[rllib] More efficient tuple flattening. #4416
Conversation
|
Can one of the admins verify this patch? |
| @@ -41,10 +42,14 @@ def transform(self, observation): | |||
| """Returns the preprocessed observation.""" | |||
| raise NotImplementedError | |||
|
|
|||
| def write(self, observation, array, offset): | |||
| """Alternative to transform for more efficient flattening.""" | |||
| raise NotImplementedError | |||
There was a problem hiding this comment.
Should this have a default implementation that calls transform() and then puts it in the array? Otherwise I can see this exception coming up if you hit a preprocessor that doesn't support it (though it seems most preprocessors have write() implemented).
There was a problem hiding this comment.
Good suggestion, I'll add that.
ericl
changed the title
|
@justinwyang any idea why Travis ran the Java tests and also the regular Python and Tune tests? This should PR should only have run
|
|
Test FAILed. |
|
Test PASSed. |
|
Test PASSed. |
|
Test FAILed. |
* More efficient tuple flattening. * Preprocessor.write uses transform by default. * lint * to array * Update test_catalog.py * Update test_catalog.py
What do these changes do?
I found tuple flattening to take up quite a lot of time (30% of total sampling), mainly due to creating lots of numpy arrays. This change creates only one array and then writes into it.
Related issue number