[Maintenance]: R-Squared computation in metrics module

[bdiptesh]

Describe the maintenance request

Currently the Coefficient of determination/R-Squared is computed using a square of correlation. We need to change it to the following:

Current impact

Currently, in case of target values or predicted values consists of only one value, the output is a NaN which is not entirely correct.

Version

v0.4.0 (Default)

Additional context

No response

[bdiptesh]

Implemented as per sklearn.metrics.r2_score. The below code(s) will give the same output(s).

import numpy as np
from sklearn.metrics import r2_score

from metrics import rsq

arr_len = 1000
y = list(np.random.sample(arr_len))
y_hat = list(np.random.sample(arr_len))

print(rsq(y, y_hat))
print(r2_score(y, y_hat))

When y consists of only one value, the output will be 0.0 as in sklearn.metrics.r2_score

import numpy as np
from sklearn.metrics import r2_score

from metrics import rsq

arr_len = 1000
y = [1.0] * arr_len
y_hat = list(np.random.sample(arr_len))

print(rsq(y, y_hat))
print(r2_score(y, y_hat))