Roelof van Dijk

The script worked, but I wanted to run this within a docker based on 18.04 and avoid docker-in-docker. Regarding `./build-static.sh` script: you could shed some docker build time (170s ->100s)...

@ooeygui, Thanks for taking up this conversation. I would like to strongly second this feature request! If the Azure Kinect is to be used as a fairly complete 'robot head',...

Hello @star0w, here is a brief recording example: try the python library [sounddevice](https://python-sounddevice.readthedocs.io/en/0.3.15/), use `sounddevice.query_devices()` to get the kinect index, `sounddevice.query_devices(kinect_index)` for information about the kinect (e.g. sample rate). and...

Hello @youssef266, as far as I remember, sounddevice yields a numpy array containing the 7 separate channels, which you could feed into a spatialization algorithm. For live spatialization, you would...

Using generators instead of list comprehensions might be slower in some instances.

I have no idea what the priority is - @rayanht is going the other way (generators -> list comprehensions) [here](https://github.com/geohot/tinygrad/pull/812/files#diff-7703e7b63c9844f019ae3609629fd9c9bab290d11ae6608ed2d09edd49bab2dbR250). The performance gain might be important in very hot functions.

I think you really want to use `is` for the nodes, ops and type checking. ``` import timeit from tinygrad.shape.symbolic import NumNode X = [NumNode(x) for x in range(100)] def...

``` import timeit def func(x): return x**3 def f1(lst): return [func(x) for x in lst] def f2(lst): return list((func(x) for x in lst)) def f3(lst): return list(map(func, lst)) for n...

Quite unfavourably (function call overhead). ``` import timeit def func(x): return x**3 def f1(lst): return [func(x) for x in lst] def f2(lst): return list((func(x) for x in lst)) def f3(lst):...

You could consider using `is` for all `dtype ==, `type ==`, `op ==` and `optype ==` checks. Easy separate MR, but I don't want to touch these files while you...