Validation split in TensorFlow - Model Pipeline Trace

Model Pipeline - Validation split

This pipeline shows how a dataset is split into training and validation parts to help the model learn well and check its performance on unseen data during training.

Data Flow - 4 Stages

1Original dataset

1000 rows x 10 columns→Initial dataset with features and labels→1000 rows x 10 columns

[[5.1, 3.5, ..., 0], [4.9, 3.0, ..., 1], ...]

↓

2Train/Validation split

1000 rows x 10 columns→Split dataset into 80% training and 20% validation→Training: 800 rows x 10 columns, Validation: 200 rows x 10 columns

Training sample: [5.1, 3.5, ..., 0], Validation sample: [6.7, 3.1, ..., 1]

↓

3Model training

Training: 800 rows x 10 columns→Train model on training data→Trained model

Model learns patterns from training samples

↓

4Validation during training

Validation: 200 rows x 10 columns→Evaluate model on validation data each epoch→Validation loss and accuracy metrics

Validation accuracy improves from 60% to 85%

Training Trace - Epoch by Epoch

Loss
0.7 |****
0.6 |*** 
0.5 |**  
0.4 |*   
0.3 |*   
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.65	0.60	Model starts learning; loss is high, accuracy low
2	0.50	0.72	Loss decreases, accuracy improves
3	0.40	0.80	Model learns better features
4	0.35	0.83	Training loss decreases steadily
5	0.30	0.85	Model converges with good accuracy

Prediction Trace - 4 Layers

Layer 1: Input layer

Layer 2: Hidden layer with ReLU

Layer 3: Output layer with Softmax

Layer 4: Prediction

Model Quiz - 3 Questions

Why do we use a validation split during training?

A To speed up training by using less data

B To increase training data size

C To check model performance on unseen data during training

D To test the model after training finishes

Key Insight

Using a validation split helps us see how well the model generalizes to new data during training. It prevents overfitting by giving feedback on unseen data, shown by improving validation accuracy and decreasing loss.