Shiny (Python) ML training example

[neelasha23]

Add Shiny example

---

📚 Documentation preview 📚: https://ploomber-doc--105.org.readthedocs.build/en/105/

[edublancas]

please deploy the app so I can test it

[neelasha23]

Addressed the comments. App is deployed here: https://crimson-glade-4646.ploomberapp.io/
After setting the MONGODB_CONNECTION_URI env variable in local u can run the two scripts as mentioned in Readme: python train.py --model model_1 --units 128 and python train.py --model model_2 --units 12
@edublancas

[lfunderburk]

Hi @neelasha23

Took a look at the app - but the plot is not showing up

[neelasha23]

Hi @neelasha23

Took a look at the app - but the plot is not showing up <img alt="Screenshot 2024-02-02 at 4 00 51 PM" width="2000"

I have cleared the db because it's a realtime monitoring. you need to connect to ur mongodb account and run training scripts from local and view the plot in the dashboard. have added instructions in the README @lfunderburk

[edublancas]

the polling logic is off: I tried the app many times (both the ploomberapp.io one and locally) and I only got it one time to work. I debugged it by printing the value of last_modified and realized that even if I started training, the returned value was the same (I tried refreshing the browser after I started training).

I changed the last_modified function to this:

def last_modified():
    n = my_collection.count_documents({})
    return n

and it worked better: as soon as I started training, the plot started updating. but I found another issue: the values in the plot did not match the values printed from the terminal and when starting to train another model, the plot didn't show a new line:

2024-02-03.8.57.55.a.m.mp4

[neelasha23]

I haven't been able to reproduce the timestamp not getting updated issue. Did you try with my account? I had selected an Indian server, so I'm doubting if that's causing a lag.
Can you please try once with a new account? If it still doesn't work I'll figure out what's happening
@edublancas

Screen.Recording.2024-02-05.at.8.27.21.PM.mov

[edublancas]

@neelasha23 the problem is most likely the datetime.now(), such function returns the timestamp in your local timezone.

but there's no need to reproduce it since I already provided a fix:

def number_of_observations():
    n = my_collection.count_documents({})
    return n

this function will trigger an update as soon as there is a new element in the collection

[neelasha23]

Fixed the polling function: https://github.com/ploomber/doc/pull/105/files#diff-c99010220192b232032cc417d9e26e799aa65a2dd2fbb5dd79b5595548293670R33-R38

Although I did not completely understand why polling the timestamp column would be an issue. Even if datetime generates in local timezone it will still get accurately stored in MongoDB. And the polling function keeps checking this stored timestamp to detect if any new row has been added. You also mentioned that even after changing the polling function the timestamps in the plot are not matching the training ones , but I don't see that happening in my local. Which is why I feel maybe there's a lag between the training and the storage/retrieval process in your case?
@edublancas

Screen.Recording.2024-02-06.at.10.50.53.AM.mov