

        # Cast
        X = torch.LongTensor(X.astype(np.int32))
        seq_lens = torch.LongTensor(seq_lens.astype(np.int32))
        y = torch.LongTensor(y.astype(np.int32))

        return X, seq_lens, y

    def create_dataloader(self, batch_size, shuffle=False, drop_last=False):
        return torch.utils.data.DataLoader(
            dataset=self, batch_size=batch_size, collate_fn=self.collate_fn,
            shuffle=shuffle, drop_last=drop_last, pin_memory=True)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
# Create datasets
train_dataset = Dataset(X=X_train, y=y_train)
val_dataset = Dataset(X=X_val, y=y_val)
test_dataset = Dataset(X=X_test, y=y_test)
print ("Datasets:\n"
    f"  Train dataset:{train_dataset.__str__()}\n"
    f"  Val dataset: {val_dataset.__str__()}\n"
    f"  Test dataset: {test_dataset.__str__()}\n"
    "Sample point:\n"
    f"  X: {train_dataset[0][0]}\n"
    f"  seq_len: {train_dataset[0][1]}\n"
    f"  y: {train_dataset[0][2]}")


Datasets:
  Train dataset: <Dataset(N=84000)>
  Val dataset: <Dataset(N=18000)>
  Test dataset: <Dataset(N=18000)>
Sample point:
  X: [  16 1491  285  142  114   24]
  seq_len: 6
  y: 3

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
# Create dataloaders
batch_size = 64
train_dataloader = train_dataset.create_dataloader(
    batch_size=batch_size)
val_dataloader = val_dataset.create_dataloader(
    batch_size=batch_size)
test_dataloader = test_dataset.create_dataloader(
    batch_size=batch_size)
batch_X, batch_seq_lens, batch_y = next(iter(train_dataloader))
print ("Sample batch:\n"
    f"  X: {list(batch_X.size())}\n"
    f"  seq_lens: {list(batch_seq_lens.size())}\n"
    f"  y: {list(batch_y.size())}\n"
    "Sample point:\n"
    f"  X: {batch_X[0]}\n"
    f" seq_len: {batch_seq_lens[0]}\n"
    f"  y: {batch_y[0]}")

Sample batch:
  X: [64, 14]
  seq_lens: [64]
  y: [64]
Sample point:
  X: tensor([  16, 1491,  285,  142,  114,   24,    0,    0,    0,    0,    0,    0,
           0,    0])
 seq_len: 6
  y: 3

Trainer

Let's create the Trainer class that we'll use to facilitate training for our experiments.

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
class Trainer(object):
    def __init__(self, model, device, loss_fn=None, optimizer=None, scheduler=None):

        # Set params
        self.model = model
        self.device = device
        self.loss_fn = loss_fn
        self.optimizer = optimizer
        self.scheduler = scheduler

    def train_step(self, dataloader):
        """Train step."""
        # Set model to train mode
        self.model.train()
        loss = 0.0

        # Iterate over train batches
        for i, batch in enumerate(dataloader):

            # Step
            batch = [item.to(self.device) for item in batch]  # Set device
            inputs, targets = batch[:-1], batch[-1]
            self.optimizer.zero_grad()  # Reset gradients
            z = self.model(inputs)  # Forward pass
            J = self.loss_fn(z, targets)  # Define loss
            J.backward()  # Backward pass
            self.optimizer.step()  # Update weights

            # Cumulative Metrics
            loss += (J.detach().item() - loss) / (i + 1)

        return loss

    def eval_step(self, dataloader):
        """Validation or test step."""
        # Set model to eval mode
        self.model.eval()
        loss = 0.0
        y_trues, y_probs = [], []

        # Iterate over val batches
        with torch.inference_mode():
            for i, batch in enumerate(dataloader):

                # Step
                batch = [item.to(self.device) for item in batch]  # Set device
                inputs, y_true = batch[:-1], batch[-1]
                z = self.model(inputs)  # Forward pass
                J = self.loss_fn(z, y_true).item()

                # Cumulative Metrics
                loss += (J - loss) / (i + 1)

                # Store outputs
                y_prob = F.softmax(z).cpu().numpy()
                y_probs.extend(y_prob)
                y_trues.extend(y_true.cpu().numpy())

        return loss, np.vstack(y_trues), np.vstack(y_probs)

    def predict_step(self, dataloader):
        """Prediction step."""
        # Set model to eval mode
        self.model.eval()
        y_probs = []

        # Iterate over val batches
        with torch.inference_mode():
            for i, batch in enumerate(dataloader):

                # Forward pass w/ inputs
                inputs, targets = batch[:-1], batch[-1]
                z = self.model(inputs)

                # Store outputs
                y_prob = F.softmax(z).cpu().numpy()
                y_probs.extend(y_prob)

        return np.vstack(y_probs)

    def train(self, num_epochs, patience, train_dataloader, val_dataloader):
        best_val_loss = np.inf
        for epoch in range(num_epochs):
            # Steps
            train_loss = self.train_step(dataloader=train_dataloader)
            val_loss, _, _ = self.eval_step(dataloader=val_dataloader)
            self.scheduler.step(val_loss)

            # Early stopping
            if val_loss < best_val_loss:
                best_val_loss = val_loss
                best_model = self.model
                _patience = patience  # reset _patience
            else:
                _patience -= 1
            if not _patience:  # 0
                print("Stopping early!")
                break

            # Logging
            print(
                f"Epoch: {epoch+1} | "
                f"train_loss: {train_loss:.5f}, "
                f"val_loss: {val_loss:.5f}, "
                f"lr: {self.optimizer.param_groups[0]['lr']:.2E}, "
                f"_patience: {_patience}"
            )
        return best_model

Vanilla RNN

RNN

Inputs to RNNs are sequential like text or time-series.

1
2
BATCH_SIZE = 64
EMBEDDING_DIM = 100

1
2
3
4
5
6
# Input
sequence_size = 8 # words per input
x = torch.rand((BATCH_SIZE, sequence_size, EMBEDDING_DIM))
seq_lens = torch.randint(high=sequence_size, size=(BATCH_SIZE, ))
print (x.shape)
print (seq_lens.shape)


torch.Size([64, 8, 100])
torch.Size([1, 64])
vanilla RNN

$$ \text{RNN forward pass for a single time step } X_t $$:

\[ h_t = tanh(W_{hh}h_{t-1} + W_{xh}X_t+b_h) \]

Variable Description
\(N\) batch size
\(E\) embeddings dimension
\(H\) # of hidden units
\(W_{hh}\) RNN weights \(\in \mathbb{R}^{HXH}\)
\(h_{t-1}\) previous timestep's hidden state \(\in in \mathbb{R}^{NXH}\)
\(W_{xh}\) input weights \(\in \mathbb{R}^{EXH}\)
\(X_t\) input at time step \(t \in \mathbb{R}^{NXE}\)
\(b_h\) hidden units bias \(\in \mathbb{R}^{HX1}\)
\(h_t\) output from RNN for timestep \(t\)

At the first time step, the previous hidden state \(h_{t-1}\) can either be a zero vector (unconditioned) or initialized (conditioned). If we are conditioning the RNN, the first hidden state \(h_0\) can belong to a specific condition or we can concat the specific condition to the randomly initialized hidden vectors at each time step. More on this in the subsequent notebooks on RNNs.

1
2
RNN_HIDDEN_DIM = 128
DROPOUT_P = 0.1

1
2
3
# Initialize hidden state
hidden_t = torch.zeros((BATCH_SIZE, RNN_HIDDEN_DIM))
print (hidden_t.size())


torch.Size([64, 128])

We'll show how to create an RNN cell using PyTorch's RNNCell and the more abstracted RNN.

1
2
3
# Initialize RNN cell
rnn_cell = nn.RNNCell(EMBEDDING_DIM, RNN_HIDDEN_DIM)
print (rnn_cell)

RNNCell(100, 128)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
# Forward pass through RNN
x = x.permute(1, 0, 2) # RNN needs batch_size to be at dim 1

# Loop through the inputs time steps
hiddens = []
for t in range(sequence_size):
    hidden_t = rnn_cell(x[t], hidden_t)
    hiddens.append(hidden_t)
hiddens = torch.stack(hiddens)
hiddens = hiddens.permute(1, 0, 2) # bring batch_size back to dim 0
print (hiddens.size())

torch.Size([64, 8, 128])

1
2
3
4
5
6
# We also could've used a more abstracted layer
x = torch.rand((BATCH_SIZE, sequence_size, EMBEDDING_DIM))
rnn = nn.RNN(EMBEDDING_DIM, RNN_HIDDEN_DIM, batch_first=True)
out, h_n = rnn(x) # h_n is the last hidden state
print ("out: ", out.shape)
print ("h_n: ", h_n.shape)

out:  torch.Size([64, 8, 128])
h_n:  torch.Size([1, 64, 128])

1
2
3
# The same tensors
print (out[:,-1,:])
print (h_n.squeeze(0))

tensor([[-0.0359, -0.3819,  0.2162,  ..., -0.3397,  0.0468,  0.1937],
        [-0.4914, -0.3056, -0.0837,  ..., -0.3507, -0.4320,  0.3593],
        [-0.0989, -0.2852,  0.1170,  ..., -0.0805, -0.0786,  0.3922],
        ...,
        [-0.3115, -0.4169,  0.2611,  ..., -0.3214,  0.0620,  0.0338],
        [-0.2455, -0.3380,  0.2048,  ..., -0.4198, -0.0075,  0.0372],
        [-0.2092, -0.4594,  0.1654,  ..., -0.5397, -0.1709,  0.0023]],
       grad_fn=<SliceBackward>)
tensor([[-0.0359, -0.3819,  0.2162,  ..., -0.3397,  0.0468,  0.1937],
        [-0.4914, -0.3056, -0.0837,  ..., -0.3507, -0.4320,  0.3593],
        [-0.0989, -0.2852,  0.1170,  ..., -0.0805, -0.0786,  0.3922],
        ...,
        [-0.3115, -0.4169,  0.2611,  ..., -0.3214,  0.0620,  0.0338],
        [-0.2455, -0.3380,  0.2048,  ..., -0.4198, -0.0075,  0.0372],
        [-0.2092, -0.4594,  0.1654,  ..., -0.5397, -0.1709,  0.0023]],
       grad_fn=<SqueezeBackward1>)

In our model, we want to use the RNN's output after the last relevant token in the sentence is processed. The last relevant token doesn't refer the <PAD> tokens but to the last actual word in the sentence and its index is different for each input in the batch. This is why we included a seq_lens tensor in our batches.

1
2
3
4
5
6
7
8
def gather_last_relevant_hidden(hiddens, seq_lens):
    """Extract and collect the last relevant
    hidden state based on the sequence length."""
    seq_lens = seq_lens.long().detach().cpu().numpy() - 1
    out = []
    for batch_index, column_index in enumerate(seq_lens):
        out.append(hiddens[batch_index, column_index])
    return torch.stack(out)

1
2
# Get the last relevant hidden state
gather_last_relevant_hidden(hiddens=out, seq_lens=seq_lens).squeeze(0).shape


torch.Size([64, 128])

There are many different ways to use RNNs. So far we've processed our inputs one timestep at a time and we could either use the RNN's output at each time step or just use the final input timestep's RNN output. Let's look at a few other possibilities.

RNN architecture

Model

1
import torch.nn.functional as F

1
HIDDEN_DIM = 100

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
class RNN(nn.Module):
    def __init__(self, embedding_dim, vocab_size, rnn_hidden_dim,
                 hidden_dim, dropout_p, num_classes, padding_idx=0):
        super(RNN, self).__init__()

        # Initialize embeddings
        self.embeddings = nn.Embedding(
            embedding_dim=embedding_dim, num_embeddings=vocab_size,
            padding_idx=padding_idx)

        # RNN
        self.rnn = nn.RNN(embedding_dim, rnn_hidden_dim, batch_first=True)

        # FC weights
        self.dropout = nn.Dropout(dropout_p)
        self.fc1 = nn.Linear(rnn_hidden_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, num_classes)

    def forward(self, inputs):
        # Embed
        x_in, seq_lens = inputs
        x_in = self.embeddings(x_in)

        # Rnn outputs
        out, h_n = self.rnn(x_in)
        z = gather_last_relevant_hidden(hiddens=out, seq_lens=seq_lens)

        # FC layers
        z = self.fc1(z)
        z = self.dropout(z)
        z = self.fc2(z)
        return z

1
2
3
4
5
6
7
# Simple RNN cell
model = RNN(
    embedding_dim=EMBEDDING_DIM, vocab_size=VOCAB_SIZE,
    rnn_hidden_dim=RNN_HIDDEN_DIM, hidden_dim=HIDDEN_DIM,
    dropout_p=DROPOUT_P, num_classes=NUM_CLASSES)
model = model.to(device) # set device
print (model.named_parameters)


<bound method Module.named_parameters of RNN(
  (embeddings): Embedding(5000, 100, padding_idx=0)
  (rnn): RNN(100, 128, batch_first=True)
  (dropout): Dropout(p=0.1, inplace=False)
  (fc1): Linear(in_features=128, out_features=100, bias=True)
  (fc2): Linear(in_features=100, out_features=4, bias=True)
)>

Training

1
from torch.optim import Adam

1
2
3
4
NUM_LAYERS = 1
LEARNING_RATE = 1e-4
PATIENCE = 10
NUM_EPOCHS = 50

1
2
3
# Define Loss
class_weights_tensor = torch.Tensor(list(class_weights.values())).to(device)
loss_fn = nn.CrossEntropyLoss(weight=class_weights_tensor)

1
2
3
4
# Define optimizer & scheduler
optimizer = Adam(model.parameters(), lr=LEARNING_RATE)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
    optimizer, mode="min", factor=0.1, patience=3)

1
2
3
4
# Trainer module
trainer = Trainer(
    model=model, device=device, loss_fn=loss_fn,
    optimizer=optimizer, scheduler=scheduler)

1
2
3
# Train
best_model = trainer.train(
    NUM_EPOCHS, PATIENCE, train_dataloader, val_dataloader)


Epoch: 1 | train_loss: 1.25605, val_loss: 1.10880, lr: 1.00E-04, _patience: 10
Epoch: 2 | train_loss: 1.03074, val_loss: 0.96749, lr: 1.00E-04, _patience: 10
Epoch: 3 | train_loss: 0.90110, val_loss: 0.86424, lr: 1.00E-04, _patience: 10
...
Epoch: 31 | train_loss: 0.32206, val_loss: 0.53581, lr: 1.00E-06, _patience: 3
Epoch: 32 | train_loss: 0.32233, val_loss: 0.53587, lr: 1.00E-07, _patience: 2
Epoch: 33 | train_loss: 0.32215, val_loss: 0.53572, lr: 1.00E-07, _patience: 1
Stopping early!

Evaluation

1
2
import json
from sklearn.metrics import precision_recall_fscore_support

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
def get_metrics(y_true, y_pred, classes):
    """Per-class performance metrics."""
    # Performance
    performance = {"overall": {}, "class": {}}

    # Overall performance
    metrics = precision_recall_fscore_support(y_true, y_pred, average="weighted")
    performance["overall"]["precision"] = metrics[0]
    performance["overall"]["recall"] = metrics[1]
    performance["overall"]["f1"] = metrics[2]
    performance["overall"]["num_samples"] = np.float64(len(y_true))

    # Per-class performance
    metrics = precision_recall_fscore_support(y_true, y_pred, average=None)
    for i in range(len(classes)):
        performance["class"][classes[i]] = {
            "precision": metrics[0][i],
            "recall": metrics[1][i],
            "f1": metrics[2][i],
            "num_samples": np.float64(metrics[3][i]),
        }

    return performance

1
2
3
# Get predictions
test_loss, y_true, y_prob = trainer.eval_step(dataloader=test_dataloader)
y_pred = np.argmax(y_prob, axis=1)

1
2
3
4
# Determine performance
performance = get_metrics(
    y_true=y_test, y_pred=y_pred, classes=label_encoder.classes)
print (json.dumps(performance["overall"], indent=2))


{
  "precision": 0.8171357577653572,
  "recall": 0.8176111111111112,
  "f1": 0.8171696173843819,
  "num_samples": 18000.0
}

Gated RNN

While our simple RNNs so far are great for sequentially processing our inputs, they have quite a few disadvantages. They commonly suffer from exploding or vanishing gradients as a result using the same set of weights (\(W_{xh}\) and \(W_{hh}\)) with each timestep's input. During backpropagation, this can cause gradients to explode (>1) or vanish (<1). If you multiply any number greater than 1 with itself over and over, it moves towards infinity (exploding gradients) and similarly, If you multiply any number less than 1 with itself over and over, it moves towards zero (vanishing gradients). To mitigate this issue, gated RNNs were devised to selectively retain information. If you're interested in learning more of the specifics, this post is a must-read.

There are two popular types of gated RNNs: Long Short-term Memory (LSTMs) units and Gated Recurrent Units (GRUs).

When deciding between LSTMs and GRUs, empirical performance is the best factor but in general GRUs offer similar performance with less complexity (less weights).

gated RNN
Understanding LSTM Networks - Chris Olah 
1
2
3
4
# Input
sequence_size = 8 # words per input
x = torch.rand((BATCH_SIZE, sequence_size, EMBEDDING_DIM))
print (x.shape)

torch.Size([64, 8, 100])


1
2
# GRU
gru = nn.GRU(input_size=EMBEDDING_DIM, hidden_size=RNN_HIDDEN_DIM, batch_first=True)

1
2
3
4
# Forward pass
out, h_n = gru(x)
print (f"out: {out.shape}")
print (f"h_n: {h_n.shape}")


out: torch.Size([64, 8, 128])
h_n: torch.Size([1, 64, 128])

Bidirectional RNN

We can also have RNNs that process inputs from both directions (first token to last token and vice versa) and combine their outputs. This architecture is known as a bidirectional RNN. 

1
2
3
# GRU
gru = nn.GRU(input_size=EMBEDDING_DIM, hidden_size=RNN_HIDDEN_DIM,
             batch_first=True, bidirectional=True)

1
2
3
4
# Forward pass
out, h_n = gru(x)
print (f"out: {out.shape}")
print (f"h_n: {h_n.shape}")


out: torch.Size([64, 8, 256])
h_n: torch.Size([2, 64, 128])

Notice that the output for each sample at each timestamp has size 256 (double the RNN_HIDDEN_DIM). This is because this includes both the forward and backward directions from the BiRNN.

Model

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
class GRU(nn.Module):
    def __init__(self, embedding_dim, vocab_size, rnn_hidden_dim,
                 hidden_dim, dropout_p, num_classes, padding_idx=0):
        super(GRU, self).__init__()

        # Initialize embeddings
        self.embeddings = nn.Embedding(embedding_dim=embedding_dim,
                                       num_embeddings=vocab_size,
                                       padding_idx=padding_idx)

        # RNN
        self.rnn = nn.GRU(embedding_dim, rnn_hidden_dim,
                          batch_first=True, bidirectional=True)

        # FC weights
        self.dropout = nn.Dropout(dropout_p)
        self.fc1 = nn.Linear(rnn_hidden_dim*2, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, num_classes)

    def forward(self, inputs:
        # Embed
        x_in, seq_lens = inputs
        x_in = self.embeddings(x_in)

        # Rnn outputs
        out, h_n = self.rnn(x_in)
        z = gather_last_relevant_hidden(hiddens=out, seq_lens=seq_lens)

        # FC layers
        z = self.fc1(z)
        z = self.dropout(z)
        z = self.fc2(z)
        return z

1
2
3
4
5
6
7
# Simple RNN cell
model = GRU(
    embedding_dim=EMBEDDING_DIM, vocab_size=VOCAB_SIZE,
    rnn_hidden_dim=RNN_HIDDEN_DIM, hidden_dim=HIDDEN_DIM,
    dropout_p=DROPOUT_P, num_classes=NUM_CLASSES)
model = model.to(device) # set device
print (model.named_parameters)


<bound method Module.named_parameters of GRU(
  (embeddings): Embedding(5000, 100, padding_idx=0)
  (rnn): GRU(100, 128, batch_first=True, bidirectional=True)
  (dropout): Dropout(p=0.1, inplace=False)
  (fc1): Linear(in_features=256, out_features=100, bias=True)
  (fc2): Linear(in_features=100, out_features=4, bias=True)
)>

Training

1
2
3
# Define Loss
class_weights_tensor = torch.Tensor(list(class_weights.values())).to(device)
loss_fn = nn.CrossEntropyLoss(weight=class_weights_tensor)

1
2
3
4
# Define optimizer & scheduler
optimizer = Adam(model.parameters(), lr=LEARNING_RATE)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
    optimizer, mode="min", factor=0.1, patience=3)

1
2
3
4
# Trainer module
trainer = Trainer(
    model=model, device=device, loss_fn=loss_fn,
    optimizer=optimizer, scheduler=scheduler)

1
2
3
# Train
best_model = trainer.train(
    NUM_EPOCHS, PATIENCE, train_dataloader, val_dataloader)


Epoch: 1 | train_loss: 1.18125, val_loss: 0.93827, lr: 1.00E-04, _patience: 10
Epoch: 2 | train_loss: 0.81291, val_loss: 0.72564, lr: 1.00E-04, _patience: 10
Epoch: 3 | train_loss: 0.65413, val_loss: 0.64487, lr: 1.00E-04, _patience: 10
...
Epoch: 23 | train_loss: 0.30351, val_loss: 0.53904, lr: 1.00E-06, _patience: 3
Epoch: 24 | train_loss: 0.30332, val_loss: 0.53912, lr: 1.00E-07, _patience: 2
Epoch: 25 | train_loss: 0.30300, val_loss: 0.53909, lr: 1.00E-07, _patience: 1
Stopping early!

Evaluation

1
from pathlib import Path

1
2
3
# Get predictions
test_loss, y_true, y_prob = trainer.eval_step(dataloader=test_dataloader)
y_pred = np.argmax(y_prob, axis=1)

1
2
3
4
# Determine performance
performance = get_metrics(
    y_true=y_test, y_pred=y_pred, classes=label_encoder.classes)
print (json.dumps(performance["overall"], indent=2))


{
  "precision": 0.8192635071011053,
  "recall": 0.8196111111111111,
  "f1": 0.8192710197821547,
  "num_samples": 18000.0
}

1
2
3
4
5
6
7
8
# Save artifacts
dir = Path("gru")
dir.mkdir(parents=True, exist_ok=True)
label_encoder.save(fp=Path(dir, "label_encoder.json"))
tokenizer.save(fp=Path(dir, 'tokenizer.json'))
torch.save(best_model.state_dict(), Path(dir, "model.pt"))
with open(Path(dir, 'performance.json'), "w") as fp:
    json.dump(performance, indent=2, sort_keys=False, fp=fp)

Inference

1
2
3
4
5
6
7
8
def get_probability_distribution(y_prob, classes):
    """Create a dict of class probabilities from an array."""
    results = {}
    for i, class_ in enumerate(classes):
        results[class_] = np.float64(y_prob[i])
    sorted_results = {k: v for k, v in sorted(
        results.items(), key=lambda item: item[1], reverse=True)}
    return sorted_results

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
# Load artifacts
device = torch.device("cpu")
label_encoder = LabelEncoder.load(fp=Path(dir, "label_encoder.json"))
tokenizer = Tokenizer.load(fp=Path(dir, 'tokenizer.json'))
model = GRU(
    embedding_dim=EMBEDDING_DIM, vocab_size=VOCAB_SIZE,
    rnn_hidden_dim=RNN_HIDDEN_DIM, hidden_dim=HIDDEN_DIM,
    dropout_p=DROPOUT_P, num_classes=NUM_CLASSES)
model.load_state_dict(torch.load(Path(dir, "model.pt"), map_location=device))
model.to(device)


GRU(
  (embeddings): Embedding(5000, 100, padding_idx=0)
  (rnn): GRU(100, 128, batch_first=True, bidirectional=True)
  (dropout): Dropout(p=0.1, inplace=False)
  (fc1): Linear(in_features=256, out_features=100, bias=True)
  (fc2): Linear(in_features=100, out_features=4, bias=True)
)


1
2
# Initialize trainer
trainer = Trainer(model=model, device=device)

1
2
3
4
5
6
7
# Dataloader
text = "The final tennis tournament starts next week."
X = tokenizer.texts_to_sequences([preprocess(text)])
print (tokenizer.sequences_to_texts(X))
y_filler = label_encoder.encode([label_encoder.classes[0]]*len(X))
dataset = Dataset(X=X, y=y_filler)
dataloader = dataset.create_dataloader(batch_size=batch_size)


['final tennis tournament starts next week']

1
2
3
4
# Inference
y_prob = trainer.predict_step(dataloader)
y_pred = np.argmax(y_prob, axis=1)
label_encoder.decode(y_pred)

['Sports']

1
2
3
# Class distributions
prob_dist = get_probability_distribution(y_prob=y_prob[0], classes=label_encoder.classes)
print (json.dumps(prob_dist, indent=2))

{
  "Sports": 0.49753469228744507,
  "World": 0.2925860285758972,
  "Business": 0.1932886838912964,
  "Sci/Tech": 0.01659061387181282
}

We will learn how to create more context-aware representations and a little bit of interpretability with RNNs in the next lesson on attention.

To cite this content, please use:

1
2
3
4
5
6
@article{madewithml,
    author       = {Goku Mohandas},
    title        = { RNNs - Made With ML },
    howpublished = {\url{https://madewithml.com/}},
    year         = {2023}
}