retrain_every

🦉 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

How many resolved live trades the bot accumulates before calling partial_fit() on the online-learning ensemble.

This field has no environment variable — to change it, edit the default in src/config.py or construct BotConfig yourself.

What it does

When a pending trade resolves as a win or loss, _result_checker in src/bot.py converts the outcome into a training label (win → the predicted direction was correct; loss → the opposite direction was correct), adds the stored feature vector as a sample, and increments a counter:

self.ensemble.add_sample(features, label)
self._samples_since_fit += 1
...
if self._samples_since_fit >= self.cfg.retrain_every:
    self.ensemble.partial_fit()
    self._samples_since_fit = 0

So retrain_every counts resolved trades (wins and losses only — draws produce no sample) between incremental partial_fit() calls on the ensemble’s online models (SGD, Passive-Aggressive, Naive Bayes). Buffered samples are not learned from until the fit runs.

This counter only governs the online models. The batch models (GBM/RF) are retrained on a separate schedule in the same function: once the batch buffer reaches 200 samples and every 100 samples thereafter, on a sliding window of the last 500 samples. That schedule is hard-coded and not affected by retrain_every.

Note the same field name does not control dataset pre-training (_load_dataset fits every 5000 samples) or journal reload (one fit after loading).

Valid values

Any positive integer.

• Lower values (e.g. 1–5): the model adapts faster to recent results but each fit is on fewer samples, so updates are noisier.
• Higher values (e.g. 20–50): smoother, slower adaptation; recent trade outcomes take longer to influence predictions.

Examples

# src/config.py
retrain_every: int = 5                  # partial_fit after every 5 resolved trades

from src.config import BotConfig
cfg = BotConfig(ssid="...", retrain_every=5)

Interactions

• Trade frequency (gated by min_wait_between_trades, trading_hours, confidence gates) determines how quickly retrain_every trades accumulate in wall-clock time.
• The learned state is persisted via PO_BRAIN_PATH (auto-saved every 10 resolved trades) and the trade journal at db_path, which is replayed on restart.

See also

• Ensemble concept
• Persistence guide
• Config reference index

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