Added support for matrices have NA value - Githubissues

alexeckert / parallelDist

R Package: Parallel Distance Matrix Computation using Multiple Threads

GNU General Public License v2.0

49 stars 9 forks source link

Added support for matrices have NA value #31

Open jlianxm opened 2 years ago

jlianxm commented 2 years ago

Hi Alex,

Thanks for your job! I have added support for matrices have NA value. Please review and let me know if this adds value.

What?

Added support for matrices have NA value. Methods include: euclidean, maximum, manhattan, Canberra, and Minkowski.
Revised the computational formula of Canberra a little, based on dist source code.

Why?

These changes support the calculation of matrices with NA value, since original dist() supports.
The original computational formula of Canberra in parDist() gives different results than dist() when there is more than one NA value in one row. The changes in the computational formula of Canberra correct the results.

How?

Added function proportion() to normalize the results, and function remove_nan() to remove the NA values in matrices.
Revised the calDistance(). The revised calDistance() first removes the NA values in a calculated vector (e.g., A - B) using remove_nan(). Then come up with an intermediate result. Finally, normalize the result and get the final result using proportion().

Testing?

The code has passed the original test cases. I also add a new unit test file (testMatrixNA.R) to test the cases when matrices have NA values.

Best regards, Jiawei