Refactor path

[yinweisu]

Issue #, if available:

• [Tabular] Improve path logic #3260

Description of changes:

• Refactor path logic in Tabular/TimeSeries, so that parent only will track relative paths to its children and passing correct path to its children when init/load
• Trainer now stores model's relative path as a list, and dynamically construct them to a path.
• Replaced usage os os.path.sep + str concatenation in path with os.path.join to avoid bugs

• Cross OS check for tabular
• Cross OS check for timeseries

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Job PR-3355-705d09d is done.
Docs are uploaded to http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-3355/705d09d/index.html

[github-actions]

Job PR-3355-18ef70f is done.
Docs are uploaded to http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-3355/18ef70f/index.html

[Innixma]

LGTM, this is an excellent and much needed refactor!

[shchur]

Thank you for the PR, this looks great!

[eromoe]

Similar error:
ValueError: path is on mount 'C:', start on mount 'E:'

I run code in E , and set path to C

TabularPredictor(label='y', eval_metric='precision', path=autgoluon_model_path)

ValueError                                Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_17032/2841266324.py in <module>
----> 1 m1 = train(X1_train, y1_train, X1_test, y1_test)

~\AppData\Local\Temp/ipykernel_17032/2534130983.py in train(X_train, y_train, X_test, y_test)
     42         m = model_cls(eval_metric=eval_metric, path=new_model_path)
     43
---> 44         m.fit(X_train, y_train)
     45
     46         if X_test.shape[0]:

e:\workspace\github_me\mlquant\mlquant\my_auto.py in fit(self, X, y, **kwargs)
     30
     31     def fit(self, X, y, **kwargs):
---> 32         return super().fit(X.assign(y=y), hyperparameters=self.hyperparameters, verbosity=0, **kwargs)
     33
     34     def predict_proba(self, data, model=None,):

~\anaconda3\lib\site-packages\autogluon\core\utils\decorators.py in _call(*args, **kwargs)
     29         def _call(*args, **kwargs):
     30             gargs, gkwargs = g(*other_args, *args, **kwargs)
---> 31             return f(*gargs, **gkwargs)
     32
     33         return _call

~\anaconda3\lib\site-packages\autogluon\tabular\predictor\predictor.py in fit(self, train_data, tuning_data, time_limit, presets, hyperparameters, feature_metadata, infer_limit, infer_limit_batch_size, fit_weighted_ensemble, calibrate_decision_threshold, num_cpus, num_gpus, **kwargs)
    984             aux_kwargs["fit_weighted_ensemble"] = False
    985         self.save(silent=True)  # Save predictor to disk to enable prediction and training after interrupt
--> 986         self._learner.fit(
    987             X=train_data,
    988             X_val=tuning_data,

~\anaconda3\lib\site-packages\autogluon\tabular\learner\abstract_learner.py in fit(self, X, X_val, **kwargs)
    157             raise AssertionError("Learner is already fit.")
    158         self._validate_fit_input(X=X, X_val=X_val, **kwargs)
--> 159         return self._fit(X=X, X_val=X_val, **kwargs)
    160
    161     def _fit(

~\anaconda3\lib\site-packages\autogluon\tabular\learner\default_learner.py in _fit(self, X, X_val, X_unlabeled, holdout_frac, num_bag_folds, num_bag_sets, time_limit, infer_limit, infer_limit_batch_size, verbosity, **trainer_fit_kwargs)
    133             time_limit_trainer = None
    134
--> 135         trainer = self.trainer_type(
    136             path=self.model_context,
    137             problem_type=self.label_cleaner.problem_type_transform,

~\anaconda3\lib\site-packages\autogluon\core\trainer\abstract_trainer.py in __init__(self, path, problem_type, eval_metric, num_classes, quantile_levels, low_memory, feature_metadata, k_fold, n_repeats, sample_weight, weight_evaluation, save_data, random_state, verbosity)
    102     ):
    103         self.path = path
--> 104         self.path = PathConverter.to_relative(self.path)
    105         self.problem_type = problem_type
    106         self.feature_metadata = feature_metadata

~\anaconda3\lib\site-packages\autogluon\common\utils\path_converter.py in to_relative(path)
     54             return path
     55         os_path_sep = PathConverter.os_path_sep()
---> 56         path_relative = os.path.relpath(path)
     57         if path.endswith(os_path_sep):
     58             if not path_relative.endswith(os_path_sep):

~\anaconda3\lib\ntpath.py in relpath(path, start)
    701         path_drive, path_rest = splitdrive(path_abs)
    702         if normcase(start_drive) != normcase(path_drive):
--> 703             raise ValueError("path is on mount %r, start on mount %r" % (
    704                 path_drive, start_drive))
    705

ValueError: path is on mount 'C:', start on mount 'E:'