Text preprocessing for RNNs in PyTorch - Model Metrics & Evaluation

Metrics & Evaluation - Text preprocessing for RNNs

Which metric matters for this concept and WHY

When preparing text for RNNs, the key metrics to watch are sequence length consistency and vocabulary coverage. These ensure the model receives clean, uniform input sequences and understands the words it sees. For model evaluation, accuracy or loss during training shows if preprocessing helped the RNN learn well.

Confusion matrix or equivalent visualization (ASCII)

    Example confusion matrix for text classification after preprocessing:

          Predicted
          Pos   Neg
    Actual
    Pos   85    15
    Neg   10    90

    TP=85, FP=10, TN=90, FN=15
    Total samples = 85+10+90+15 = 200
    

Precision vs Recall tradeoff with concrete examples

In text tasks, like spam detection, precision means how many flagged messages are truly spam. High precision avoids marking good emails as spam.

Recall means how many actual spam messages are caught. High recall avoids missing spam.

Preprocessing affects this tradeoff: poor tokenization or missing words can lower recall by hiding spam clues. Overly aggressive cleaning might remove important words, hurting precision.

What "good" vs "bad" metric values look like for this use case

Good preprocessing leads to:

• High accuracy (e.g., >85%) on validation data
• Balanced precision and recall (both >80%)
• Stable loss decreasing over epochs

Bad preprocessing causes:

• Low accuracy (<60%) or unstable training
• Very low recall or precision (e.g., <50%)
• Overfitting or underfitting signs

Metrics pitfalls (accuracy paradox, data leakage, overfitting indicators)

• Accuracy paradox: High accuracy can be misleading if classes are imbalanced (e.g., many non-spam emails).
• Data leakage: Using test data during preprocessing (like fitting tokenizer on all data) inflates metrics falsely.
• Overfitting: Very low training loss but high validation loss means preprocessing or model is too tailored to training data.
• Ignoring sequence length: Not padding/truncating sequences properly can cause inconsistent input and poor model performance.

Self-check question

Your RNN text classifier has 98% accuracy but only 12% recall on spam messages. Is it good for production? Why or why not?

Answer: No, it is not good. The model misses most spam messages (low recall), which is critical for spam detection. High accuracy is misleading here because most emails are not spam, so the model just predicts non-spam well but fails to catch spam.

Key Result

For RNN text preprocessing, balanced precision and recall above 80% indicate good input preparation and model learning.

Practice

1. Why do we split text into tokens before feeding it to an RNN?

easy

A. Because RNNs process sequences of numbers, not raw text

B. To reduce the size of the dataset

C. To make the text look nicer

D. Because tokens are easier to print

Solution

1. Step 1: Understand RNN input requirements

   RNNs work with sequences of numbers, not raw text strings.

2. Step 2: Role of tokenization

   Splitting text into tokens converts sentences into smaller units that can be mapped to numbers.

3. Final Answer:

   Because RNNs process sequences of numbers, not raw text -> Option A

4. Quick Check:

   Tokenization = Convert text to numbers [OK]

Hint: RNNs need numbers, so split text into tokens first [OK]

Common Mistakes:

• Thinking tokens are for making text prettier
• Believing tokenization reduces dataset size
• Confusing tokens with characters

2. Which PyTorch function is commonly used to pad sequences to the same length for batch processing?

easy

A. torch.nn.utils.rnn.pad_sequence

B. torch.tensor.pad

C. torch.pad_sequences

D. torch.nn.pad

Solution

1. Step 1: Identify PyTorch padding utilities

   PyTorch provides pad_sequence in torch.nn.utils.rnn to pad variable-length sequences.

2. Step 2: Check other options

   Functions like torch.tensor.pad or torch.nn.pad do not exist; torch.pad_sequences is not a PyTorch function.

3. Final Answer:

   torch.nn.utils.rnn.pad_sequence -> Option A

4. Quick Check:

   Use pad_sequence to pad RNN inputs [OK]

Hint: Remember: pad_sequence is in torch.nn.utils.rnn [OK]

Common Mistakes:

• Using non-existent torch.pad_sequences
• Confusing tensor.pad with pad_sequence
• Trying to pad manually without this function

3. Given the following code, what is the shape of the padded batch tensor?

import torch
from torch.nn.utils.rnn import pad_sequence

seq1 = torch.tensor([1, 2, 3])
seq2 = torch.tensor([4, 5])
seq3 = torch.tensor([6])
batch = pad_sequence([seq1, seq2, seq3], batch_first=True, padding_value=0)
print(batch.shape)

medium

A. (1, 3)

B. (3, 1)

C. (3, 3)

D. (3, 6)

Solution

1. Step 1: Understand input sequences

   Sequences have lengths 3, 2, and 1 respectively.

2. Step 2: pad_sequence with batch_first=true

   All sequences are padded to length 3 (max length), batch dimension is first, so shape is (3 sequences, 3 elements each).

3. Final Answer:

   (3, 3) -> Option C

4. Quick Check:

   Batch size = 3, max seq length = 3 [OK]

Hint: Batch shape = (number sequences, max sequence length) [OK]

Common Mistakes:

• Confusing batch_first=true with false
• Assuming padding adds length beyond max sequence
• Mixing up batch and sequence dimensions

4. What is wrong with this code snippet for preparing text sequences for an RNN?

import torch
from torch.nn.utils.rnn import pad_sequence

sentences = [[1, 2, 3, 4], [5, 6], [7]]
tensors = [torch.tensor(s) for s in sentences]
padded = pad_sequence(tensors)
print(padded.shape)

medium

A. torch.tensor cannot convert lists to tensors

B. pad_sequence is missing batch_first=true, so shape is unexpected

C. pad_sequence requires padding_value argument

D. The input lists must be numpy arrays, not lists

Solution

1. Step 1: Check pad_sequence default behavior

   By default, pad_sequence returns tensor with shape (max_seq_len, batch_size), not batch first.

2. Step 2: Effect on output shape

   Without batch_first=true, the printed shape will be (4, 3) instead of expected batch-first (3, 4) shape.

3. Final Answer:

   pad_sequence is missing batch_first=true, so shape is unexpected -> Option B

4. Quick Check:

   Use batch_first=true for (batch, seq_len) shape [OK]

Hint: Always add batch_first=true for batch as first dimension [OK]

Common Mistakes:

• Assuming pad_sequence pads automatically without batch_first
• Thinking torch.tensor can't convert lists
• Believing padding_value is mandatory

5. You have a batch of sentences tokenized as integer lists of different lengths. You want to feed them into an RNN in PyTorch. Which sequence of steps is correct for preprocessing?

hard

A. Tokenize text -> Pad sequences -> Convert tokens to integers -> Feed to RNN

B. Pad raw text strings -> Tokenize padded strings -> Convert tokens to integers -> Feed to RNN

C. Convert raw text to tensor -> Tokenize tensor -> Pad sequences -> Feed to RNN

D. Tokenize text -> Convert tokens to integers -> Pad sequences with pad_sequence(batch_first=true) -> Convert to tensor batch

Solution

1. Step 1: Tokenize text and convert tokens to integers

   First, split text into tokens, then map tokens to integers using a vocabulary.

2. Step 2: Pad sequences and prepare batch tensor

   Pad integer sequences to equal length using pad_sequence with batch_first=true, then feed the tensor batch to the RNN.

3. Final Answer:

   Tokenize text -> Convert tokens to integers -> Pad sequences with pad_sequence(batch_first=true) -> Convert to tensor batch -> Option D

4. Quick Check:

   Tokenize -> Integer map -> Pad -> Batch tensor [OK]

Hint: Tokenize first, then integer map, then pad sequences [OK]

Common Mistakes:

• Padding raw text instead of token integers
• Converting raw text directly to tensor
• Padding before converting tokens to integers