turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Data Mining
- /
- Inverse iteration for eigenvectors fails

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

2 weeks ago

Hello,

I am using the Principal Component node in SAS Miner for variable reduction. For a subset of the data, I am getting "Inverse iteration for eigenvectors fails" error. The data has about 800 columns and 3314 rows.

The Principal Component node runs fine when I run it with 16053 rows and 800 columns. In this 16053 rows, the previous 3314 rows is also included.

Is there a way to get past this error?

Thanks

Arindam

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

2 weeks ago

Do you have lots of missing values in the data? If so, the algorithm might be dropping the missing cases and you are ending up with fewer (valid) rows than columns. Try performing listwise deletion on the 3314 rows so that you keep only the complete cases. Are there more than 800 complete cases?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

2 weeks ago

But PCA should work just fine with fewer rows than columns (unless there's something about the way it is implemented in EM)

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

2 weeks ago

@PaigeMiller is correct. Sorry for the confusion.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

2 weeks ago

It must be due to some nature of the data in play at the time of matrix inversion. If I take out a further section of my original data then PC node works. So it is definately not failing due to the amount of records, more like the relationship between the records at the time of PC derivation. So even if a PC calculation of 1000 records can fail, but a subset of the same records may not fail. Similarly if the 1000 records are included in a dataset of 2000 records, then also it may not fail.