Neural Networks

The WiseTrader Toolbox includes a fully configurable neural network engine you can train and run directly from AmiBroker formulas. It supports both feed-forward (MLP) and recurrent (LSTM and GRU) networks, a wide range of modern training algorithms, and built-in tools for controlling overfitting — the single biggest factor in whether a network's predictions are any good.

If you are new to this, read the pages in order. If you just need a setting, jump to the Settings Reference.

Learn the concepts

Introduction — what a neural network is, the train → predict workflow, and the one idea that matters most (overfitting).
Architectures: MLP, LSTM & GRU — feed-forward versus recurrent networks, and how to size and choose one.
Activation Functions — the building blocks of each layer and when to use each.
Input & Output Scaling — how to prepare your data so the network can learn from it.

Train a good network

Training Algorithms — the optimizers (iRPROP+, Adam, AdamW and more), learning-rate schedules and minibatch training.
Accuracy & Overfitting — the step-by-step recipe for predictions that hold up on new data.

Reference & examples

Settings Reference — every configuration function in one place.
Neural Network Functions — the training, prediction and indicator functions you call from AFL.
Tutorial 1: Predicting Returns — a feed-forward network that predicts the next bar's return, and the habits that keep predictions honest.
Tutorial 2: Direction Classifier — a practical up/down classifier using several indicators, with a first real pass at overfitting control.
Tutorial 3: Advanced Techniques — a recurrent LSTM that reads a sequence of bars to anticipate the next swing, plus the modern training settings for harder problems.
Tutorial 4: Filtering Trade Signals — the most practical use of a network in trading: instead of predicting price, let it filter the entries of a system you already trade, keeping the likely winners. The most accessible technique here, and usually the one that works best.