FileNotFoundError after I used tune.run()

Ray Tune
1

I am trying to use tune.run to get multiple trials for my experiment. But after I used the tune.run, I got a FileNotFoundError like below:

Traceback (most recent call last):
File"/Users/zhaopengqian/miniconda3/envs/flow/lib/python3.7/sitepackages/ray/tune/function_runner.py", line 248, in run
self._entrypoint()
File"/Users/zhaopengqian/miniconda3/envs/flow/lib/python3.7/sitepackages/ray/tune/function_runner.py", line 316, in entrypoint
self._status_reporter.get_checkpoint())
File"/Users/zhaopengqian/miniconda3/envs/flow/lib/python3.7/sitepackages/ray/tune/function_runner.py", line 580, in _trainable_func
output = fn()
File “/Users/zhaopengqian/Desktop/Experiment/deeprl_signal_control的/main.py”, line 218, in train_config
dirs = init_dir(base_dir)
File “/Users/zhaopengqian/Desktop/Experiment/deeprl_signal_contro的/utils.py”, line 34, in init_dir
os.mkdir(base_dir)
FileNotFoundError: [Errno 2] No such file or directory: ‘results/ma2c_1’

And the utils.py like below which is just to create a directory and file:

if not os.path.exists(base_dir):
os.mkdir(base_dir)

When I put this part before tune.run the propram operate correct. So, is the tune.run() change the os.path so that it can`t find the correct path?
How can i fix it?

2

Hey @Qian_Zhao, yes that’s right, the Ray Tune trials are not run in the same working directory as the driver.

You have a couple options here:

  1. Use absolute paths instead of relative paths inside your training function/trainable
  2. Starting from Ray 1.12, inside your training function/Trainable, you can access the original working dir via the TUNE_ORIG_WORKING_DIR environment variable. So you can change to this directory inside the training function, but it is recommended to do so for read-only purposes.
3

Thank you very much! I solved this problem by using the absolute paths.