Simon Meister

I second that. I often use the plugin to create a whitelist for the contents of a table/list, where by default everything is selected. This makes the ui rather bloated...

Just in case anyone would like to look at a alternative work-in-progress implementation without openai-baselines dependency and complete action space: https://github.com/simonmeister/pysc2-rl-agents.

Which of the games are you training? Yes, we also encountered it occasionally in DefeatRoaches and DefeatZerglingsAndBanelings (if I remember correctly), however it only was for some runs and I...

@Dongyang7 In the original DeepMind paper, functions which require rectangle selection used two different spatial output layers for the top left and bottom right position (if that was what you...

In our A2C re-implementation (https://github.com/simonmeister/pysc2-rl-agents), we are also able to train MoveToBeacon in about 4000 to 8000 episodes for most runs (using 32 parallel environments), with mostly complete inputs and...

@Inoryy 1) My suggestion would be to use log2(categories) channels (or log with some other base) for each embedding, as e.g. unit_type has 1850 categories and 1 output channel might...

@pekaalto Regarding the computational efficiency, i meant it in the following way: let's assume we have 2 feature layers and we want to have a specific _total_ output dimensionality (e.g....

@chris-chris @avolny They only state that there is a single output channel 1x1 convolution for _a_ spatial output. In my interpretation, there may be multiple spatial and non-spatial outputs. For...

@avolny By the way, you can have a look at https://github.com/simonmeister/pysc2-rl-agents/blob/master/rl/agents/a2c/agent.py for a example implementation. It predicts the complete action space with all argument types, works fine for MoveToBeacon and...

@JIElite in the FullyConv agent, the spatial policy should be taken from the convolutional stream, with a 1-channel convolution as output and no further fully-connected layers. This is then interpreted...