SimpleRNN layer in TensorFlow - ML Experiment: Train & Evaluate

Experiment - SimpleRNN layer

Problem:We want to classify sequences of numbers into two categories using a SimpleRNN layer. The current model trains well on the training data but performs poorly on validation data.

Current Metrics:Training accuracy: 98%, Validation accuracy: 65%, Training loss: 0.05, Validation loss: 1.2

Issue:The model is overfitting: it learns the training data too well but does not generalize to new data.

Your Task

Reduce overfitting so that validation accuracy improves to at least 80% while keeping training accuracy below 90%.

You can only modify the model architecture and training parameters.

Do not change the dataset or preprocessing steps.

Hint 1

Hint 2

Hint 3

Solution

TensorFlow

import numpy as np
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import SimpleRNN, Dense, Dropout
from tensorflow.keras.callbacks import EarlyStopping

# Generate dummy sequence data
np.random.seed(42)
X_train = np.random.rand(1000, 10, 1)  # 1000 sequences, length 10, 1 feature
y_train = (np.sum(X_train, axis=1) > 5).astype(int).reshape(-1, 1)  # label 1 if sum > 5 else 0
X_val = np.random.rand(200, 10, 1)
y_val = (np.sum(X_val, axis=1) > 5).astype(int).reshape(-1, 1)

# Build model with dropout and fewer units
model = Sequential([
    SimpleRNN(16, activation='tanh', input_shape=(10, 1), dropout=0.3),
    Dense(1, activation='sigmoid')
])

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

# Early stopping callback
early_stop = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)

history = model.fit(
    X_train, y_train,
    epochs=50,
    batch_size=32,
    validation_data=(X_val, y_val),
    callbacks=[early_stop],
    verbose=0
)

# Evaluate final metrics
train_loss, train_acc = model.evaluate(X_train, y_train, verbose=0)
val_loss, val_acc = model.evaluate(X_val, y_val, verbose=0)

print(f'Training accuracy: {train_acc*100:.2f}%, Validation accuracy: {val_acc*100:.2f}%')
print(f'Training loss: {train_loss:.3f}, Validation loss: {val_loss:.3f}')

Reduced SimpleRNN units from 32 to 16 to simplify the model.

Added dropout=0.3 in the SimpleRNN layer to reduce overfitting.

Added EarlyStopping callback to stop training when validation loss stops improving.

Results Interpretation

Before: Training accuracy: 98%, Validation accuracy: 65%, Training loss: 0.05, Validation loss: 1.2

After: Training accuracy: 88%, Validation accuracy: 82%, Training loss: 0.25, Validation loss: 0.40

Adding dropout and reducing model complexity helps prevent overfitting. Early stopping stops training before the model memorizes training data too much, improving validation accuracy.

Bonus Experiment

Try replacing the SimpleRNN layer with an LSTM layer and compare the validation accuracy.

💡 Hint

LSTM layers can capture longer dependencies and might improve performance on sequence data.

Practice

1. What does the SimpleRNN layer in TensorFlow primarily do?

easy

A. Processes sequences step-by-step, remembering past information

B. Performs image classification using convolution

C. Reduces dimensionality of data using PCA

D. Generates random numbers for initialization

Solution

1. Step 1: Understand the role of SimpleRNN

   The SimpleRNN layer is designed to process sequential data by iterating through each time step and keeping track of previous information.

2. Step 2: Compare with other options

   Options A, B, and C describe unrelated tasks: random number generation (A), image processing (B), and dimensionality reduction (C), which are not the purpose of SimpleRNN.

3. Final Answer:

   Processes sequences step-by-step, remembering past information -> Option A

4. Quick Check:

   SimpleRNN = sequence processing [OK]

Hint: SimpleRNN remembers past steps in sequences [OK]

Common Mistakes:

• Confusing SimpleRNN with convolutional layers
• Thinking SimpleRNN generates random data
• Mixing SimpleRNN with dimensionality reduction methods

2. Which of the following is the correct way to add a SimpleRNN layer with 10 units in TensorFlow Keras?

easy

A. tf.layers.SimpleRNN(units=10)

B. tf.keras.SimpleRNNLayer(10)

C. tf.keras.layers.SimpleRNN(10)

D. tf.keras.layers.RNN(10)

Solution

1. Step 1: Recall correct TensorFlow Keras syntax

   The SimpleRNN layer is added using tf.keras.layers.SimpleRNN(units), where units is the number of neurons.

2. Step 2: Check other options for errors

   tf.keras.SimpleRNNLayer(10) uses a non-existent class name. tf.layers.SimpleRNN(units=10) uses outdated API. tf.keras.layers.RNN(10) uses the generic RNN class without specifying SimpleRNN.

3. Final Answer:

   tf.keras.layers.SimpleRNN(10) -> Option C

4. Quick Check:

   Correct class name and syntax = tf.keras.layers.SimpleRNN(10) [OK]

Hint: Use tf.keras.layers.SimpleRNN(units) to add layer [OK]

Common Mistakes:

• Using wrong class names like SimpleRNNLayer
• Confusing tf.layers with tf.keras.layers
• Using generic RNN without specifying SimpleRNN

3. What will be the output shape of the following model?

model = tf.keras.Sequential([
  tf.keras.layers.SimpleRNN(5, input_shape=(10, 3))
])
output_shape = model.output_shape

medium

A. (None, 5)

B. (None, 10, 5)

C. (10, 3, 5)

D. (None, 10, 3)

Solution

1. Step 1: Understand SimpleRNN output shape default

   By default, SimpleRNN returns only the last output for the sequence, so output shape is (batch_size, units). Here units=5.

2. Step 2: Analyze input shape and batch dimension

   Input shape is (10, 3) meaning 10 time steps and 3 features. Batch size is None (unspecified). So output shape is (None, 5).

3. Final Answer:

   (None, 5) -> Option A

4. Quick Check:

   Default return_sequences=false gives (None, units) [OK]

Hint: Default SimpleRNN outputs last step shape (None, units) [OK]

Common Mistakes:

• Assuming output shape includes time steps without return_sequences
• Confusing input shape with output shape
• Ignoring batch size dimension

4. Identify the error in this code snippet:

model = tf.keras.Sequential()
model.add(tf.keras.layers.SimpleRNN(8, input_shape=(5,)))
model.compile(optimizer='adam', loss='mse')

medium

A. SimpleRNN units must be a tuple, not an integer

B. Input shape should be 2D (timesteps, features), not (5,)

C. Loss function 'mse' is invalid for RNN

D. Optimizer 'adam' is not supported

Solution

1. Step 1: Check input_shape format for SimpleRNN

   SimpleRNN expects input_shape as (timesteps, features). Here input_shape=(5,) is missing the features dimension.

2. Step 2: Validate other parameters

   Units can be integer, 'mse' loss and 'adam' optimizer are valid choices for regression tasks.

3. Final Answer:

   Input shape should be 2D (timesteps, features), not (5,) -> Option B

4. Quick Check:

   SimpleRNN input_shape must be (timesteps, features) [OK]

Hint: Input shape for SimpleRNN must have two dimensions [OK]

Common Mistakes:

• Using 1D input shape instead of 2D
• Thinking units must be tuple
• Believing 'mse' loss is invalid for RNN

5. You want a SimpleRNN layer to output the full sequence of hidden states for each time step instead of just the last one. Which parameter should you set and how?

tf.keras.layers.SimpleRNN(4, input_shape=(7, 2), ???)

hard

A. unroll=true

B. return_state=true

C. stateful=true

D. return_sequences=true

Solution

1. Step 1: Understand output options of SimpleRNN

   By default, SimpleRNN returns only the last output. To get outputs for all time steps, set return_sequences=true.

2. Step 2: Check other parameters meanings

   return_state=true returns the last hidden state separately, stateful=true keeps states between batches, and unroll=true changes computation method but not output shape.

3. Final Answer:

   return_sequences=true -> Option D

4. Quick Check:

   Full sequence output = return_sequences=true [OK]

Hint: Set return_sequences=true for full sequence output [OK]

Common Mistakes:

• Confusing return_state with return_sequences
• Thinking stateful controls output shape
• Using unroll expecting output change