Neural Network Wizard

Introduction

The Neural Network Wizard is a desktop companion to the WiseTrader Toolbox. It lets you design a neural network for the financial markets visually — choosing its inputs, its prediction target and its architecture — and then generates the AmiBroker Formula Language (AFL) that trains the network and plots its prediction on a chart.

Tip

Never built a neural network before? Read Neural Networks for Traders first. It explains, with no maths, what a network is, what "training" does, and the one pitfall that trips up every beginner — overfitting. Every setting in this manual makes more sense once those ideas click.

Note

The Wizard does not run a neural network itself. It is a code generator: it produces ready-to-paste AFL that calls the WiseTrader Toolbox neural-network functions inside AmiBroker. You need the WiseTrader Toolbox plugin installed and licensed for the generated formulas to run.

What the Wizard produces

From a single network design the Wizard emits up to two AFL formulas:

FormulaPurpose
Training FormulaTrains the network on your data — across a pool of symbols if you supply one — and saves the trained weights to a .net file.
Indicator FormulaLoads the trained network and plots its prediction on a chart, so you can use it like any other indicator or as a signal source for a system.

Both formulas use the standard WiseTrader Toolbox functions, so the output is plain AFL you can read, edit and combine with the rest of your trading system. The Wizard can also package the two together as a single Combined Formula — one self-contained file that trains on demand and plots automatically; see Generating the Formula.

Two kinds of network

When you start the Wizard you pick one of two operating modes, which shapes everything that follows:

ModeDescription
StandardTrains once on your data (split into train and test sets) across a pool of symbols, then reuses the saved network. Best for most users and the only mode that supports the sequence models (LSTM/GRU/TCN/Transformer) and network-to-AFL export.
Walk-ForwardRe-trains continuously on a rolling window of recent bars on a single symbol. More adaptive but slower, single-symbol, and produces no saved network.
Tip

If you are new to neural networks in AmiBroker, start with a Standard network. It is faster to iterate on and works the same way as the rest of this guide unless a section is explicitly marked Walk-Forward.

What's in this guide