Repository navigation
Implement argmax on a PMMatrix #80
Description
Implement something similar to this one : https://docs.scipy.org/doc/numpy/reference/generated/numpy.argmax.html on PMMatrix
Implement something similar to this one : https://docs.scipy.org/doc/numpy/reference/generated/numpy.argmax.html on PMMatrix
Activity
nicolas-cellier-aka-nice commentedon Aug 3, 2018
The numpy example is quite stupid because ambiguous: we don't know if 5 is the 0-based index of the maximum value, or the maximum value itself!
Also we don't know what happens with negative values... I presume it's the maximum value and not the maximum magnitude, but it would be good to have an example for disambiguating.
Answering a single index only make sense for an underlying flatten collection like in Smallapack.
But this is not the case of Didier Besset matrix implementation (Array of Array).
Can we access a PMMatrix with a single flat index?
Otherwise we should answer a row and a column index (or alternatively provide a block [:rowIndex :colIndex | ] as argument).
nicolas-cellier-aka-nice commentedon Aug 3, 2018
Note that in Smallapack I've implemented
findMax:with a block argument which is(self collect: aBlock) argmax.I also have
findAll:which is equivalent of Matlab find (same as select: but answers the indices rather than the elements).SergeStinckwich commentedon Aug 3, 2018
Yes the Numpy example is quite stupid ... Apparently without axis, the matrix is flattened before finding the argmax.
Following discussion #79 and #80, I rename the flatten methods and ad…
Merge dev commits to master for the 1.0 release (#145)