Keras as TensorFlow's high-level API - Model Pipeline Trace

Model Pipeline - Keras as TensorFlow's high-level API

This pipeline shows how Keras, a simple and friendly tool inside TensorFlow, helps build and train a model to recognize handwritten digits. It takes raw images, prepares them, learns patterns, and then predicts new digits.

Data Flow - 7 Stages

1Data Loading

70000 rows x 28 x 28 pixels→Load MNIST handwritten digits dataset→70000 rows x 28 x 28 pixels

Image of digit '5' as 28x28 grayscale pixels

↓

2Preprocessing

70000 rows x 28 x 28 pixels→Normalize pixel values from 0-255 to 0-1→70000 rows x 28 x 28 pixels

Pixel value 150 becomes 0.588

↓

3Train/Test Split

70000 rows x 28 x 28 pixels→Split dataset into training (60000) and testing (10000)→60000 rows x 28 x 28 pixels (train), 10000 rows x 28 x 28 pixels (test)

Training image of digit '3', test image of digit '7'

↓

4Model Building

60000 rows x 28 x 28 pixels→Define a simple neural network with Flatten, Dense layers→Model ready to train on input shape (28,28)

Input layer flattens 28x28 to 784 features

↓

5Model Training

60000 rows x 28 x 28 pixels→Train model with labels for 5 epochs→Trained model with learned weights

Model learns to recognize digit '1' patterns

↓

6Evaluation

10000 rows x 28 x 28 pixels→Evaluate model accuracy on test set→Accuracy score (e.g., 0.98)

Model correctly predicts 9800 out of 10000 digits

↓

7Prediction

1 row x 28 x 28 pixels→Model predicts digit class probabilities→Array of 10 probabilities summing to 1

Prediction: [0.01, 0.02, 0.85, ..., 0.01] means digit '2' likely

Training Trace - Epoch by Epoch

Loss
0.4 | *
0.3 |  *
0.2 |   *
0.1 |    *
0.0 |     *
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.35	0.90	Model starts learning basic digit shapes
2	0.18	0.95	Accuracy improves as model refines features
3	0.12	0.97	Model captures more details, fewer mistakes
4	0.09	0.98	Loss decreases steadily, accuracy near perfect
5	0.07	0.98	Training converges with high accuracy

Prediction Trace - 4 Layers

Layer 1: Input Layer

Layer 2: Dense Layer 1 (ReLU)

Layer 3: Dense Layer 2 (Softmax)

Layer 4: Prediction Output

Model Quiz - 3 Questions

What does the preprocessing step do to the pixel values?

A Scales pixel values to between 0 and 1

B Converts images to color

C Removes half of the pixels

D Changes image size to 14x14

Key Insight

Keras simplifies building and training neural networks by providing easy-to-use layers and training loops. This helps beginners quickly create models that learn from data and improve accuracy over time.