Input shape specification in TensorFlow - ML Experiment: Train & Evaluate

Experiment - Input shape specification

Problem:You are building a neural network to classify images of size 28x28 pixels in grayscale. The current model does not specify the input shape correctly, causing errors or poor training.

Current Metrics:Model training fails or accuracy is very low due to incorrect input shape.

Issue:The input shape is not properly defined in the first layer of the model, leading to shape mismatch errors or inability to train.

Your Task

Correctly specify the input shape in the model so it matches the data shape (28x28 grayscale images). The model should train successfully and achieve at least 80% accuracy on the test set.

Do not change the dataset or model architecture except for input shape.

Use TensorFlow and Keras only.

Hint 1

Hint 2

Hint 3

Solution

TensorFlow

import tensorflow as tf
from tensorflow.keras import layers, models

# Load MNIST dataset
(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.mnist.load_data()

# Reshape data to add channel dimension (grayscale = 1 channel)
train_images = train_images.reshape((-1, 28, 28, 1)).astype('float32') / 255.0
test_images = test_images.reshape((-1, 28, 28, 1)).astype('float32') / 255.0

# Build model with correct input shape
model = models.Sequential([
    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
    layers.MaxPooling2D((2, 2)),
    layers.Flatten(),
    layers.Dense(64, activation='relu'),
    layers.Dense(10, activation='softmax')
])

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

# Train model
model.fit(train_images, train_labels, epochs=5, batch_size=64, validation_split=0.2)

# Evaluate model
test_loss, test_acc = model.evaluate(test_images, test_labels)
print(f'Test accuracy: {test_acc:.4f}')

Added input_shape=(28, 28, 1) to the first Conv2D layer to match grayscale image shape.

Reshaped training and test images to have shape (num_samples, 28, 28, 1).

Normalized pixel values to range 0-1 by dividing by 255.

Results Interpretation

Before: Model training failed or accuracy was very low due to input shape mismatch.

After: Model trains successfully and achieves about 85% accuracy on test data.

Specifying the correct input shape is essential for the model to understand the data format and train properly. For image data, the input shape must include height, width, and channels.

Bonus Experiment

Try changing the input shape to (28, 28) without the channel dimension and observe what error or behavior occurs.

💡 Hint

TensorFlow Conv2D layers expect 3D input (height, width, channels). Omitting channels causes shape mismatch errors.

Practice

1. What does the input_shape parameter specify in a TensorFlow Keras model?

easy

A. The size and format of the input data the model expects

B. The number of layers in the model

C. The learning rate for training

D. The number of output classes

Solution

1. Step 1: Understand the role of input_shape

   The input_shape tells the model what size and type of data it will receive as input.

2. Step 2: Differentiate from other parameters

   Other parameters like number of layers, learning rate, or output classes do not describe input data format.

3. Final Answer:

   The size and format of the input data the model expects -> Option A

4. Quick Check:

   input_shape = data size/type [OK]

Hint: Input shape defines data size, not layers or learning rate [OK]

Common Mistakes:

• Confusing input_shape with number of layers
• Thinking input_shape sets learning rate
• Mixing input_shape with output classes

2. Which of the following is the correct way to specify an input shape for a model expecting 28x28 grayscale images in TensorFlow?

easy

A. tf.keras.layers.Input(shape=(28, 28))

B. tf.keras.layers.Input(shape=(28, 28, 1))

C. tf.keras.layers.Input(shape=(1, 28, 28))

D. tf.keras.layers.Input(shape=(784,))

Solution

1. Step 1: Identify the correct shape for grayscale images

   Grayscale images have height, width, and 1 channel, so shape is (28, 28, 1).

2. Step 2: Check each option

   tf.keras.layers.Input(shape=(28, 28, 1)) matches (28, 28, 1). tf.keras.layers.Input(shape=(28, 28)) misses channel dimension. tf.keras.layers.Input(shape=(1, 28, 28)) has wrong channel position. tf.keras.layers.Input(shape=(784,)) flattens input, not raw shape.

3. Final Answer:

   tf.keras.layers.Input(shape=(28, 28, 1)) -> Option B

4. Quick Check:

   Grayscale image shape = (height, width, 1) [OK]

Hint: Grayscale images need channel=1 as last dimension [OK]

Common Mistakes:

• Omitting the channel dimension
• Placing channel dimension first incorrectly
• Using flattened input shape instead of 2D+channel

3. What will be the output shape of the following TensorFlow model's first layer?

model = tf.keras.Sequential([
  tf.keras.layers.Input(shape=(32, 32, 3)),
  tf.keras.layers.Conv2D(16, 3)
])

medium

A. (None, 32, 32, 3)

B. (None, 32, 32, 16)

C. (None, 30, 30, 3)

D. (None, 30, 30, 16)

Solution

1. Step 1: Understand Conv2D output shape calculation

   Conv2D with kernel size 3 and default 'valid' padding reduces height and width by 2 (3-1) each.

2. Step 2: Calculate output dimensions

   Input shape is (32, 32, 3). Output height and width = 32 - 3 + 1 = 30. Number of filters = 16, so output shape is (None, 30, 30, 16).

3. Final Answer:

   (None, 30, 30, 16) -> Option D

4. Quick Check:

   Conv2D valid padding reduces size by kernel-1 [OK]

Hint: Valid padding shrinks size by kernel_size - 1 [OK]

Common Mistakes:

• Assuming output size equals input size without padding
• Confusing number of channels with number of filters
• Ignoring batch size dimension (None)

4. Identify the error in this TensorFlow input layer code:

input_layer = tf.keras.layers.Input(shape=28, 28, 1)

medium

A. The shape argument should be a single tuple, not separate values

B. The input layer must specify batch size explicitly

C. The channel dimension should be first, not last

D. Input layers cannot have 3D shapes

Solution

1. Step 1: Check the syntax of shape argument

   The shape parameter must be a single tuple, e.g., (28, 28, 1), not separate arguments.

2. Step 2: Verify other options

   Batch size is optional and not required here. Channel last is standard. Input layers can have 3D shapes for images.

3. Final Answer:

   The shape argument should be a single tuple, not separate values -> Option A

4. Quick Check:

   Shape must be tuple like (28, 28, 1) [OK]

Hint: Use parentheses to group shape dimensions as a tuple [OK]

Common Mistakes:

• Passing shape dimensions as separate arguments
• Forcing batch size in input shape
• Misplacing channel dimension

5. You want to build a model that accepts variable-length sequences of 10 features each. Which input shape specification is correct for the first layer?

hard

A. tf.keras.layers.Input(shape=(10,))

B. tf.keras.layers.Input(shape=(10, None))

C. tf.keras.layers.Input(shape=(None, 10))

D. tf.keras.layers.Input(shape=(None,))

Solution

1. Step 1: Understand variable-length sequences

   Variable-length means the sequence length is unknown, so use None for that dimension.

2. Step 2: Identify feature dimension position

   Each sequence element has 10 features, so feature dimension is fixed at 10, sequence length is variable.

3. Step 3: Match shape to (sequence_length, features)

   The correct shape is (None, 10), meaning variable sequence length and fixed 10 features per step.

4. Final Answer:

   tf.keras.layers.Input(shape=(None, 10)) -> Option C

5. Quick Check:

   Variable length = None in first dimension [OK]

Hint: Use None for variable dimension, fixed size for features [OK]

Common Mistakes:

• Swapping sequence length and feature dimensions
• Using fixed size for variable-length dimension
• Omitting feature dimension