Thanks, the problem was I solved a covariance matrix manually, but my fellow mate using the covariance syntax. That's why there was difference in Matlab and python results. But now when I use the covariance syntax in python, matlab and python gives the same results.
On Sat, Oct 10, 2020, 21:37 Peter J. Holzer <hjp-pyt...@hjp.at> wrote: > On 2020-10-07 07:53:55 +0200, Marco Sulla wrote: > > If you want to avoid float problems, you can use Decimal: > > Decimal doesn't avoid floating point problems, because it is a floating > point format. For example: > > Python 3.8.5 (default, Jul 28 2020, 12:59:40) > [GCC 9.3.0] on linux > Type "help", "copyright", "credits" or "license" for more information. > >>> from decimal import * > >>> a = Decimal(3) > >>> a > Decimal('3') > >>> b = Decimal(1E50) > >>> b > Decimal('100000000000000007629769841091887003294964970946560') > >>> c = Decimal(2) > >>> a + b - c - b > Decimal('8112996705035029053440') > >>> b - b + a - c > Decimal('1') > >>> a + (b - b) - c > Decimal('1') > >>> a + b - b - c > Decimal('8112996705035029053438') > >>> > > Mathematically, all four expressions should have the result 1, but with > floating point numbers they don't because intermediate results are > rounded. > > For comparison, here are the results with float: > > >>> a + b - c - b > 0.0 > >>> b - b + a - c > 1.0 > >>> a + (b - b) - c > 1.0 > >>> a + b - b - c > -2.0 > >>> > > One could argue that these are at least closer to the truth, although I > think that's just luck, > > Decimal does have two advantages over float: > > a) It's precision is configurable and even by default higher. So on > average, the error is smaller (but still not zero). > > b) It uses decimal numbers like we learned in school. So it will make > the same errors as we make when we use pencil and paper, which is > less confusing to laypersons than the seemingly arbitrary errors from > converting from decimal to binary and back. > > > The disadvantages are of course higher memory consumption and lower > speed. Also, I'm very confident that the engineers at Intel and AMD knew > what they were doing when they designed the FP units of their > processors. I'm slightly less confident about the authors of the Decimal > module. And I'm much less confident that the average Python programmer > can implement a matrix multiplication with Decimal which is as > numerically stable as what Matlab or Pandas provide using IEEE-754 > arithmetic. > > hp > > PS: I recently read an interesting article on the Android calculator. > That goes to extreme lengths to avoid unexpected rounding errors. > It is also very slow, but still faster than a human can look, so it > doesn't matter. > > -- > _ | Peter J. Holzer | Story must make more sense than reality. > |_|_) | | > | | | h...@hjp.at | -- Charles Stross, "Creative writing > __/ | http://www.hjp.at/ | challenge!" > -- > https://mail.python.org/mailman/listinfo/python-list > -- https://mail.python.org/mailman/listinfo/python-list