October 23, 2018 at 6:53 pm #28967
Hello! I’d like to get started on the AIDO, but I have a few problems stopping me.
1. I’d really much like to do it in Pytorch, but the template is missing from the quick start page;
2. Since I can’t do it in Pytorch for now, I tried using the Tenserflow template, but the information on it is fairly inaccurate (folders are mentioned that do not exist in repo, and the Learning part of the guide is missing);
3. Finally I tried running the agent to record logs, but I get the error mentioned in this post
Also, and pardon me if this is written somewhere, is there a way to run the solution.py on our local machines? I’d like to know if my image processing is working or not, how can I print out results to find out?
I’m a little stumped, any leads?
October 23, 2018 at 10:49 pm #28973
- This topic was modified 8 months ago by lmandrile.
Hey, thanks for reaching out and thanks for sticking to PyTorch. Excellent choice! 🙂
For the missing documentation: everything that you need to get started is actually on the Github pages – don’t know if you saw them.
This repo tells you how to train a semi-working model for driving (documentation contained in README): https://github.com/duckietown/challenge-aido1_LF1-baseline-RL-sim-pytorch
And once you have trained model following that repo, you can use this other repo to submit your solution (you have to copy over your model, documentation contained in README): https://github.com/duckietown/challenge-aido1_LF1-template-pytorchOctober 23, 2018 at 10:53 pm #28974
Oh yes and for the local evaluation: that’s a little tricky. So the way it works is that there is a “cloud” of participating computers that all run the evaluation server. Whenever a user submits a solution, a free evaluation server will grab your submission and evaluate it. The first computer that takes your submission, will mark it as being handled – so that no other evaluator touches it or overwrites the results.
You can run the evaluation server on your local machine. And you can even set it so that it only evaluates your submission and not the ones from other people. But the problem is that this only works when nobody else is evaluating your submission. And that’s only possible when our evaluation servers are at max capacity and there is a waiting line.
So until there is more traffic, your solution will be immediately evaluated online AFAIK.
You must be logged in to reply to this topic.