`ValueError: assignment destination is read-only` in `BaseSLearner.predict` with `catboost` and `polars`

[PelinV]

Describe the bug
Using polars (cast to numpy on fit and predict) get an error in predict on X_new[:, 0] = 1 (118 row in slearner.py)
Cause CatBoostRegressor (and CatBoostClassifier) change writeable flag on array

To Reproduce
Steps to reproduce the behavior:

n_train_rows = 10_000
n_test_rows = 5_000
n_features = 10
feature_names = [f'X_{i}' for i in range(n_features)]

X_np = np.random.randint(0, 100, (n_train_rows, n_features))
y_np = np.random.randint(0, 100, (n_train_rows, 1))
X_test_np = np.random.randint(0, 100, (n_train_rows, n_features))

X_pd = pd.DataFrame(np.random.randint(0, 100, (n_train_rows, n_features)), columns=feature_names)
y_pd = pd.DataFrame(np.random.randint(0, 100, (n_train_rows, 1)), columns=['y'])
X_test_pd = pd.DataFrame(np.random.randint(0, 100, (n_train_rows, n_features)), columns=feature_names)

X_pd = pl.DataFrame(np.random.randint(0, 100, (n_train_rows, n_features)), schema=feature_names)
y_pd = pl.DataFrame(np.random.randint(0, 100, (n_train_rows, 1)), schema=['y'])
X_test_pd = pl.DataFrame(np.random.randint(0, 100, (n_train_rows, n_features)), schema=feature_names)

# 1. Numpy (no errors, save all writeable flags)
X = X_np
y = y_np
X_test = X_test_np
print(
    'Before',
    f'{X.flags.writeable=}',  # True
    f'{y.flags.writeable=}',  # True
    f'{X_test.flags.writeable=}',  # True
    sep='\n',
)
model = CatBoostRegressor(iterations=10)  # `10` for debug
model.fit(X, y)
model.predict(X_test)
print(
    'After',
    f'{X.flags.writeable=}',  # True
    f'{y.flags.writeable=}',  # True
    f'{X_test.flags.writeable=}',  # True
    sep='\n',
)
s_model = BaseSLearner(CatBoostRegressor(iterations=10))
s_model.fit(X, y)
s_model.predict(X_test)

# 2. Pandas -> Numpy (no errors, save all writeable flags)
X = X_pd.to_numpy()
y = y_pd.to_numpy()
X_test = X_test_pd.to_numpt()
print(
    'Before',
    f'{X.flags.writeable=}',  # True
    f'{y.flags.writeable=}',  # True
    f'{X_test.flags.writeable=}',  # True
    sep='\n',
)
model = CatBoostRegressor(iterations=10)  # `10` for debug
model.fit(X, y)
model.predict(X_test)
print(
    'After',
    f'{X.flags.writeable=}',  # True
    f'{y.flags.writeable=}',  # True
    f'{X_test.flags.writeable=}',  # True
    sep='\n',
)
s_model = BaseSLearner(CatBoostRegressor(iterations=10))
s_model.fit(X, y)
s_model.predict(X_test)


# 3. Polars -> Numpy (ValueError, do not save writeable flags)
X = X_df.to_numpy()
y = y_df.to_numpy()
X_test = X_test_df.to_numpy()
print(
    'Before',
    f'{X.flags.writeable=}',  # True
    f'{y.flags.writeable=}',  # False
    f'{X_test.flags.writeable=}',  # True
    sep='\n',
)
model = CatBoostRegressor(iterations=10)  # `10` for debug
model.fit(X, y)
model.predict(X_test)
print(
    'After',
    f'{X.flags.writeable=}',  # False (was True)
    f'{y.flags.writeable=}',  # False
    f'{X_test.flags.writeable=}',  # False (was True)
    sep='\n',
)
s_model = BaseSLearner(CatBoostRegressor(iterations=10))
s_model.fit(X, y)
try:
    s_model.predict(X_test)  # ValueError
except ValueError:
    pass


# 4. Polars -> Pandas -> Numpy (ValueError, do not save writeable flags)
X = X_df.to_pandas().to_numpy()
y = y_df.to_pandas().to_numpy()
X_test = X_test_df.to_pandas().to_numpy()
print(
    'Before',
    f'{X.flags.writeable=}',  # True
    f'{y.flags.writeable=}',  # True
    f'{X_test.flags.writeable=}',  # True
    sep='\n',
)
model = CatBoostRegressor(iterations=10)  # `10` for debug
model.fit(X, y)
model.predict(X_test)
print(
    'After',
    f'{X.flags.writeable=}',  # False (was True)
    f'{y.flags.writeable=}',  # True (was False)
    f'{X_test.flags.writeable=}',  # False (was True)
    sep='\n',
)
s_model = BaseSLearner(CatBoostRegressor(iterations=10))
s_model.fit(X, y)
try:
    s_model.predict(X_test)  # ValueError
except ValueError:
    pass

Expected behavior
Add X_new.flags.writeable = True to fix this error

Environment (please complete the following information):

Linux-4.18.0-305.72.1.el8_4.x86_64-x86_64-with-glibc2.28
Python 3.11.9 (main, Jun 19 2024, 10:02:06) [GCC 8.5.0 20210514 (Red Hat 8.5.0-22)]

causalml==0.15.5
catboost==1.2.8
numpy==2.2.6
pandas==2.2.3
polars==1.32.2

Additional context
Add any other context about the problem here.