--hidden

🦉 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

Hidden layer sizes for the neural model, as a comma-separated list of integers.

What it does

train.py parses the string before building the config:

hidden = tuple(int(h) for h in args.hidden.split(",") if h.strip())

so "128,64" becomes (128, 64); blank segments are ignored (a trailing comma is harmless), but non-numeric segments raise ValueError. The tuple flows into TorchMLPClassifier(hidden=...), and _build_net() (src/training/torch_model.py) builds one block per entry:

input → [Linear(prev, h) → ReLU → Dropout(0.2)]  per h  → Linear(last, 2)

The input size is the feature-vector length (set at fit() time); the output is 2 logits (PUT/CALL). Dropout is fixed at 0.2 per hidden layer.

The architecture is also pickled into the brain: at load time the network is rebuilt from hidden plus the stored CPU state dict, so the same tuple travels with the file.

Choosing a value

• 128,64 (two layers, tapering) is a sensible size for the modest feature vector; the training guide’s “bigger” recipe is 256,128,64.
• More/larger layers increase capacity, training time, and GPU memory (--batch-size is the other memory knob); dropout and the ensemble’s holdout-accuracy weighting limit — but don’t remove — overfitting risk.
• A single layer (--hidden 64) is valid. An empty value (--hidden "") produces logistic regression on the features (input straight to the 2-logit output layer) — legal but rarely what you want.

Examples

python train.py --hidden 256,128,64 --epochs 60 --lr 5e-4
python train.py --hidden 64                 # small, fast net

python train.py --hidden "256,128,64"

Interactions

• --epochs, --lr — bigger networks usually want more epochs and/or a lower LR.
• --batch-size — lower it if a large network hits GPU out-of-memory.
• --no-nn — makes this flag a no-op.

See also

• src.training.torch-model API reference
• Ensemble concept

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