Issue using .fit()

[aaronnkang]

Every time I try use .fit(n) to fit a pwlf with n number of segments I get the following error:

The map-like callable must be of the form f(func, iterable), returning a sequence of numbers the same length as 'iterable'

Any ideas?

[cjekel]

I haven't seen this before. It's possible something I use from scipy has changed. Can you post or send me a full stack trace of the error? Just so I can see which line in pwlf is triggering this use l issue.

…

On Thu, Apr 14, 2022, 06:40 aaronnkang ***@***.***> wrote: Every time I try used .fit(n) to fit a pwlf with n number of segments I get the following error: The map-like callable must be of the form f(func, iterable), returning a sequence of numbers the same length as 'iterable' Any ideas? — Reply to this email directly, view it on GitHub <#94>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADJ5Z4L77G6XVACZ2NXGASTVFAN6FANCNFSM5TN2BDSA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[cjekel]

Can you tell me which version of python and scipy are you using?

…

On Thu, Apr 14, 2022, 07:23 Charles Jekel ***@***.***> wrote: I haven't seen this before. It's possible something I use from scipy has changed. Can you post or send me a full stack trace of the error? Just so I can see which line in pwlf is triggering this use l issue. On Thu, Apr 14, 2022, 06:40 aaronnkang ***@***.***> wrote: > Every time I try used .fit(n) to fit a pwlf with n number of segments I > get the following error: > > The map-like callable must be of the form f(func, iterable), returning a > sequence of numbers the same length as 'iterable' > > Any ideas? > > — > Reply to this email directly, view it on GitHub > <#94>, or > unsubscribe > <https://github.com/notifications/unsubscribe-auth/ADJ5Z4L77G6XVACZ2NXGASTVFAN6FANCNFSM5TN2BDSA> > . > You are receiving this because you are subscribed to this thread.Message > ID: ***@***.***> >

[camjay99]

I am getting the same issue, it appears the DifferentialEvolutionSolver is generating infs/nans that cause the error. I am running Python 3.9.13 and Scipy 1.9.1.

---

ValueError Traceback (most recent call last)
File ~\anaconda3\envs\gee\lib\site-packages\scipy\optimize_differentialevolution.py:1116, in DifferentialEvolutionSolver._calculate_population_energies(self, population)
1115 try:
-> 1116 calc_energies = list(
1117 self._mapwrapper(self.func, parameters_pop[0:S])
1118 )
1119 calc_energies = np.squeeze(calc_energies)

File ~\anaconda3\envs\gee\lib\site-packages\scipy_lib_util.py:407, in _FunctionWrapper.call(self, x)
406 def call(self, x):
--> 407 return self.f(x, *self.args)

File ~\anaconda3\envs\gee\lib\site-packages\pwlf\pwlf.py:590, in PiecewiseLinFit.fit_with_breaks_opt(self, var)
588 try:
589 # least squares solver
--> 590 ssr = self.lstsq(A, calc_slopes=False)
592 except linalg.LinAlgError:
593 # the computation could not converge!
594 # on an error, return ssr = np.inf
595 # You might have a singular Matrix!!!

File ~\anaconda3\envs\gee\lib\site-packages\pwlf\pwlf.py:1491, in PiecewiseLinFit.lstsq(self, A, calc_slopes)
1490 if self.weights is None:
-> 1491 beta, ssr, _, _ = linalg.lstsq(A, self.y_data,
1492 lapack_driver=self.lapack_driver)
1493 # ssr is only calculated if self.n_data > self.n_parameters
1494 # in this case I'll need to calculate ssr manually
1495 # where ssr = sum of square of residuals

File ~\anaconda3\envs\gee\lib\site-packages\scipy\linalg_basic.py:1135, in lstsq(a, b, cond, overwrite_a, overwrite_b, check_finite, lapack_driver)
1134 a1 = _asarray_validated(a, check_finite=check_finite)
-> 1135 b1 = _asarray_validated(b, check_finite=check_finite)
1136 if len(a1.shape) != 2:

File ~\anaconda3\envs\gee\lib\site-packages\scipy_lib_util.py:287, in _asarray_validated(a, check_finite, sparse_ok, objects_ok, mask_ok, as_inexact)
286 toarray = np.asarray_chkfinite if check_finite else np.asarray
--> 287 a = toarray(a)
288 if not objects_ok:

File ~\anaconda3\envs\gee\lib\site-packages\numpy\lib\function_base.py:627, in asarray_chkfinite(a, dtype, order)
626 if a.dtype.char in typecodes['AllFloat'] and not np.isfinite(a).all():
--> 627 raise ValueError(
628 "array must not contain infs or NaNs")
629 return a

ValueError: array must not contain infs or NaNs

The above exception was the direct cause of the following exception:

RuntimeError Traceback (most recent call last)
Cell In [31], line 1
----> 1 evi = process_point(all_data, 54, 1)

Cell In [25], line 30, in process_point(dataset, latitude, longitude)
27 print(df)
29 # Fits a piecewise linear regression to data
---> 30 model = fit_pwlf(df, break_guesses)
32 # Calculate anomaly
33 df['EVI_pred'] = model.predict(df['doy_nrml'])

Cell In [27], line 60, in fit_pwlf(df, break_guesses)
58 def fit_pwlf(df, break_guesses):
59 model = pwlf.PiecewiseLinFit(df['doy_nrml'], df['EVI_nrml'])
---> 60 model.fit(8)#fit_guess(break_guesses)
61 return model

File ~\anaconda3\envs\gee\lib\site-packages\pwlf\pwlf.py:771, in PiecewiseLinFit.fit(self, n_segments, x_c, y_c, bounds, **kwargs)
769 # run the optimization
770 if len(kwargs) == 0:
--> 771 res = differential_evolution(min_function, bounds,
772 strategy='best1bin', maxiter=1000,
773 popsize=50, tol=1e-3,
774 mutation=(0.5, 1), recombination=0.7,
775 seed=None, callback=None, disp=False,
776 polish=True, init='latinhypercube',
777 atol=1e-4)
778 else:
779 res = differential_evolution(min_function,
780 bounds, **kwargs)

File ~\anaconda3\envs\gee\lib\site-packages\scipy\optimize_differentialevolution.py:392, in differential_evolution(func, bounds, args, strategy, maxiter, popsize, tol, mutation, recombination, seed, callback, disp, polish, init, atol, updating, workers, constraints, x0, integrality, vectorized)
375 # using a context manager means that any created Pool objects are
376 # cleared up.
377 with DifferentialEvolutionSolver(func, bounds, args=args,
378 strategy=strategy,
379 maxiter=maxiter,
(...)
390 integrality=integrality,
391 vectorized=vectorized) as solver:
--> 392 ret = solver.solve()
394 return ret

File ~\anaconda3\envs\gee\lib\site-packages\scipy\optimize_differentialevolution.py:984, in DifferentialEvolutionSolver.solve(self)
979 self.feasible, self.constraint_violation = (
980 self._calculate_population_feasibilities(self.population))
982 # only work out population energies for feasible solutions
983 self.population_energies[self.feasible] = (
--> 984 self._calculate_population_energies(
985 self.population[self.feasible]))
987 self._promote_lowest_energy()
989 # do the optimization.

File ~\anaconda3\envs\gee\lib\site-packages\scipy\optimize_differentialevolution.py:1123, in DifferentialEvolutionSolver._calculate_population_energies(self, population)
1119 calc_energies = np.squeeze(calc_energies)
1120 except (TypeError, ValueError) as e:
1121 # wrong number of arguments for _mapwrapper
1122 # or wrong length returned from the mapper
-> 1123 raise RuntimeError(
1124 "The map-like callable must be of the form f(func, iterable), "
1125 "returning a sequence of numbers the same length as 'iterable'"
1126 ) from e
1128 if calc_energies.size != S:
1129 if self.vectorized:

RuntimeError: The map-like callable must be of the form f(func, iterable), returning a sequence of numbers the same length as 'iterable'

[cjekel]

Thanks for that full traceback. I think I know what the issue is, it's this block of code in the lstsq fit

        try:
            ssr = self.lstsq(A)

        except linalg.LinAlgError:
            # the computation could not converge!
            # on an error, return ssr = np.print_function
            # You might have a singular Matrix!!!
            ssr = np.inf
        if ssr is None:
            ssr = np.inf
            # something went wrong...
        self.ssr = ssr

The new version of DE seems not to be able to handle np.inf. I have not been following how DE has been changing in scipy. It has gone through several changes throughout the years. It might be a bug with scipy's DE if this used to be supported.

As a hotfix, can you try using a very very big number instead of np.inf?

[camjay99]

Sorry for the false alarm, after further exploration, I discovered my preprocessing was generating a NaN that was causing the issue. The code appears to work fine as is.