Bert model

[susiejojo]

Changes made:

Should solve #2 and #3 .

• Setup a basic Flask backend
• Setup endpoint /predict for fetching predictions
• Setup Fetch API request and response handling to the model
• Create a predict_type function in predict_types.py to return the predicted user personality type.
• Return a json response from the endpoint to display the response in following format: {"type" : <personality_name> }

Tests to be checked

• Check by entering some text in the browser form by running index.html on your browser and flask run on your terminal together.
• Check browser console for the expected json response.
• Check your terminal to see if the whole BERT model gets downloaded or no. It will probably download the very first time you run it, (warning, it's 500 MB).
• Make sure BERT model doesn't download for all but the first time that you test.
• For the next few times, keep track of how long it takes to print the response on the browser console after sending the request. For me, it's around 1 min (which we can account for by keeping a loader, or nothing even)

What to do next:

• Test same on Twitter API responses.
• Return probabilities of each class instead of one label.
• Send this response json to frontend for displaying on the page.

[V2dha]

Similar to @sh-biswas, the model returns a JSON response within 15 seconds and the model gets downloaded only the very first time so everything's working good. Just one thing, there was an error in the terminal about the incompatibility in shape of the tensor, the shape of the tensor required was (None, 1500) but shape it got was (None, 8). But despite that it is giving the json results so I guess its okay.

[susiejojo]

Similar to @sh-biswas, the model returns a JSON response within 15 seconds and the model gets downloaded only the very first time so everything's working good. Just one thing, there was an error in the terminal about the incompatibility in shape of the tensor, the shape of the tensor required was (None, 1500) but shape it got was (None, 8). But despite that it is giving the json results so I guess its okay.

Yeah so that's the bug why I can't get model.json. I'll look into it, but since it's working atm ig we should be fine.