🐞 Customize your bug

Please check the colab documentation as for more intuitive play with this feature

🐞 Customize your bug

Bugs are, terrifying things, to be exposed to your students, especially the ones not familiar with python.

But to fix bugs, first we have to conquer the fear. The fear over its complexity and verbosity

A friendlier experience

Here we try to offer a smooth experience over frequently met bugs, as this can be configured freely, even outside of this library

Import and deploy the bugbook

from unpackai.bug import ishell, BUGBOOK, itb

If you ran this, you'll have different experience encountering bugs:

• There will be download report button, so you can send your Mentor/TA a detailed report
• The error prompt will be more friendly structured, it will show message from unpackai
• Still shows the original traceback, if needed

Simple unified message

# configure the bug mapping using error type name as the key (case sensitive)
BUGBOOK["KeyError"] = f"KeyError usually means there is no such key in a dictionary, but also can mean there is no such column the dataframe table"

Now try run the following to see an error

import pandas as pd
pd.DataFrame({"week":[1,2,3]})["weeks"]

Check the UnpackAI Tips in return , congrats on your first reconfigured error report!

Simple conditional message output

Certainly we can be much more smarter than that, we can try to customize the message text according to the error detail.

Instead of setting a string to BUGBOOK, we can set a function, that will takeetype, evalue, tb as inputs, but return a string message

def customized_keyerror(etype, evalue, tb):
    # create structured traceback
    stb = itb.structured_traceback(etype, evalue, tb)
    # string format of trace back
    sstb = itb.stb2text(stb)

    # if this is a pandas keyerror?
    if 'pandas' in sstb:
        key = str(evalue)
        return f"""
        There's no <strong style="font-weight: bold; color: var(--ansi-red);">column {key}</strong> in the dataframe
        <br>
        <p>Try to...</p>
        """
    else:
        return f"There's no key {key} in the dictionary"

BUGBOOK["KeyError"] = customized_keyerror

Hence we got our flexibility, check the UnpackAI Tips again

pd.DataFrame({"week":[1,2,3]})["weeks"]

About the function you passed in

Inputs

Notice the input arguments of the function, it will always be

• etype: error type, etype.__name__=='KeyError'
• evalue: default error message, in the above case str(evalue)=='weeks'
• tb: traceback, in the above case

itb.structured_traceback(etype, evalue, tb).stb2text(stb)

will be the entire traceback text

Return

The return of the function will always be a string, it's actually a HTML string to render. So images, js, widgets, you name it...

With this flexibility, your own imagination will be the limit

For contributors

You can put bugbook under different module 🔧

If you're developing unpackai library, eg. if you are developing nlp module

If within the NLP module, certain error can only leads to certain solution under this subject

You can put the following under unpackai/nlp/__init__.py:

from unpackai.bug import BUGBOOK, ishell

def customized_file_notfound(et, ev, tb):
    if 'gdrive' in str(ev):
        return f"""
        Please check if the google drive is mounted...
        <button>reload google drive</button>"""
    elif '.txt' in str(ev):
        return f"""
        Please check the text data is ...
        <button>upload text again</button>"""
    else:
        return str(ev)

BUGBOOK['FileNotFoundError'] = customized_file_notfound

In such way, the error handling for CV, tabular and NLP can be different and specific

You can further improve the interface 🌈

• You can change the bug page by editing this html template
• YOu can change the error report page by editing this html template

Fancier features

Well, if the above things didn't buffle you, let's not stop here

Now you know this format to set new configuration

BUGBOOK['FileNotFoundError'] = customized_file_notfound

But for the key of configuration FileNotFoundError, for now is a string to filter the error, means we do not apply this rule to other error, only the FileNotFoundError

This key is acting as a filter. The key can also be a function instead of a string

def pandas_filter(et, ev, tb) -> bool:
    # create structured traceback
    stb = itb.structured_traceback(et, ev, tb)
    # string format of trace back
    sstb = itb.stb2text(stb)

    # return True if the traceback has keyword pandas
    # no matter what error type it is
    if 'pandas' in sstb:
        return True
    return False

Now, use the filter function pandas_filter as the key

BUGBOOK[pandas_filter] = "This is a pandas error, please refer to the <a href='...'>documentation here</a>"

pd.DataFrame({"week":[1,2,3]}).not_a_method()