Will Jessup
Will Jessup
simple jsfiddle w/ the `type="module"` error. https://jsfiddle.net/wdzsbgvL/
Adding on here, can this feature support initializing from a list that contains numpy arrays? `a = mx.array( [np.array(1), np.array(1)])`
Overall i'm trying to get this faster than the equivalent in pytorch, which will do 100 episodes, Total steps = 6358, with Avg time per step: 0.00089. Currently my time...
Thanks for the comments. is this the right way to implement the warmup? Regarding #3, would I put an mx.eval() with no passed argument above the first timer? ``` #...
Sure the code I was replace was from a DQN with replay memory. We need to select only the non terminated cases to make predictions, so: ` terminated_batch = torch.tensor([False,...
For simplicity I didn't show the part where we're selecting random samples from the replay memory and instead just showing example tensors. The idea here is that in the replay...
no need to differentiate. the numpy method looks good. will try it out!
this works if you change default.secret.json to secret.js and export the object: `module.exports = { "main": { "id": 0, "token": "qwerty123" }, "hackathon": { "id": 1, "token": "qwerty234" }, "workshop":...
Friendly bump. Is there a better way to share the benchmark w/ you?
Thanks for looking! It does seem that all terminated states cannot be replayed with a different action. I wonder why not. Anyway, can you rewind 2 actions and it won't...