%head fails with "head failed. must be real number, not str"

[roblem]

Steps to reproduce:

sysuse auto

Run

%head

fails with

head failed.
must be real number, not str

However, omitting the variable rep78 (which has missing data)

%head make price mpg headroom trunk weight length turn displacement gear_ratio foreign

gives the desired result:

Note: I am getting this error on a campus jupyterlab instance and believe the version of nbstata is fairly new as this was implemented last week, but can't verify at the moment.

[hugetim]

Thanks for reporting this. I am indeed able to reproduce it with the latest version of nbstata v0.6.4 (and the Aug 30 update of Stata 18). I'll take a look now to see if there is a relatively quick fix.

[hugetim]

I don't understand the problem yet, but I have a workaround that seems to work: changing the missing config value either in the configuration file or in a cell like:

%set missing = pandas

[roblem]

Just verifying: this works with current v0.6.4, right?

[roblem]

And is there a downside to using the pandas option?

[hugetim]

Yes, the workaround works with v0.6.4.

As far as a downside, I'm not aware of one. It doesn't make a visible difference in this example, and it may not ever do so. (This configuration option dates back to before I overhauled these magics to display numbers according to their Stata format, which may have made it redundant. It's that conversion to Stata format where the problem seems to lie--at this point, my best guess is that it worked with Pandas 1.4.2 but something changed by Pandas 2.1.0 that has made it not work.)

[hugetim]

Sure enough, another workaround is to add the noformat option to the %head magic. (In your example, this only removes the thousands comma from the display of "price" and "weight", but it may be a bigger deal in other cases, such as when using Stata's native date format.)

[hugetim]

Ok, I've traced it down to a change in pandas.api.types.is_string_dtype. For Pandas 1.x, pd.api.types.is_string_dtype(pd.Series([1, "."]))(pd.Series([1, "."])) returns True, but for Pandas 2.x, it returns False. I'm going to instead use not pandas.api.types.is_numeric_dtype, which is more consistent across Pandas major versions. Hoping to ship a new version shortly.