02_end_to_end_machine_learning_project.ipynb -- TypeError: object supporting the buffer API required about handson-ml HOT 4 CLOSED

Hi Thiago,

Thanks for your kind words and for your feedback. I just checked, I get the same error as you. It seems that identifier used to be a string and somehow it is now an int. Weird, probably a change I made, sorry about that. Anyway, the fix is simple (I just pushed it): replace hash(identifier) by hash(str(identifier).encode("ascii")). Basically MD5 only works on a byte array (on another object that implements the buffer API). So I first convert the int to a string, then encode the string to a byte array. No pretty, but it works.

from handson-ml.

Apparently what happened is that in a previous version of Pandas, integers were returned as numpy.int64, which happen to implement the buffer API. Now they are returned as plain ints, hence the bug. This led me to a cleaner solution: simply convert the identifier to a numpy.int64, and the bug is fixed. It beats converting it to a unicode string then encoding it to ascii.

Side note: I wrote this chapter using scikit-learn 0.17.1, and you will get a bunch of deprecation warnings right now if you run it using the latest scikit-learn version 0.18. I'm going through the code now to get rid of these warnings. It's hard to write a book on a moving target!

from handson-ml.

Thanks very much for the Help! =)

from handson-ml.

My pleasure. FYI, I upgraded all notebooks to sklearn 0.18, so the deprecated warnings are gone. Don't hesitate to mention all problems you run into, I can still fix the book in the next couple of weeks. Thanks!

from handson-ml.