search

cocreature / thrill

Thrill - An EXPERIMENTAL Algorithmic Distributed Big Data Batch Processing Framework in C++
http://project-thrill.org
Other
0 stars 0 forks source link

Unreasonably high error rate for some combinations of precision and sampling size #9

Closed cocreature closed 7 years ago

cocreature commented 7 years ago

I’ve just generated some graphs on random data (100 iterations for each size) and some of the results look very suspicious:

Precision 10 index

Precision 7 precision_7

Precision 13 precision_8

Since the error rates for smaller and bigger sample sizes look alright, it seems like it might be related to the Bias correction down for medium sized arrays. I’ll try to take a more detailed look in the next few days.

cocreature commented 7 years ago

There was a typo in the bias correction which resulted in it never being used. After fixing this the plots look fine.