Error with DPGAN and other GAN based generators #614
-
|
Hi, I'm using Synth-smartnoise for trying out some of the experiments generating differentially private synthetic data. What I'm stuck with is the GAN based generators throws me an error with specific datasets. DPGAN throws the following error. "TypeError: float() argument must be a string or a number, not 'tuple'" This happens at line 246 of dpctgan.py. If I read the data as read_csv(path, header=None) the error goes away, however resulting data becomes unusable as the header is treated as one of the rows in dataset during data generation and it corrupts the entire generated dataset. I tried dropping the 1st row after reading the csv and then passing the resulting data as sample. That also resulted in the same above mentioned error. The issue I have noticed are with these specific datasets. Can you suggest, if I'm missing something or is this a defect that no one noticed ? Any guidance/suggestion would help. Regards |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 2 replies
-
|
Pinging @joshua-oss, who's more familar with smartnoise-synth. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for reporting this. I think it may be this bug where the one-hot encoder expects 2 or more categories. If you can confirm that your data has a column with only one category, there is a PR under review that will fix the issue. We can expedite and cut a release. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you Joshua. Your suggestion worked. Yes, the data set has column with only one categorical value. I dropped the column and tried. It worked. Regards |
Beta Was this translation helpful? Give feedback.
-
|
Excellent, thanks again for reporting. The fix has been deployed to smartnoise-synth v0.3.3 on PyPi |
Beta Was this translation helpful? Give feedback.
-
|
Great !! Thank you very much for the support and quick action.
Regards
Kiran
…On Thu, Dec 1, 2022 at 5:09 AM joshua-oss ***@***.***> wrote:
Excellent, thanks again for reporting. The fix has been deployed to
smartnoise-synth v0.3.3 on PyPi
—
Reply to this email directly, view it on GitHub
<#614 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/A4NG2NPWLQXQ63VD2DPSIS3WK7QT3ANCNFSM6AAAAAASMRKAEM>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
Thanks for reporting this. I think it may be this bug where the one-hot encoder expects 2 or more categories.
opendp/smartnoise-sdk#509
If you can confirm that your data has a column with only one category, there is a PR under review that will fix the issue. We can expedite and cut a release.