Mismatch in hyperopt and backtesting

[luca-palese]

Mismatch in hyperopt and backtesting

[xmatthias]

what you're missing is the actual commands you used for both backtesting and hyperopt.

[luca-palese]

Right!
docker compose run --rm freqtrade hyperopt --hyperopt-loss OnlyProfitHyperOptLoss --strategy FAdxSmaStrategy --config user_data/config.json -e 400 --timerange 20210101-20230101

docker compose run --rm freqtrade backtesting --config user_data/config.json --strategy FAdxSmaStrategy --timerange 20210101-20240301

[xmatthias]

Well your "problem" is the careless / wrong usage of parameter.value.

You can't use .value in populate_indicators() (unless you specify --analyze-per-epoch - which will make backtesting significantly slower). This method is only called once by default - so no changes to it's values will be actually used.

if memory constraints allow - it's instead recommended to create all parameters upfront (e.g. 'ATR_{val}') to then be used as part of the other populate_*_trend() methods - as described in the optimizing indicator parameters documentation section.

If memory doesn't allow (e.g. if your space is too big - you can keep your strategy as is - but you MUST use --analyze-per-epoch - which will call populate_indicators() for every epoch (this is described on the same section in the "performance tip").

[luca-palese]

I will read those, thanks man XD.

[xmatthias]

Essentially, you either use <parameter>.range: ... in your populate_indicators() - or you use --analyze-per-epoch - so you can keep using .value in populate_indicators().