Most recent numpy version failing tests

[JMGilbert]

In the most recent version of numpy, we need to replace ** with np.float_power

import xarray as xr
import numpy as np
dims = ["dim1", "dim2"]
array = xr.DataArray(
    data=np.array([[5.0, 10.0], [3.0, 2.0]]),
    dims=dims,
    coords=np.array([[1, 2], [3, 4]]),
)
weights = xr.DataArray(
    data=np.array([[0.4, 0.2], [0.1, 0.3]]),
    dims=dims,
    coords=np.array([[1, 2], [3, 4]]),
)  # sum of weights equals 1
eta = 10
exp = 1 / (1 - eta)
utility = np.true_divide(array ** (1 - eta), (1 - eta))
exp_utility = utility.weighted(weights).mean(dim=dims)
ce_array = (exp_utility * (1 - eta)) ** exp
print(ce_array.values == ((np.divide(array ** (1 - 10), 1 - 10) * weights).values.sum() * (1 - 10)) ** (1 / (1 - 10)))

prints False

while

import xarray as xr
import numpy as np
dims = ["dim1", "dim2"]
array = xr.DataArray(
    data=np.array([[5.0, 10.0], [3.0, 2.0]]),
    dims=dims,
    coords=np.array([[1, 2], [3, 4]]),
)
weights = xr.DataArray(
    data=np.array([[0.4, 0.2], [0.1, 0.3]]),
    dims=dims,
    coords=np.array([[1, 2], [3, 4]]),
)  # sum of weights equals 1
eta = 10
exp = 1 / (1 - eta)
utility = np.true_divide(np.float_power(array, (1 - eta)), (1 - eta))
exp_utility = utility.weighted(weights).mean(dim=dims)
ce_array = np.float_power((exp_utility * (1 - eta)), exp)
print(ce_array.values == np.float_power(((np.divide(np.float_power(array, (1 - 10)), 1 - 10) * weights).values.sum() * (1 - 10)),(1 / (1 - 10))))

returns True (the desired result)

[JMGilbert]

#63 fixes the np.float_power issue.

[JMGilbert]

The risk aversion gwr damage function tests still fail for numpy 1.24.1 -- may have to do with numpy/numpy#22956

[kemccusker]

dscim's dependency is pinned to numpy==1.23.3 for now

[brews]

Yeah, I could get it to numpy==1.24.0 but beyond that it starts to fail. Afraid I can't decipher what's going on in testing. Unfortunately, the logs to the failed tests in #63 have expired.

Normally, it can be helpful to check for deprecation warnings in local tests or CI run logs in the section after CI shows test results, but before test coverage gets uploaded. This usually calls out behavior that will break in future releases. We have warnings but it's unclear to me if they are directly related.

Also, looking at numpy's changelog it's kinda weird that tests fail on v1.24.1 and not v1.24.0. Numpy v1.24.0 should have all the breaking behavior changes.

[brews]

A reminder that #11 might be tangled into this?

[brews]

Wat.

All the tests are now passing with numpy v1.25.2 in PR #119. Can we merge the PR and close this?

[JMGilbert]

So strange... I can't think of any reasons not to merge though!

[brews]

Cool. Thanks for the input. Merging away.

[brews]

Closed with PR #119