--epochs

🦉 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

Number of full passes over the training set for the neural model (PyTorch MLP).

What it does

_train_nn() constructs TorchMLPClassifier(epochs=cfg.nn_epochs, ...). In fit() (src/training/torch_model.py), the training loop runs for epoch in range(self.epochs), each epoch iterating a shuffled DataLoader over the scaled training features with cross-entropy loss and the AdamW optimizer. Progress is logged on epoch 1 and every 5th epoch:

   MLP epoch 5/30  loss=0.6893  acc=53.2%

There is no early stopping or validation-based scheduling — the model trains for exactly this many epochs. Overfitting is mitigated structurally instead: dropout 0.2 after every hidden layer and AdamW weight decay 1e-5 (both fixed, not CLI-exposed). After training, the model’s holdout accuracy (on the --test-split set it never saw) becomes its vote weight in the ensemble, so an overfit net is down-weighted rather than trusted.

Choosing a value

• 30 is a reasonable default for the default 128,64 network on hundreds of thousands of samples.
• Watch the logged in-training accuracy: if loss is still falling at the last epoch, more epochs (e.g. --epochs 60) may help — but judge by the holdout accuracy line (MLP holdout accuracy: ...), not the training one.
• Training time scales linearly with epochs; on CPU this is the dominant cost of the whole run.
• Because there is no early stopping, very high epoch counts mostly buy overfitting; the ensemble weighting limits the damage but wastes time.

Examples

python train.py --epochs 60 --hidden 256,128,64 --lr 5e-4
python train.py --cpu --epochs 50

Interactions

• --no-nn — makes this flag a no-op.
• --lr, --batch-size, --hidden — the other MLP hyperparameters.

See also

• src.training.torch-model API reference
• Training guide

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