Backtest visualisation

[jpmediadev]

What do you use to visualize the results as a GUI ?
https://kernc.github.io/backtesting.py/ - this thing looks very nice

[cjdsellers]

I've been using Jupyter notebooks with some custom plots. One option is we eventually add some example notebooks, or we could go the route of integrating something more substantial such as a Backtester.py.

It does look quite good. I'll investigate more closely over the next few days and likely open an issue for further discussions.

[jpmediadev]

Some examples with your notebooks will be great!

I have idea to use ClickHouse + Grafana for backtest visualisation and also like realtime dashboard ,
how best to implement it - use Logger subclass with separate thread or write a special class at the strategy level?

[Troubladore]

I'm thinking about a UI as well - any progression in thoughts on this space? Was actually contemplating a standard JS/Angular web front end, but I'd prefer to contribute something everyone could use

[limx0]

I'm also interesting in this topic.

[cjdsellers]

I think we can work towards some more basic backtest visualizations in the same vein as some other open source Python trading platforms currently out there. Note that empyrical is currently being used for backtest performance metrics (also used by pyfolio).

A JS/Angular front end is a more substantial undertaking, which I had intended to include in a future licensed add-on to NautilusTrader.

The first step would be building a web server which could expose a REST and WebSocket API, leveraging something like FastAPI. This is something an Angular front end could then 'plug into'. Its currently not my intention to add the machinery for this into the open source repo, however I'm happy to discuss topics relating to the above via email.

[limx0]

Another good looking library https://github.com/ranaroussi/quantstats

[cjdsellers]

This looks very good actually.

[thematz1]

I checked this out (quantstats) and landed on Cerebro ( https://github.com/mementum/backtrader ) from there. I think they accomplish many nice things with a similar idea as Nautilus Trader, but in pure python.

Could be interesting to take a look! They have an argument to plot the results with Pyfolio or Quantstats after running the test. I also find it interesting how they allow you to easily input any type of bar data (1m aggregations, 1 hour aggregations, etc) which could help to limit the ram required to run a backtest (at the trade-off of accuracy).

I thought I would share incase anyone wanted to take a peak and get some ideas for feature implementation here.

[cjdsellers]

Thanks for sharing @thematz1. I've been aware of Backtrader for a while and have played with it in the past. It doesn't look like the project is too active these days - however I'm aware of many forks which have improved on it, both public and private.

So we currently use the empyrical library for backtest performance stats, which is what pyfolio calls into under the hood. Unfortunately that library is also not very active, so its prime time we seek an alternative.

Feel free to continue investigating this theme and post any ideas you have! Possibly an integration with a more active and built out project such as quantstats could be the way to go?

[yohplala]

Feel free to continue investigating this theme and post any ideas you have! Possibly an integration with a more active and built out project such as quantstats could be the way to go?

Hi Christopher,
Following your invitation, I can highlight that quantstats is already used by what I identify as Nautylus trader's 'main competitor', Jesse-AI.
Their forked repo: https://github.com/jesse-ai/crypto-quantstats

[thematz1]

Im wondering how i could just output the returns to CSV in order to simply use the series in one of these visualizers. I was planning on investigating what the input should be from an actual trading strategy into quantstats. I figured posting it here could save some time if someone who has tried reads it. Any ideas?

[OnlyC]

While we don't have the perfect solution yet, can you suggest how should I start to build a FastAPI server to expose backtest
data, result, indicator's data ...?
@cjdsellers
I've done something like this to qtpylib, but then I found nautilus-trader. Now I need to do it again :)

[DeevsDeevs]

Hey, @cjdsellers, awesome project, thank you!
Any updates about this visualisation part? i found nothing in docs about it
Are there any dirty solution (if we don't have stable solution for that)