The implementation of the Morris function is wrong

[mbaudin47]

The implementation of the Morris function at _MorrisSensitivity.py is wrong:

• b2 is wrongly initialized because of a shallow copy,
• b2 is wrongly initialized because k is not updated,
• b3 is wrongly initialized because of a shallow copy,
• b4 is wrongly initialized because of a shallow copy,
• y is overwritten by a dot product, instead of being updated.

Let us detail these problems.

The index k is not updated to initialize b2 from gamma

At:

otbenchmark/otbenchmark/_MorrisSensitivity.py

Line 77 in ce6d551

self.b2[i][j] = gamma[k]

the index k is used, but never updated.

The code should write:

k = 0
for i in range(6, 20):
    for j in range(20):
        self.b2[i][j] = gamma[k]
        k = k + 1

The value y is overwritten by a dot product instead of being updated

At:

otbenchmark/otbenchmark/_MorrisSensitivity.py

Line 105 in ce6d551

y = w.dot(b1)

the value of y is overwritten, instead of being updated.

The correct code is:

y += w.dot(b1)

The variable b2 is initialized from a shallow copy

At:

otbenchmark/otbenchmark/_MorrisSensitivity.py

Line 63 in ce6d551

self.b2 = [[0.0] * 20] * 20

the value of b2 is wrongly defined because it is a shallow copy: many entries share the same memory!
The correct code is:

b2 = [[0.0 for _ in range(20)] for _ in range(20)]

The variable b3 is initialized from a shallow copy

At:

otbenchmark/otbenchmark/_MorrisSensitivity.py

Line 80 in ce6d551

self.b3 = [[[0.0] * 20] * 20] * 20

the same happens for b3. The correct code is:

self.b3 = [[[0.0 for _ in range(20)] for _ in range(20)] for _ in range(20)]

The variable b4 is initialized from a shallow copy

And so for b4 at:

otbenchmark/otbenchmark/_MorrisSensitivity.py

Line 86 in ce6d551

self.b4 = [[[[0.0] * 20] * 20] * 20] * 20

where the correct code is:

self.b4 = [
    [[[0.0 for _ in range(20)] for _ in range(20)] for _ in range(20)]
        for _ in range(20)
    ]

[mbaudin47]

The implementation at: https://gist.github.com/mbaudin47/b7c1502c41ec97d55ab2eb6ee80d6698 fixes these issues. It also improves the speed by a factor 10 to 100.