Open noahcgreen opened 2 years ago
Hi there, thanks. Especially, thanks for confirming that all of your package versions match what's in the README.
This issue didn't used to exist, and the code in this repo hasn't changed, so it seems likely due to some dependency shifting underneath us in a way that isn't captured by just the package versions.
Hi @noahcgreen, so I just ran a test on my side, and my tests are passing. My computer though had all of these package versions installed from before.
Could you share your output of pip list
? I wonder if you are asking for certain pip versions but they are not all getting installed as asked for?
Hi, it seems that the unit tests do not work out of the box. I'm working in a clean conda environment with Python 3.7.13. All of the prerequisites are installed with the versions described in the readme, as well as CUDA and cuDNN (Tensorflow has GPU access).
Here's the complete output from the test script:
Test script output
Here's the last part of that formatted a bit more nicely:
The issue seems to be that several fields of the observation returned by
BlockPushMultimodal._compute_state()
need to be converted to np arrays with dtypenp.float32
. After doing that and running the test again, I get the following error instead:This is the same error as in #14 so I'd guess these issues are related.
Any thoughts?