AI Mode

Step 1 : Select a categorical target

A categorical target looks ahead in the data and labels each candle; the model then learns to predict those labels from your indicators. Choose one of six:

Step 2 : Configure the target settings

Each target exposes its own parameters as sliders. The first three are the target's main thresholds and its forward-looking window — for Stoploss-Takeprofit Based these are Stop Loss, Take Profit and Max Period; for Rally Based, Buy Margin, Sell Margin and Max Period; for Normalized Stoploss-Takeprofit Based, ATR Multiplier, Reward:Risk and Max Period. Targets with additional parameters show them as extra sliders alongside.

These settings decide what counts as a good trade for this signal. A modest take-profit over a long window produces many positive labels and a model that signals often; a demanding reward-to-risk ratio produces few, and a model that signals rarely but with more conviction.

A preview chart updates as you move the sliders, so you can see the labelling the model will be trained on before committing to it. For what each individual parameter means, follow the link for your chosen target in the table above.

Step 3 : Advanced ML Engine Settings

This optional panel controls how the model is trained. It is the wizard's view of the model and feature-processor configuration described in New Model; left untouched, training uses sensible defaults.

You would typically change these when a model underfits or overfits — increasing memory to give it more context, or switching scaler when features have very different ranges.

Step 4 : Select the indicators

The indicators you pick here are the inputs (features) the model learns from. They are grouped by timeframe: one selector per timeframe configured in Main Settings, so an indicator computed on the base timeframe and the same indicator on a helper timeframe are chosen separately.

The available indicators come from your market data's Features, and you can select at most 40 in total across all timeframes.

Add Custom Indicator lets you build one that is not in the list, choosing the indicator, its window length and its input settings. Custom indicators are built with the same step-based builder as Manual signals — see IQ Wizard: Manual Signal Builder — and once created they appear alongside the predefined indicators for the timeframe you built them on. A custom indicator you expect to reuse can be saved as a template and loaded again from the same builder — see IQ Wizard Templates.

Step 5 : Decision Mode

Decision Mode decides whether the model's decision threshold is shared or calculated per pair:

• Global — one threshold derived from the whole training set and applied to every pair.

• Pairwise — a separate threshold for each pair.

This is the same Mode setting described in Signal Generator, which covers how the thresholds are derived and what happens when a pair has no data in the training range.

Step 6 : Confidence Level

Confidence Level is the minimum certainty the model must have in its own prediction before the signal is allowed to fire.

Raising it produces fewer but higher-conviction signals; lowering it produces more signals of lower average quality.

Step 7 : Save the signal

Two buttons finish the signal:

• Save — adds the signal to the Entry or Exit box you opened the dialog from. This is the action that actually puts the signal into your strategy.

• Save as Template — stores the configuration for reuse. Templates are available to any future signal, in this strategy or another one; see IQ Wizard Templates.

When you open the dialog from an existing template, these become Save Template (overwrite the template) and Save as New Template (keep the original and store a copy).

Step 1: Select a numerical target

Where Classification targets assign labels, a numerical target assigns each candle a value — a rate of return, a momentum score, a reward-to-risk figure — and the model learns to predict that number. Four are offered:

The output range matters more here than it does for Classification, because it is the scale you will be dividing up in Step 3. The wizard repeats the range for whichever target you select, so you can see what you are working with before defining your ranges.

Step 2: Configure the target settings

The chosen target's parameters appear below it, most importantly its forward-looking window length, which sets how far ahead the target measures. As with Classification, an illustration updates alongside your settings. Follow the target's link above for the meaning of each parameter.

Step 3: Define prediction ranges and labels

This step has no equivalent in Classification. The model outputs a number, and this table turns that number into a signal by mapping ranges onto labels. Each row takes:

Add one row per range you want to distinguish. A Not Match Value is applied to any prediction that falls outside every range you defined, so nothing is left unlabelled.

Step 4: Configure the ML engine

Identical to Classification — the same Advanced ML Engine Settings panel covering model, memory, sampling and scaler. See Step 3 — Advanced ML Engine Settings.

Step 5: Select the indicators

Identical to Classification: indicators grouped per timeframe, the same limit on how many can be selected, and the same Add Custom Indicator option for building one with your own indicator, window length and input settings.

Step 6: Save the signal

The same Save and Save as Template buttons described in Step 7 — Save the signal.