pmf-pymc example is broken

[fonnesbeck]

The referenced data for this example seems not to exist, and print statements are legacy Python.

[twiecki]

ping @macks22

[macks22]

I assume you're referring to the "jokes" directory? The csv file, the train/test split and the pmf-map-d5 directory are all present. As for "legacy" print statements, would you like them converted to Python 3.X function-style print statements? If so, will a v4 notebook be compatible or are only < v4 notebooks supported?

[twiecki]

@macks22 That's how I had remembered it. @fonnesbeck why weren't you able to run the files? Perhaps we aren't referencing the correct directory.

Changing to print() is 2 and 3 compatible so that one is an easy choice.

I think ipython nb v4 is fine.

[ColCarroll]

This is still broken -- looks like the files can be found here: http://eigentaste.berkeley.edu/dataset/

They are ~97kb. I will add them back in a separate PR from #1444, and update print statements.

---------------------------------------------------------------------------
FileNotFoundError                         Traceback (most recent call last)
<ipython-input-6-46c88078b60d> in <module>()
      7 # Let's see for ourselves. Load the jokes.
      8 joke_dir = '../data/jokes'
----> 9 files = [os.path.join(joke_dir, fname) for fname in os.listdir(joke_dir)]
     10 jokes = [fname for fname in files if fname.endswith('txt')]
     11 nums = [filter(lambda c: c.isdigit(), fname) for fname in jokes]

FileNotFoundError: [Errno 2] No such file or directory: '../data/jokes'

[twiecki]

Can we download them on the fly if they don't exist?

[ColCarroll]

Looking closer, the files are used in one place, and only to access the
best and worst rated jokes. I see three choices:

1. It could be removed, but I think it adds more character to the example
   (if we're rating jokes, we should get to see the jokes)
2. It could be hard coded. This would be sad for anyone who wanted to
   interact with the notebook.
3. We download on the fly. I can't figure out how to do that in a way that
   is python 2/3 compatible, other than adding the requests library. Also
   there's some light post-processing (the files are available as html, but we
   might want to display them as text).
4. We add 100 files to pymc3/examples/data/jokes, 1.txt through 100.txt.
   I've already done this, and it runs as expected.
5. We combine all 100 into jokes.csv, with line number corresponding to
   joke number. This is a <50kb file.

I like choice 5 the best.

Another problem is that pmf.draw_samples(5000) appears to take hours to
run (didn't complete last night). I'm going to lower the number of samples
to just make sure it runs, and maybe someone else can work on the
performance.

On Thu, Oct 13, 2016 at 4:41 AM Thomas Wiecki notifications@github.com
wrote:

Can we download them on the fly if they don't exist?

—
You are receiving this because you modified the open/close state.

Reply to this email directly, view it on GitHub
#830 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/ACMHEFukl7ztNoScgh4IaWoG4fEKnGaPks5qze6fgaJpZM4GAZuK
.

[twiecki]

Closed by #1447.