How to create environment that uses multiple different episodes with 200 time steps each
R470R opened this issue · comments
Hello, I want to make a RL for forex, in wich the environment is feeded with multiple episodes of 200 time steps each.
How can I do it?
Ok same symbol or different, if same I fiddled with env_dummy, no fun. But
if different symbols am happy with gym-mtsim on rrlib:)
Hope that helps.
As for the how, the way that quickly worked for me was . . .
Using gym-anytrading to try a sine wave that looks like your preferred
primary symbol. (Some math about Fourier,Bessel and Euler )
Then use lib or ai algorithm you made and RL on that base.
Works well on small time frames, bigger time frames where too bullish and
had too many drawdowns (for min lot and min deposit only)
On Thu, 8 Dec 2022 at 20:55, R470|2 ***@***.***> wrote:
Hello, I want to make a RL for forex, in wich the environment is feeded
with multiple episodes of 200 time steps each.
How can I do it?
—
Reply to this email directly, view it on GitHub
<#81>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA7JXS752PCMHYKPSPTBIPLWMI4JHANCNFSM6AAAAAASYQ5NOA>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
--
Kudakwashe Mafutah
[image: https://]about.me/kmafutah
<https://about.me/kmafutah?promo=email_sig&utm_source=product&utm_medium=email_sig&utm_campaign=edit_panel&utm_content=plaintext>
Ahh gym-anytrading is your favorite puppy. Plus I recall pickling the data(
saving train data save as binary made it perform three cents better at
session close for eurusd)
Good luck share if you make that dollar extra.
On Thu, 8 Dec 2022 at 22:30, R470|2 ***@***.***> wrote:
@kmafutah <https://github.com/kmafutah> same symbol, is like each episode
is 1 signal
—
Reply to this email directly, view it on GitHub
<#81 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA7JXS3E55RBUDQ6ZO5ZOULWMJHOPANCNFSM6AAAAAASYQ5NOA>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
--
Kudakwashe Mafutah
[image: https://]about.me/kmafutah
<https://about.me/kmafutah?promo=email_sig&utm_source=product&utm_medium=email_sig&utm_campaign=edit_panel&utm_content=plaintext>
Am saying for me reading binary was faster than reading csv. During
retraining and training RL for day trading small time frames.
On Thu, 8 Dec 2022 at 23:07, R470|2 ***@***.***> wrote:
What do you mean @kmafutah <https://github.com/kmafutah> , feed the
environment with binary data?
—
Reply to this email directly, view it on GitHub
<#81 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA7JXS5C4TMRBQXTIHL7UFLWMJLZTANCNFSM6AAAAAASYQ5NOA>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
--
Kudakwashe Mafutah
[image: https://]about.me/kmafutah
<https://about.me/kmafutah?promo=email_sig&utm_source=product&utm_medium=email_sig&utm_campaign=edit_panel&utm_content=plaintext>
Ok @kmafutah , but i don't understand how should I tell the environment that each episode has it's own points and the learning should be transfered (I will use DQN), but I think there's must be a simple solution
You are telling to pickle each episode?
Ok already understood, tomorrow I tell how it had gone
It works well, but I'm going deeper, I will create a custom env myself