using .quantile_mut() on Array1::<f64> #58

Activity

1. LukeMathWalker commented on Dec 4, 2019

   

   LukeMathWalker

   on Dec 4, 2019

   Member

   More actions

   Rust's floats (f64 and f32) do not implement the Ord trait because of how f64::NAN and f32::NAN behave in comparisons - see this little snippet https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=6190375f8d1a48bc1a885ce38981ebea

   If you remove Not a Number from the equation (using something like noisy-floats) than you recover the Ord trait and you can call .quantile_mut.

   As an action item on our side, we should probably provide a PartialOrd version of quantile_mut, as we do for the other summary function.
2. LukeMathWalker

   added

   EnhancementNew feature or requestNew feature or request

   Good first issueGood for newcomersGood for newcomers

   on Dec 4, 2019
3. CharlieBickerton commented on Dec 4, 2019

   

   CharlieBickerton

   on Dec 4, 2019

   Last edited by CharlieBickerton

   Author

   More actions

   Thanks for the quick response @LukeMathWalker , that's annoying f32/f64 doen't impelment Ord . I made my array n64's and it works great.

   It was potentially me reading the docs wrong as I had the impression that the q value needed to be noisy rather than the array which didn't make much sense to me. Thanks for clearning it up!

   👍React with 👍1Reacted by Luca Palmieri
4. jturner314 commented on Dec 4, 2019

   

   jturner314

   on Dec 4, 2019

   Member

   More actions

   The quantile_axis_skipnan_mut method on the QuantileExt trait may also be useful for your application since it can work directly on f32/f64 in the presence of NaN values. We should probably add an analogous method to Quantile1dExt; @LukeMathWalker is that what you were referring to in your comment?
5. CharlieBickerton commented on Dec 5, 2019

   

   CharlieBickerton

   on Dec 5, 2019

   Author

   More actions

   @jturner314 that sounds like a great idea, thanks for pointing out quantile_axis_skipnan_mut, I could add an axis to my 1d arrays and use this but I think it's cleaner to stay in 1d.
6. LukeMathWalker commented on Dec 5, 2019

   

   LukeMathWalker

   on Dec 5, 2019

   Member

   More actions

   Yes, that's what I meant @jturner314.
7. jturner314 commented on Dec 5, 2019

   

   jturner314

   on Dec 5, 2019

   Member

   More actions

   @CharlieBickerton Note that 1d arrays have a single axis (axis 0). quantile_axis_skipnan_mut returns an array with one less axis, so the result of calling it on a 1d array is a 0d array. To turn an 0d array into its element, you can use .into_scalar(). So, you could do this:

   use ndarray::prelude::*;
   use ndarray_stats::{interpolate::Nearest, QuantileExt};
   use noisy_float::types::n64;
   
   fn main() -> Result<(), Box<dyn std::error::Error>> {
       let mut arr: Array1<f64> = array![0., std::f64::NAN, 1., std::f64::NAN, 2., 3., 4.];
       assert_eq!(
           2.,
           arr.quantile_axis_skipnan_mut(Axis(0), n64(0.5), &Nearest)?
               .into_scalar()
       );
       Ok(())
   }

   It would be more convenient to have a .quantile_skipnan_mut() method in Quantile1dExt, as @LukeMathWalker pointed out.