Model Training Fundamentals

Essential workflow for training neural networks in Keras, covering the basic steps from model creation to prediction.

Training Your First Model

Section 1.40 - The Training Workflow

Four Essential Steps

1. Create the model
2. Compile it with optimizer and loss
3. Fit to training data
4. Make predictions

Complete Training Example

import numpy as np
from keras import Sequential, layers

# Generate sample data
X_train = np.random.normal(size=(1000, 20))  # 1000 samples, 20 features
y_train = (X_train.sum(axis=1) > 0).astype(int)  # Binary classification

# Step 1: Create model
model = Sequential([
    layers.Dense(64, activation='relu', input_shape=(20,)),
    layers.Dense(32, activation='relu'),
    layers.Dense(1, activation='sigmoid')
])

# Step 2: Compile
model.compile(
    optimizer='adam',               # Algorithm to update weights
    loss='binary_crossentropy',    # How to measure errors
    metrics=['accuracy']           # What to report during training
)

# Step 3: Fit
history = model.fit(
    X_train, y_train,
    epochs=10,           # Number of training passes
    batch_size=32,       # Samples per gradient update
    validation_split=0.2 # 20% of data for validation
)

# Step 4: Predict
X_new = np.random.normal(size=(100, 20))  # New data
predictions = model.predict(X_new)         # Get model outputs

Section 1.41 - Key Parameters Explained

Compilation Options

model.compile(
    optimizer='adam',    # Common choices: 'adam', 'sgd', 'rmsprop'
    loss='mse',         # For regression: 'mse', 'mae'
                        # For classification: 'binary_crossentropy', 'categorical_crossentropy'
    metrics=['accuracy'] # What to track: 'accuracy', 'mae', custom metrics
)

Training Parameters

model.fit(
    X_train, y_train,        
    epochs=10,              # More epochs = more training (sometimes too much)
    batch_size=32,          # Smaller = slower but more stable
    validation_split=0.2,   # Hold out data to check performance
    shuffle=True,           # Mix data between epochs
    verbose=1              # 0=silent, 1=progress bar, 2=one line per epoch
)

Section 1.42 - Quick Tips

Common Gotchas

1. Input Shape:
   • Must match your data
   • Include batch dimension: (batch_size, features)
2. Data Types:
   • Convert inputs to float32: X_train = X_train.astype('float32')
   • Classification targets to int32: y_train = y_train.astype('int32')
3. Validation:
   • Always use validation data to check for overfitting
   • Either validation_split or separate validation_data=(X_val, y_val)

Note

We’ll dive deeper into:

• Loss function selection
• Optimizer tuning
• Batch size effects
• Validation strategies
• Overfitting prevention

in later sections of the course.

Section 1.43 - Task-Specific Templates

Regression (Predict Numbers)

model = Sequential([
    layers.Dense(64, activation='relu', input_shape=(n_features,)),
    layers.Dense(1)  # No activation for regression
])
model.compile(optimizer='adam', loss='mse', metrics=['mae'])

Binary Classification (Yes/No)

model = Sequential([
    layers.Dense(64, activation='relu', input_shape=(n_features,)),
    layers.Dense(1, activation='sigmoid')  # Output between 0 and 1
])
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

Multiclass Classification (Categories)

model = Sequential([
    layers.Dense(64, activation='relu', input_shape=(n_features,)),
    layers.Dense(n_classes, activation='softmax')  # One probability per class
])
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

Important

These templates provide starting points - you’ll learn to customize them as we progress through the course.