Edge question added; minor improvements (sections for Hyperopt and Edge)
This commit is contained in:
parent
e4369e06bc
commit
a1b00f9053
23
docs/faq.md
23
docs/faq.md
@ -1,4 +1,6 @@
|
||||
# freqtrade FAQ
|
||||
# Freqtrade FAQ
|
||||
|
||||
### Freqtrade commons
|
||||
|
||||
#### I have waited 5 minutes, why hasn't the bot made any trades yet?!
|
||||
|
||||
@ -34,7 +36,9 @@ perform anymore BUYS?
|
||||
|
||||
You can use the `/forcesell all` command from Telegram.
|
||||
|
||||
### How many epoch do I need to get a good Hyperopt result?
|
||||
### Hyperopt module
|
||||
|
||||
#### How many epoch do I need to get a good Hyperopt result?
|
||||
Per default Hyperopts without `-e` or `--epochs` parameter will only
|
||||
run 100 epochs, means 100 evals of your triggers, guards, .... Too few
|
||||
to find a great result (unless if you are very lucky), so you probably
|
||||
@ -68,3 +72,18 @@ but it will give the idea. With only these triggers and guards there is
|
||||
already 8*10^9*10 evaluations. A roughly total of 80 billion evals.
|
||||
Did you run 100 000 evals? Congrats, you've done roughly 1 / 100 000 th
|
||||
of the search space.
|
||||
|
||||
### Edge module
|
||||
|
||||
#### Edge implements interesting approach for controlling position size, is there any theory behind it?
|
||||
|
||||
The Edge module is mostly a result of brainstorming of [@mishaker](https://github.com/mishaker) and [@creslinux](https://github.com/creslinux) freqtrade team members.
|
||||
|
||||
You can find further info on expectancy, winrate, risk management and position size in the following sources:
|
||||
* https://www.tradeciety.com/ultimate-math-guide-for-traders/
|
||||
* http://www.vantharp.com/tharp-concepts/expectancy.asp
|
||||
* https://samuraitradingacademy.com/trading-expectancy/
|
||||
* https://www.learningmarkets.com/determining-expectancy-in-your-trading/
|
||||
* http://www.lonestocktrader.com/make-money-trading-positive-expectancy/
|
||||
* https://www.babypips.com/trading/trade-expectancy-matter
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user