Reduce decimal precision of gdp_education.csv?

JMSLab / Template

Template for research repository using scons.

9 stars 1 forks source link

Reduce decimal precision of gdp_education.csv? #30

Closed jmshapir closed 3 years ago

jmshapir commented 3 years ago

Per 3c6e0d2 when I compile the Template I get many small changes in gdp_education.csv at high precision.

Should we round the last few digits so that different machines will produce more similar output?

mcaceresb commented 3 years ago

@jmshapir I've made the change in f2794048eb81f13c34ae9a2b26e553c70e88538b. I did 9 digits precision; LMK if you want more or fewer.

One thing to flag is that there is another csv import/export in Stata here but I do not know how to specify rounding in Stata's export. If you have not had issues with that file we can leave it as is and trust Stata will export to the same precision, unlike Python. (The alternative being writing a work-around to explicitly specify the precision in Stata.)

jmshapir commented 3 years ago

Thanks @mcaceresb!

I tested the #30 branch on my machine and reproduced the two CSVs exactly, so I think we're good here.

On the specifics:

On build.py, I'm good with 9 digits.
On takelogs.do, I think we can leave as is. (As you say, I think Stata's default rounding precision should mean this works the same or nearly the same on most machines.)

From my side you can open a pull request when ready and assign @veli-m-andirin as reviewer.

mcaceresb commented 3 years ago

Continued in #31

mcaceresb commented 3 years ago

In this issue we set the numeric precision when exporting to .csv from Python so the file doesn't change when re-compiling in different machines.