Description
Is it possible to set the number of scenarios to a small number, such as 1?
import dataclasses
from waymax import config as _config
from waymax import dataloader
# Create the data iterator
config = dataclasses.replace(_config.WOD_1_0_0_VALIDATION, max_num_objects=max_num_objects)
data_iter = dataloader.simulator_state_generator(config=config)
# Get a scene (this takes ~ 80 seconds for the first time, < 1 seconds after that )
scenario = next(data_iter)
Motivation
It would be helpful to this because calling next(data_iter) with the default settings for the first time takes approximately 80 seconds. I assume this is due to the iterator being jitted.
Speeding this up would facilitate faster debugging/testing.
What I tried
I went through the config.py file and code in the dataloader folder but couldn't find a way to reduce the number of loaded scenes. Would it be possible to include something like max_num_rg_points for the number of files?
WOD_1_0_0_TRAINING = DatasetConfig(
path='gs://waymo_open_dataset_motion_v_1_0_0/uncompressed/tf_example/training/training_tfexample.tfrecord@1000',
max_num_rg_points=20000,
data_format=DataFormat.TFRECORD,
)
Thank you for considering this issue and, of course, for releasing waymax!
Description
Is it possible to set the number of scenarios to a small number, such as 1?
Motivation
It would be helpful to this because calling
next(data_iter)with the default settings for the first time takes approximately 80 seconds. I assume this is due to the iterator being jitted.Speeding this up would facilitate faster debugging/testing.
What I tried
I went through the
config.pyfile and code in thedataloaderfolder but couldn't find a way to reduce the number of loaded scenes. Would it be possible to include something likemax_num_rg_pointsfor the number of files?Thank you for considering this issue and, of course, for releasing
waymax!