Quick Panel Guide

AiDataTaskRunner is a tool (panel) that allows you to generate and train ML models for trading bots. This guide walks you through the entire process — from data generation to model training.

Requirements

Have the latest functional release of the repo installed.
A compatible bot — in this guide we'll use EasySbAi: https://forge.mql5.io/nique_372/EasySbAi
A main working folder named EasySbAI (the folder name depends on how you named the bot in EasySbAi). Inside, files must follow this structure:

EasySbAi/
├── XAUUSD/
│   ├── M1/
│   │   ├── config_a.set
│   │   └── config_b.set
│   └── H4/
│       └── config_c.set
└── GBPUSD/
    └── D1/
        └── config_d.set
        .
        .
        .
        (other .set files can go here)

At this point, you should already have optimized .set files for specific timeframes and symbols.

Pre-step (Optional)

This step only applies if you are using calendar features. If so, run the script:

Generador Archivos CSV Calendario.ex5

Starting State

After running a few optimizations and creating the working folder (main folder), the initial structure looks like this:

EasySbAi/
└── XAUUSD/
    └── M5/
        ├── bueno.set
        ├── decente.set
        ├── conservador.set
        └── label.set

Step-by-Step

1. Run the Runner

Execute Runner.ex5 on any chart.

2. Run the Panel

Launch the panel, fill in all parameters with the correct paths, and execute it.

Note: If you installed the panel via a release, you may have placed it under Experts\ or a subfolder. In that case, you will need to update the EA parameters, since the defaults point to the project's own directory paths.

3. Configure the YAML Workflow

Go to the "Workflows" section (or "Flujos de trabajo" in Spanish). Click the ? icon and confirm by clicking Accept.

The YAML file will be loaded into the panel. Click Open — this will launch your default application for .yaml files. If you don't have one, you can install VS Code with a YAML extension (recommended), or use Notepad for a quicker start.

The loaded YAML content matches this template: https://forge.mql5.io/nique_372/AiDataTaskRuner/src/branch/main/FilesInstaler/res/templates/basic_data_generation.yaml

If you are using a different bot, update the main folder and the bot paths accordingly. If you are using EasySbAI, no changes are needed.

4. Execute the YAML Workflow

Close VS Code (or your editor), return to the panel, and click the button below the "Workflow" label.

Once execution finishes, an alert will appear (behavior depends on your YAML configuration). This means the workflow has completed. You can then go to the "View" tab to check execution times and any error codes from failed passes.

5. Prepare the Features for Training

This step depends on how your features were defined during data generation — either embedded in the EA or passed as parameters.

For EasySbAi, features are embedded and generic. Copy the features file from: https://forge.mql5.io/nique_372/EasySbAi/raw/branch/main/Resrc/features.fgblc

At this point in the tutorial, the folder structure should look like this:

EasySbAi/
└── XAUUSD/
    └── M5/       # All subfolders were created by the panel; each contains CSV files,
                  # scaler data from the data generation step, and the corresponding .set file
        ├── bueno_0/
        ├── conservador_0/
        ├── decente_0/
        ├── label_0/
        ├── bueno.set
        ├── decente.set
        ├── conservador.set
        └── label.set

    └── Features/
        └── features_model.fgblc   # Place it here, at the same level as the timeframe folders,
                                   # and rename it according to what is specified in the JSON config.

└── Config/
    └── move_files.json            # Temporary file generated by the workflow,
                                   # used to tell the panel which files to move.

6. Select the Training Config JSON

Navigate to the Training Models section.

Select the JSON configuration file using the first label field. To make this easier, click ? — the file picker will open with the file already pre-selected. Click Open and you'll see something like this:

{
    "general": {
        "_comment": "Place here the relative path to the 'symbol folder'. The panel will automatically train all models across all timeframes and labels.",
        "path": "C:\\Users\\USER\\AppData\\Roaming\\MetaQuotes\\Terminal\\Common\\Files\\EasySbAi\\EURUSD",
        "_comment_1": "You must specify which features file to use for each ML model type. These paths are relative to:",
        "_comment_2": "1. Symbol folder  | 2. Timeframe folder | 3. label_labelid folder — internally a cascade/override system is used.",
        "_comment_3": "The file must be in .csv format. If you have a .fgblc file, use the built-in editor to load and compile it to CSV.",
        "_comment_4": "If the .fgblc is embedded in the EA, extract it to a .fgblc file and place it in FeatureEditor\\My\\",
        "_comment_5": "(path relative to the panel folder), rename it based on the CSV names, load it in the editor, and compile it.",
        "features_pred_file": "Features\\features_model.csv",
        "features_tp_file": "Features\\features_model.csv",
        "features_sl_file": "Features\\features_model.csv",
        "file_name_idx": "idx.txt",
        "file_name_features_ptr": "features_ptr.txt"
    },
    "clasificacion": {
        "target_col": " salida",
        "model_name": "ModelPred",
        "num_features": 25,
        "validation_split": 0.2,
        "n_trials": 75,
        "k_folds": 5,
        "random_seed": 42,
        "hilos": 2,
        "jobs_optuna": 12,
        "final_hilos": 20,
        "data_csv_file": "data_pred.csv"
    },
    "regresion": {
        "target_col": " salida",
        "model_name_tp": "ModelTP",
        "model_name_sl": "ModelSL",
        "num_features": 25,
        "validation_split": 0.2,
        "n_trials": 75,
        "k_folds": 5,
        "random_seed": 42,
        "hilos": 2,
        "jobs_optuna": 12,
        "final_hilos": 20,
        "_comment": "Paths for TP and SL output files. Their location is relative to the MainFolder (e.g. EasySbAi).",
        "data_csv_file_tp": "data_tp.csv",
        "data_csv_file_sl": "data_sl.csv"
    }
}

Two things need to be updated to get started. First, set general::path to point to your symbol folder:

"path": "C:\\Users\\USER\\AppData\\Roaming\\MetaQuotes\\Terminal\\Common\\Files\\EasySbAi\\XAUUSD"

You can also adjust training parameters such as n_trials for Optuna (in this example it's lowered to 40). Pay attention to data_csv_* fields — if your data file names differ from the defaults, update them accordingly. Otherwise, no further changes are needed.

7. Train the Model

Select the Python executable (or .exe) to use for training. In the releases, a pre-built .exe is provided — in this example it's located in the Common folder. Click Browse, navigate to Common, and select it:

Next, optionally select a log file where Python will write its output. This step is optional — the file does not need to exist beforehand, but it must have a .log extension.

Finally, click Run to start the training process and wait for it to complete.