added more precondition

[SumanSudhir]

#517
Precondiiton for function:

• replacing
• all
• any
• max
• min
• argmax
• argmin
• sum
• product
• mean
• variance
• cumulativeSum
• cumulativeProduct

[SumanSudhir]

@dan-zheng I have change precondition of areAxesInRange() as it is not necessary that the input axes tensor must have rank 1. Eg: cumlativeSum() and moments the former require rank 0 tensor while later takes rank 1 tensor as input

[Shashi456]

In cumprod and cumsum, if the axis is plural then I think the documentation must read, All axes must be in range ..., Like the previous conditions.
If it's a singular value, is calling the function necessary instead of just doing the check locally ?

[SumanSudhir]

Previously it was given axis instead of axes and also the same comment was written above that axis must be in the range -rank..<rank. So I followed the same pattern in writing the precondition. Changing the input variable from axis to axes may conflict with some written test cases.

[Shashi456]

You don't have to write all axis, it's just referring to a singular axis. You can start the statement with Axis must be in range...

[SumanSudhir]

I think rank 0 tensor means that there is only one value in axis tensor. So previously it was written as singular.
var input: Tensor<Int32> = Tensor(5)
input.rank is giving 0 as output

[SumanSudhir]

@dan-zheng can you please review it?

[dan-zheng]

@dan-zheng I have change precondition of areAxesInRange() as it is not necessary that the input axes tensor must have rank 1. Eg: cumlativeSum() and moments the former require rank 0 tensor while later takes rank 1 tensor as input

Thanks for the clarification!

The correct fix is actually to add another precondition helper method called isAxisInRange taking a Tensor<Int32>, not to call areAxesInRange(_: Tensor<Int32>) on a scalar Tensor<Int32> representing a single axis.

Tensor.isAxisInRange(_: Tensor<Int32>) should have a precondition:

precondition(axis.rank == 0, "Axis must have rank 0.")

And Tensor.areAxesInRange(_: Tensor<Int32>) should have the old precondition:

precondition(axes.rank == 1, "Axis must have rank 1.")

Suggested change

	precondition(areAxesInRange(axis), "Axis must be in the range `[-rank, rank)`.")
	precondition(isAxisInRange(axis), "Axis must be in the range `[-rank, rank)`.")

[dan-zheng]

Nice!

[dan-zheng]

Weirdly, there's a test failure:

/swift-apis/Tests/TensorFlowTests/OperatorTests/LinearAlgebraTests.swift:77: error: LinearAlgebraTests.testSVD : XCTAssertEqual failed: ("-3.0785367") is not equal to ("-3.0785267") +/- ("1e-05") -
/swift-apis/Tests/TensorFlowTests/OperatorTests/LinearAlgebraTests.swift:89: error: LinearAlgebraTests.testSVD : XCTAssertEqual failed: ("-3.0785367") is not equal to ("-3.0785267") +/- ("1e-05") -
Test Case 'LinearAlgebraTests.testSVD' failed (0.012 seconds)

Is this a flaky test? Rerunning tests now.