Precision problem while solving Cody Problems

While solving Cody problems, I have often encountered that my answer is not accepted because of the precision doesn't match the required answer. Matlab always rounds the double format to 4 digits after decimal. My logic seems correct on paper, but fails while applied as a code due to the precision problem.

Here's an example

Can somebody help me how to change this or how I can output more than 4 digits?

I know I can use vpa, but cody doesn't accept vpa and also it is symbolic.

[Walter Roberson]

MATLAB does not round to 4 digits unless you are displaying and you have format short in effect.

Internally MATLAB uses IEEE 754 double precision, which is about 15 1/2 to 16 1/2 digits.

[Walter Roberson]

Really the test suite should not have been coded in terms of isequal() and should have been coded to accept tolerance instead. I will bring this to the attention of some cody people.

[Dyuman Joshi]

I get that Matlab uses IEE 754 double precision but why does it output answer upto 4 digits only? (Edit : kindly ignore my stupidity)

I agree that the test suite should have been coded to accept tolerance instead.

[Stephen23]

"Matlab always rounds the double format to 4 digits after decimal."

I doubt that. How did you check this? Please show us the exact code that you used to show how MATLAb has performed rounding on the stored data value.

"...why does it output answer upto 4 digits only?"

I doubt that it does. Much more likely is that your algorithm accumulates floating point error in a different way to the provided/expected answer, and thus fails the test (even though it is equivalent given the precision of the data type).

Note that the square brackets are totally superflous: square brackets are a concatentation operator (not a "list" operator, which MATLAB does not have) and should be removed completely:

>> polyval(5:-1:0,0.1)
ans =  0.12345

https://www.mathworks.com/matlabcentral/answers/35676-why-not-use-square-brackets

[Walter Roberson]

MATLAB defaults to format short because for most work, about 4 digits is as much accuracy as can be justified by the precision of the inputs, and much of the time people do not care about the additional digits and the digits just make it more difficult to see the information they are looking for. Most of the time, it is more valuable to use the saved space to output more entries per line instead of fewer more precise entries. In the case where people do need the higher precision then format longg command can be used or the system preference can be modified. I personally often need higher precision but most people do not.

[Dyuman Joshi]

"...why does it output answer upto 4 digits only?"

I guess this was my misunderstanding of the format and precision used by MATLAB and partially a mistake as I forgot about format long.

"Note that the square brackets are totally superflous: square brackets are a concatentation operator (not a "list" operator, which MATLAB does not have) and should be removed completely"

I was not aware of this. Thanks for the info.

Answer 1

[goc3]

There are quite a few problems that have been posted to Cody that erroneously use isequal() to test answers, rather than using a tolerance. Each faulty problem needs to have its test suite fixed/updated to properly test against your function output. We have made an effort to fix many of these faulty problems. If you find a Cody problem that still exhibits this discrepancy in required precision, post a comment on that problem to alert us.