Dataset & DataLoader

Load and process data for neural network training.

---

Dataset Pipeline

module NN;

data is NN.LoadCSV("train.csv");
loader is NN.DataLoader(data, batchSize: 32, shuffle: true);

for (batch in loader) {
    predictions is model.Forward(batch.input);
    loss is NN.CrossEntropy(predictions, batch.labels);
    loss.Backward();
    optimizer.Step();
}

---

Data Loading

Method	Description
NN.LoadCSV(path)	Load data from CSV file
NN.LoadTensor(path)	Load serialized tensor data

---

DataLoader

The DataLoader handles:

• Batching — Splits data into mini-batches
• Shuffling — Randomize order each epoch
• Iteration — Works with for-in loops

---

Next Steps

• GPU Training — Accelerate training on GPU