Building Your First Neural Network

# CHAPTER 7

Building Your First Neural Network #

1. Introduction #

2. Learning Objectives #

• Use the Sequential model API in Keras.

• Understand and implement Dense layers.

• Define Input and Output layer shapes.

• Build a multi-layer Neural Network architecture.

• Compile a model for training.

3. The Sequential Model #

import tensorflow as tf
from tensorflow.keras.models import Sequential

# Initialize an empty Sequential model
model = Sequential()

4. Adding Dense Layers #

from tensorflow.keras.layers import Dense, Flatten

# Building the architecture
model = Sequential([
    # Input Layer (Flatten converts a 2D image into a 1D vector of pixels)
    Flatten(input_shape=(28, 28)),
    
    # Hidden Layer 1 (128 neurons, ReLU activation)
    Dense(128, activation='relu'),
    
    # Hidden Layer 2 (64 neurons, ReLU activation)
    Dense(64, activation='relu'),
    
    # Output Layer (10 neurons because we have 10 digits to predict: 0-9)
    Dense(10, activation='softmax')
])

5. Compiling the Model #

1. 1. Optimizer: The algorithm that updates the Weights during Backpropagation (e.g., adam or sgd).

1. 2. Loss Function: The mathematical formula that calculates how wrong the model's guesses are (e.g., sparsecategoricalcrossentropy).

1. 3. Metrics: What we want to monitor during training (e.g., accuracy).

*(We will dive deeply into Activations, Optimizers, and Loss functions in Chapter 8).*

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

6. Mini Project: Handwritten Digit Classifier #

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten

# 1. Load Data
mnist = tf.keras.datasets.mnist
(X_train, y_train), (X_test, y_test) = mnist.load_data()

# 2. Preprocess Data (Scale pixels to be between 0 and 1)
X_train = X_train / 255.0
X_test = X_test / 255.0

# 3. Build Architecture
model = Sequential([
    Flatten(input_shape=(28, 28)),
    Dense(128, activation='relu'),
    Dense(10, activation='softmax')
])

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

# 5. Train! (epochs=5 means the model sees the entire dataset 5 times)
print("Training started...")
model.fit(X_train, y_train, epochs=5)

# 6. Evaluate on unseen Test data
test_loss, test_acc = model.evaluate(X_test, y_test, verbose=2)
print(f"Test Accuracy: {test_acc*100:.2f}%")

7. Analyzing Model Architecture #

model.summary()

This prints a beautiful table showing every layer, the shape of the data passing through it, and exactly how many "Parameters" (Weights and Biases) the model needs to learn. For this simple model, it is over 100,000 parameters!

8. Common Mistakes #

• Forgetting the Input Shape: The very first layer of your model MUST define the inputshape. If you don't tell the model what size the incoming data is (e.g., a 28x28 image), it doesn't know how many weights to create, and the code will crash.