A free tool i just build to find algorithmic strategies

Colleague said:
So i made couple more improvements on http://algostrategies.io/

First i noticed that the whole app can become unresponsive once there is a lot of data uploaded.
Sadly this fix makes old uploaded data useless so you need to upload the data again.
Also playing around i noticed it can crash when chrome decides that there is too much data uploaded.

Another change is about the stocks/data. So i did try apple and noticed quickly that the chart sucks when the data changes overtime rapidly (so from 10$ to 400$). Basically the equity on the chart goes sideways and only last fraction goes suddenly up. My fix is to have 2 charts, one is basic, second/new is adjusted so it takes percentage changes rather then changes in absolute numbers.
Lastly i also fixed the fitness function so that the profit factor is counted based on percentage changes.

sooo i think its better.

I do my research on apple again tomorrow :)
More...

I tried http://algostrategies.io/ on a one-year, daily data set for iShares MSCI Russia ETF (ERUS). With 2 workers/parellism, it seemed to use about 1/3 to 1/2 of CPU (i4790K 4 cores/8-logicial-processors). And I think the memory usage was about 300MB, but I wasn't looking that too carefully. I wasn't monitoring how long it took to run, but it might have been about 15-20 minutes which seems long for such a small dataset for 10 running cycles.

Overall, it looks like a good start for something you could use yourself or sell/rent to subscribers.

To make the tool more useful,
  • Add entry times and exit times for each simulated trade in the statistics.
  • In addtion to using prices as input, support input files having their own values of user-supplied indicators.
  • Support a user-supplied argument for the maximum number of bars to look back for creating rules.
    For example, looking at one bar of data at a time with user-supplied indicators would let users create rules on a set of assets instead of just on a single asset.
  • Support uploading files after the training is done for forward testing.

I was unable to see the code or statistics for a stored strategy (the buttons had no effect in my chrome browser), and deleting the strategy worked without an "Are you sure?" type of prompt.
 
ph1l said:
I tried http://algostrategies.io/ on a one-year, daily data set for iShares MSCI Russia ETF (ERUS). With 2 workers/parellism, it seemed to use about 1/3 to 1/2 of CPU (i4790K 4 cores/8-logicial-processors). And I think the memory usage was about 300MB, but I wasn't looking that too carefully. I wasn't monitoring how long it took to run, but it might have been about 15-20 minutes which seems long for such a small dataset for 10 running cycles.

Overall, it looks like a good start for something you could use yourself or sell/rent to subscribers.

To make the tool more useful,
  • Add entry times and exit times for each simulated trade in the statistics.
  • In addtion to using prices as input, support input files having their own values of user-supplied indicators.
  • Support a user-supplied argument for the maximum number of bars to look back for creating rules.
    For example, looking at one bar of data at a time with user-supplied indicators would let users create rules on a set of assets instead of just on a single asset.
  • Support uploading files after the training is done for forward testing.

I was unable to see the code or statistics for a stored strategy (the buttons had no effect in my chrome browser), and deleting the strategy worked without an "Are you sure?" type of prompt.
More...
thx for the feedback,

yeah i wish i had subscribers, i dont have any (besides one friend who i asked to join my patreon lol)

>I was unable to see the code or statistics for a stored strategy
sounds like a bug but it works for me, so it will be hard to fix.

> and deleting the strategy worked without an "Are you sure?"
yes i did rush my solution so i didnt care about that much.

>Add entry times and exit times for each simulated trade in the statistics
if you look at the stats, it has the "start" datetime and that i think is enough simply because it trades always exactly one bar

> In addtion to using prices as input, support input files having their own values of user-supplied indicators.

this is a unique request, that i would not do right away. it would take some time.

other requests are also very valid...

Anyway i am in the middle of a job interview/process and they asked me to create a project
based on some description. its not easy :D definitely harder then Algo Strategies hahaha so i need to work on that for a week or so.... I wish there would be people paying for my side project but that simply wont happen that easily.

Btw i did try to run it on data for Apple (daily yahoo data since 1.1.2000) i found some good swing strategies, eventually i will post them.
 
I was thinking again about the speed again, its true its slow. Perhaps if i would make use of some randomness rather then trying every possible combination...
I also did some caching in the past but typically no 2 function calls are same, and it can also blow the memory easily.
As always i still have ideas also on this issue, will see.
 
You must log in or register to reply here.
Back
Top