Evan Walters
Evan Walters
Perfect! Thank you!
I second this, if Bryant is down. I might work on it.
Roger that. I might go for Deribit
No I have not! But now I want to after you've reminded me -- orthogonal used to work very well for me in RL experiments with complex networks. Maybe I'll...
Little update, definitely getting some interesting results with the orthogonal init. At first glance, it seems less finicky than kaiming init, more stable. I'm going to turn the experiment above...
For what it's worth, I was able to get it to work with data_init=False, because the problem in your error code, `self._naked_clone_layer.set_weights(self.layer.get_weights())`, is only called if data_init=True.
I actually thought this might be a problem before I even tried tfa weightnorm on TPU... The real question for this issue is, how to do weightnorm paper's data-dependent weight...
I have also thought about this. I am wondering if showing the agent several layers of the order book will benefit it or not, it needs to be experimented with....
I’ll be testing this later today using tensorflow 2.0 in a colab notebook, I’ll report back if I have to make any changes to make it run smoothly. Edit: never...
Quick question regarding host_callback.call within jitted functions and multiprocessing... Say I have a machine with a GPU and 4 CPU cores. If I jit and vmap a fn that calls...