Add simple_efficiency tests to cover numpy arrays and pandas Series

[aman-coder03]

Description

pvlib.transformer.simple_efficiency supports vectorized operations via
NumPy, but the current test coverage primarily exercises scalar inputs.

This issue proposes adding tests that explicitly cover
numpy.ndarray inputs for:

• input_power
• load_loss
• no_load_loss

including the edge case load_loss = 0.

[echedey-ls]

For reference, this is the file that contains the tests: https://github.com/pvlib/pvlib-python/blob/main/tests/test_transformer.py
It is indeed testing with pd.Series and scalar inputs. I don't oppose nor support checking extra types for simplicity.

[cwhanse]

@echedey-ls I think the three device parameters are only tested as float, not for the range of numeric described in the docstring.

[echedey-ls]

@cwhanse you are right. At first I didn't think it would make much sense to run this function on many transformers at the same time, but maybe it does to pick the best one from a list to fit an specific scenario. I still wonder if added complexity is beneficial. I think documenting this broadcast via a small docstring example is more important than testing it.

[cwhanse]

Many transformers, or perhaps the loss parameters vary with time (temperature dependent?). Either way, it's not the usual use of the function, and we could just change the docstring type from numeric to float.

[echedey-ls]

Many transformers, or perhaps the loss parameters vary with time (temperature dependent?).

AFAIK, loss parameters are just given at nominal conditions. While physically there surely is a dependence on temperature, I hardly think that is measured frequently and with enough reproducibility. Lab testing according to IEC 60076-1 does not include that.

we could just change the docstring type from numeric to float.

I also support this.

[markcampanelli]

I note that not supporting vectorized parameters could make the function harder to use for sensitivity analyses and uncertainty quantification.

[cwhanse]

@markcampanelli I was thinking that the length of parameter vectors for load_loss and no_load_loss would match the length of input_power, i.e., the parameters are indexed by time rather than representing alternative values.

[mikofski]

In my experience the load/no-load parameters are constant.

[kumaradityaapril]

I appreciate the conversation; this background is useful.

It seems reasonable to refrain from suggesting complete vectorization support for the loss parameters based on the comments thus far. Adding tests that cover input_power as a NumPy array (reflecting time-varying load) while maintaining load_loss and no_load_loss as scalars could be a compromise.
Without enlarging the intended API, this would increase coverage.

If that strategy seems appropriate, I'd be happy to work on it.

[cwhanse]

@kumaradityaapril that would be welcome! Scope is to increase test coverage and change the docstring descriptions for load_loss and no_load_loss from numeric to float.

[kumaradityaapril]

I appreciate the explanation. I'm excited to work on this.
As advised, I'll concentrate on expanding test coverage and updating the docstrings for load_loss and no_load_loss.