The new LinearSVR/SVM do not have a fit_intercept parameter. So AutoML gives an error.

[mail4umar]

When using the AutoML feature for the battery example in examples/business/battery

from verticapy.machine_learning.vertica.automl import AutoML

model = AutoML("battery_data.battery_autoML", 
               estimator = "native")
model.fit("battery_data.finaldata_battery_5", 
          X = ["min_temp",
               "max_temp",
               "min_volt",
               "time_to_reach_minvolt",
               "time_to_reach_maxtemp"],
          y = "SOH")

It throws an error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In [48], line 5
      1 from verticapy.machine_learning.vertica.automl import AutoML
      3 model = AutoML("battery_data.battery_autoML", 
      4                estimator = "native")
----> 5 model.fit("battery_data.finaldata_battery_5", 
      6           X = ["min_temp",
      7                "max_temp",
      8                "min_volt",
      9                "time_to_reach_minvolt",
     10                "time_to_reach_maxtemp"],
     11           y = "SOH")

File /opt/venv/lib/python3.10/site-packages/verticapy/machine_learning/vertica/automl/supervised.py:564, in AutoML.fit(self, input_relation, X, y, return_report)
    554     print(
    555         f"\n\033[1m\033[4mTesting Model - {str(elem.__class__).split('.')[-1][:-2]}\033[0m\033[0m\n"
    556     )
    557 param_grid = gen_params_grid(
    558     elem,
    559     self.parameters["nbins"],
   (...)
    562     self.parameters["optimized_grid"],
    563 )
--> 564 gs = grid_search_cv(
    565     elem,
    566     param_grid,
    567     input_relation,
    568     X,
    569     y,
    570     metric=self.parameters["metric"],
    571     cv=self.parameters["cv"],
    572     pos_label=self.parameters["pos_label"],
    573     cutoff=self.parameters["cutoff"],
    574     training_score=True,
    575     skip_error="no_print",
    576     print_info=self.parameters["print_info"],
    577 )
    578 if (
    579     gs["parameters"] != []
    580     and gs["avg_score"] != []
   (...)
    584     and gs["score_train_std"] != []
    585 ):
    586     result.values["model_type"] += [
    587         str(elem.__class__).split(".")[-1][:-2]
    588     ] * len(gs["parameters"])

File /opt/venv/lib/python3.10/site-packages/verticapy/_utils/_sql/_collect.py:244, in save_verticapy_logs.<locals>.func_prec_save_logs(*args, **kwargs)
    241 json_dict = {**json_dict, **kwargs}
    242 save_to_query_profile(name=name, path=path, json_dict=json_dict)
--> 244 return func(*args, **kwargs)

File /opt/venv/lib/python3.10/site-packages/verticapy/machine_learning/model_selection/hp_tuning/cv.py:1038, in grid_search_cv(estimator, param_grid, input_relation, X, y, metric, cv, average, pos_label, cutoff, training_score, skip_error, print_info, **kwargs)
   1036 # testing all the config
   1037 for config in all_configuration:
-> 1038     estimator.set_params(config)
   1039 # applying all the config
   1040 data = []

File /opt/venv/lib/python3.10/site-packages/verticapy/machine_learning/vertica/base.py:1197, in VerticaModel.set_params(self, parameters, **kwargs)
   1195         warnings.warn(warning_message, Warning)
   1196     new_parameters[p] = parameters[p]
-> 1197 self.__init__(name=self.model_name, **new_parameters)

File /opt/venv/lib/python3.10/site-packages/verticapy/_utils/_sql/_vertica_version.py:115, in check_minimum_version.<locals>.func_prec_check_minimum_version(*args, **kwargs)
    112 name = object_name if fun_name == "__init__" else fun_name
    113 vertica_version(MINIMUM_VERTICA_VERSION[name])
--> 115 return func(*args, **kwargs)

File /opt/venv/lib/python3.10/site-packages/verticapy/_utils/_sql/_collect.py:244, in save_verticapy_logs.<locals>.func_prec_save_logs(*args, **kwargs)
    241 json_dict = {**json_dict, **kwargs}
    242 save_to_query_profile(name=name, path=path, json_dict=json_dict)
--> 244 return func(*args, **kwargs)

TypeError: LinearSVR.__init__() got an unexpected keyword argument 'fit_intercept'

Upon investigation I can see that the following lines inside verticapy/machine_learning/model_selection/hp_tuning/param_gen.py are the cause:

    elif isinstance(estimator, (vml.LinearSVC, vml.LinearSVR)):
        if optimized_grid == 0:
            params_grid = {
                "tol": [1e-4, 1e-6, 1e-8],
                "C": [elem / 1000 for elem in range(1, 5000, math.ceil(5000 / nbins))],
                "fit_intercept": [False, True],
                "intercept_mode": ["regularized", "unregularized"],
                "max_iter": [100, 500, 1000],
            }
            

There are other lines following above which also need to be changed.

The above line of fit_intercept should be taken out. That is IF we have decided to deprecate the fit_intercept option.

Also we can add the parameter intercept_scaling in there as well for grid optimization.

[oualib]

It was corrected in 1.0.4