Model Training Pipeline - Bellatrix Capital

5

Available Models

0

Trained Models

-

Best Accuracy

Never

Last Training

Quick Train — Preset Pipelines

Crypto — All Models

Train XGBoost, LightGBM, DNN, LSTM, GNN on 21 crypto assets

~45 min

US Stocks — Tree Models

Train XGBoost & LightGBM on downloaded US stock daily data

~30 min

Deep Learning Suite

Train LSTM & DNN models with hyperparameter tuning

~60 min

Ensemble Builder

Train and combine all model predictions with optimized weights

~90 min

Custom Training Configuration

Select Models to Train

XGBoost LightGBM DNN LSTM GNN

Training Data Source

Target Variable

Feature Set

Lookback Window Rows of history per sample

Step Size Sample every Nth row

Grid Operations

Raw Lookback Values Differences (Δ) Rolling Statistics Rate of Change

Estimated features: ~160 (8 base × 5 lookback × 2 ops + 80 current) Full Feature Config

Train/Test Split

Train: 80% Test: 20%

Hyperparameter Settings

XGBoost Parameters

Max Depth

Learning Rate

N Estimators

Subsample

LightGBM Parameters

DNN Parameters

LSTM Parameters

Enable Bayesian Hyperparameter Optimization

Cross-Validation (5-fold time-series split)

Log to MLflow Tracking

Training in Progress

XGBoost

Waiting

LightGBM

Waiting

DNN

Waiting

LSTM

Waiting

GNN

Waiting

Training Log

[INFO] Training pipeline initialized.

Model Results & Comparison

Model	Data Source	Target	Accuracy	Precision	Recall	F1 Score	Sharpe (Backtest)	Trained At	Status	Actions
No trained models yet. Start a training pipeline above.

Model Accuracy Comparison

Training Loss Curves

MLflow Experiment Tracking

#run-001 XGBoost Accuracy: 0.723 F1: 0.698 Feb 28, 2026

#run-002 LightGBM Accuracy: 0.715 F1: 0.691 Feb 28, 2026

#run-003 LSTM Accuracy: 0.688 F1: 0.672 Feb 27, 2026