np.asarray(PandasSeries), with default `copy=None` returns read-only array (pandas 3.0.0)

[NimaSarajpoor]

Summary

As reported in this comment, some tests are failing.

The error shows:

# in core.preprocess_non_normalized

T[~np.isfinite(T)] = np.nan
        ^^^^^^^^^^^^^^^^^^
E       ValueError: assignment destination is read-only

This error occurs because the array T that is returned by the following line of the function core.preprocess_non_normalized is read-only when its input is pandas.Series.

stumpy/stumpy/core.py

Line 2248 in cc6546b

T = _preprocess(T, copy)

---

Minimum Reproducible Example:

import pandas as pd 
import numpy as np

print(pd.__version__)  # 3.0.0
print(np.__version__)  # 2.3.5


T_B = np.array([ 584.,  -11.,   23.,   79., 1001.,    0.,  -19.])
ps = pd.Series(T_B)
T = np.asarray(ps)
T[0] = 100

# Error 
# ValueError: assignment destination is read-only

---

Investigation
Further investigation shows that np.shares_memory(T, ps.array) returns True for the example above, and according to read-only array section in pandas Copy-on-Write (CoW):

Accessing the underlying NumPy array of a DataFrame will return a read-only array if the array shares data with the initial DataFrame:

[seanlaw]

Whatever solution we come up with, we should create a test to check this. This is gonna be tricky.

We'll need to confirm whether there are any instances where we set copy=False? If not, maybe we simply remove the option and always make a costly copy? I don't know

1. In nearly all cases (maybe 100%??), copy is set to True so whether it is a pandas dataframe, numpy array, or polars dataframe, we are already making a copy in nearly all cases and it should never affect the user's original dataframe unless copy is set to False
2. Given the point above, I think it is actually safe to set T.flags.writeable = True after the T = np.asarray(T) line because if T was not a copy then whomever called the function must have deliberately/purposely set copy=False and they must accept that their pandas dataframe can be overwritten! However, if copy=True then we've made a copy of T (even as a dataframe) and that can certainly be (safely) overwritten.

[seanlaw]

Okay, I tried this:

T = np.asarray(T)
T.flags.writeable = True

but it failed with:

ValueError: cannot set WRITEABLE flag to True of this array

So, maybe simply check the state and make a copy as needed:

T = np.asarray(T)
if not T.flags.writeable:
    T = T.copy()

And then we should update all of the docs that have copy as a parameter that this does NOT control the output array but, instead, only whether we make a copy of the input