--no-nn

🦉 AthenaAI is an AI-powered ensemble machine-learning trading bot for PocketOption, built on BinaryOptionsToolsV2 → Get it on GitLab
💬 Need help? Join the Chipa Discord · ⚡ Go no-code with ChipaEditor

Skip the PyTorch neural model entirely; train only the scikit-learn ensemble (online SGD/PA/NB plus batch GBM/RF).

What it does

In Trainer.run(), step 6 is guarded by if cfg.use_nn: — with --no-nn, _train_nn() is never called, so no TorchMLPClassifier is created and no MLP entry is added to the ensemble’s batch models. Everything else (online models, GBM/RF, walk-forward test, brain save) runs unchanged; the walk-forward result then reflects the sklearn ensemble alone.

Note that even without --no-nn, the neural model is skipped automatically (with an informational log) when PyTorch is not installed. --no-nn makes the skip intentional and silent about torch.

Choosing a value

• Use it for fast iteration: the MLP’s 30 epochs are usually the slowest training step, especially on CPU. The training guide pairs it with a --max-candles cap for quick experiments.
• Use it if you don’t want a torch dependency in the resulting brain at all. (A brain with an MLP still loads on torch-less machines — the MLP degrades to a neutral 0.5 vote — but contributes nothing there.)
• For deployment brains, leave the NN enabled: its holdout accuracy becomes its vote weight, so a weak net gets less say rather than hurting the ensemble.

Examples

python train.py --no-nn
python train.py --max-candles 100000 --no-nn     # fast experiment loop

Interactions

• Makes --epochs, --batch-size, --lr, --hidden, and --cpu irrelevant (they only affect the neural model; --cpu also changes the logged device banner).
• --check-gpu — check the device you’d be giving up.

See also

• Ensemble concept
• GPU guide
• Training guide

💬 Unsure how this interacts with the rest of the configuration? Ask in the Chipa Discord — or prototype strategies no-code with ChipaEditor.