Issues while calling model.fit_predict

[lucamcwood]

Hello,
I am performing a grid search optimization and Claspy is a part of this search.
Sometimes while trying some combinations of hyper-parameters I get two types of error:

1. Python int too large to convert to C long
2. Negative dimensions are not allowed

I am trying to collect the parameter combinations that lead to those errors. I will keep you informed.

_
ERROR:root:Exception occurred: Python int too large to convert to C long
Traceback (most recent call last):
File ".../cp_algorithms.py", line 468, in predict
cps = self.clasp_model.fit_predict(batch)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/claspy/segmentation.py", line 315, in fit_predict
return self.fit(time_series).predict(sparse)
^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/claspy/segmentation.py", line 201, in fit
self.window_size = max(1, map_window_size_methods(self.window_size)(time_series) // 2)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/claspy/window_size.py", line 107, in suss
score = 1 - (_suss_score(time_series, window_size, stats) - min_score) / (max_score - min_score)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/claspy/window_size.py", line 35, in _suss_score
roll_mean = roll.mean().to_numpy()[window_size:]
^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/pandas/core/window/rolling.py", line 2223, in mean
return super().mean(
^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/pandas/core/window/rolling.py", line 1551, in mean
return self._apply(
^^^^^^^^^^^^
File ".../Dude_py311/lib64/python3.11/site-packages/pandas/core/window/rolling.py", line 663, in _apply
return self._apply_blockwise(homogeneous_func, name, numeric_only)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/pandas/core/window/rolling.py", line 503, in _apply_blockwise
return self._apply_series(homogeneous_func, name)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/pandas/core/window/rolling.py", line 487, in _apply_series
result = homogeneous_func(values)
^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/pandas/core/window/rolling.py", line 658, in homogeneous_func
result = calc(values)
^^^^^^^^^^^^
File ".../Dude_py311/lib64/python3.11/site-packages/pandas/core/window/rolling.py", line 655, in calc
return func(x, start, end, min_periods, *numba_args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "pandas/_libs/window/aggregations.pyx", line 255, in pandas._libs.window.aggregations.roll_mean
OverflowError: Python int too large to convert to C long
ERROR:root:Exception occurred: Python int too large to convert to C long
_

_
ERROR:root:Exception occurred: negative dimensions are not allowed
Traceback (most recent call last):
File ".../cp_algorithms.py", line 468, in predict
cps = self.clasp_model.fit_predict(batch)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/claspy/segmentation.py", line 315, in fit_predict
return self.fit(time_series).predict(sparse)
^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/claspy/segmentation.py", line 242, in fit
profile = np.full(shape=self.n_timepoints - self.window_size + 1, fill_value=-np.inf, dtype=np.float64)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib64/python3.11/site-packages/numpy/core/numeric.py", line 344, in full
a = empty(shape, dtype, order)
^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: negative dimensions are not allowed
_

[ermshaua]

Looks like the window size makes problems here. Make sure that the window size is always a lot smaller than the time series size, otherwise the algorithm runs into problems.

As an upper bound for the window size, you could choose maybe 0.1 * time series size. Also, the window size should not be smaller than at least 3 or 5 values. So use such a constant as your lower bound.

Let me know if this helps!