search

mila-iqia / blocks-extras

A collection of extensions to the Blocks framework
MIT License
27 stars 40 forks source link

Updating Plot for Bokeh 0.11 #35

Open dwf opened 8 years ago

dwf commented 8 years ago

Bokeh's server functionality has been completely rewritten and reimagined (for the better, on the whole). Currently it does not have an "index page" of the sort that we rely upon for plots now. It'd be possible to build one, either into Bokeh or a custom one for Blocks. bokeh/bokeh#3087 contains a discussion I've had with the Bokeh developers about it.

dmitriy-serdyuk commented 8 years ago

We might consider switching to Pastalog.

As far as I understand, it doesn't have so much functionality as Bokeh. For example, it doesn't seem to be able to plot images.

dwf commented 8 years ago

It's also a gigantic pain the ass to install so far...

dmitriy-serdyuk commented 8 years ago

It was very straightforward for me: conda is a magical tool, it has anything I needed.

dmitriy-serdyuk commented 8 years ago

Do you mean that it's painful to install node.js? Pretty much the same amount of effort is needed to install Bokeh.

dwf commented 8 years ago

Their PyPI package is "legacy Python" only, so I'm trying directly from GitHub. It seemed that conda did not have a recent enough nodejs (they demand >= 5, conda only has 4.4.x). And as far as I can tell, their build instructions from source are wrong, wrong, wrong.

dmitriy-serdyuk commented 8 years ago

Another disadvantage comparing to Bokeh is that it dumps everything on the same chart. I'd like to see cost/error rate/gradient norms at the same time on different plots since it can have very different scale.

But it's very straightforward to compare different models.

dwf commented 8 years ago

Have you actually used it successfully? It just says "waiting for data" to me, despite me sending "post" requests to it.

On Thu, Jun 9, 2016 at 5:33 PM, dmitriy-serdyuk notifications@github.com wrote:

Another disadvantage comparing to Bokeh is that it dumps everything on the same chart. I'd like to see cost/error rate/gradient norms at the same time on different plots since it can have very different scale.

But it's very straightforward to compare different models.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/mila-udem/blocks-extras/issues/35#issuecomment-225033672, or mute the thread https://github.com/notifications/unsubscribe/AADrLuqDiNbUw_M0I8urhMIZBlL0PGPoks5qKIbFgaJpZM4GhW6v .

dmitriy-serdyuk commented 8 years ago

I was able to install it at the lab machines and having the same issues as you on my mac.

It looks pretty cool, take a look at eos4:5008 .

dmitriy-serdyuk commented 8 years ago

One more point: it's impossible to clear the plot from python :(

dmitriy-serdyuk commented 8 years ago

Somehow, build fails when I try to compile the npm package in Python3 environment, but it works in Python2. After I start a server, it is accessible from both versions of python.

dmitriy-serdyuk commented 8 years ago

I wrote a proof of concept script which runs a flask application inside tornado used by bokeh server. Please take a look: https://gist.github.com/dmitriy-serdyuk/bd6cfcc49b5bce0bf2975d93d085053a

Flask writes data points into a tinydb database when requested. Bokeh reads the same database every 2 seconds and updates a plot.