Douglas Blank

Actually actually, Calysto Scheme spells that command `string->symbol`

Command-completion works very well to see what is defined. And `(dir)` lists all variables and macros.

Thanks for any hints you can give! This is the entire log on startup: ``` [D 2016-08-19 19:21:52.836 JupyterHub application:529] Looking for jupyterhub_config in None [D 2016-08-19 19:21:52.837 JupyterHub application:549]...

What should NotebookApp.server_extensions be set to?

I ran out of time to track this down. I'll leave this for now. Perhaps someone can debug on a jupyterhub server and report back. Thanks!

Conx was written for tf 1. So, you could down-grade tf. But we're also going to redo conx in the next few weeks to work with a more modern tf.

I'm not sure this is true. What depends on image_data_format? Can't we just set it to what we want?

I think some pre-existing models have this built in to the structure. I don't think this is an issue unless you are using models built elsewhere.

Or even better: do it in Keras.

Or in keras, and could use the actual error functions. That would allow crossentropy and other loss functions to be handled correctly.