[BUG] Diviner is incompatible with pmdarima>=2.0.0

[harupy]

pmdarima 2.0.0 broke diviner:

https://github.com/mlflow/mlflow/runs/7974149279?check_suite_focus=true#step:12:417

________________________ test_diviner_pyfunc_save_load _________________________

grouped_pmdarima = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
model_path = PosixPath('/tmp/pytest-of-runner/pytest-0/test_diviner_pyfunc_save_load0/model')

    def test_diviner_pyfunc_save_load(grouped_pmdarima, model_path):
    
        mlflow.diviner.save_model(diviner_model=grouped_pmdarima, path=model_path)
        loaded_pyfunc = pyfunc.load_pyfunc(model_uri=model_path)
    
>       model_predict = grouped_pmdarima.predict(n_periods=10, return_conf_int=True, alpha=0.075)

grouped_pmdarima = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
loaded_pyfunc = mlflow.pyfunc.loaded_model:
  flavor: mlflow.diviner

model_path = PosixPath('/tmp/pytest-of-runner/pytest-0/test_diviner_pyfunc_save_load0/model')

tests/diviner/test_diviner_model_export.py:98: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:405: in predict
    return self._run_predictions(prediction_config, exog=exog, **predict_kwargs)
        alpha      = 0.075
        exog       = None
        inverse_transform = True
        n_periods  = 10
        predict_col = 'yhat'
        predict_kwargs = {}
        prediction_config =    n_periods  alpha  return_conf_int  inverse_transform key2 key1 key0
0         10  0.075             True           ...          True               True    L    A    K
3         10  0.075             True               True    Z    X    B
        return_conf_int = True
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:335: in _run_predictions
    for idx, row in processing_data.iterrows()
        df         =    n_periods  alpha  return_conf_int  inverse_transform key2 key1 key0
0         10  0.075             True           ...          True               True    L    A    K
3         10  0.075             True               True    Z    X    B
        exog       = None
        n_periods_col = 'n_periods'
        predict_kwargs = {}
        processing_data =    n_periods  alpha  return_conf_int  ...  key1 key0 grouping_key
0         10  0.075             True  ...     P    T...ue  ...     A    K    (L, A, K)
3         10  0.075             True  ...     X    B    (Z, X, B)

[4 rows x 8 columns]
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:335: in <listcomp>
    for idx, row in processing_data.iterrows()
        .0         = <generator object DataFrame.iterrows at 0x7fddfef0a6d0>
        exog       = None
        idx        = 0
        n_periods_col = 'n_periods'
        predict_kwargs = {}
        row        = n_periods                   10
alpha                    0.075
return_conf_int           True
inverse_transform        ...  H
key1                         P
key0                         T
grouping_key         (H, P, T)
Name: 0, dtype: object
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:301: in _predict_single_group
    **predict_kwargs,
        alpha      = 0.075
        exog       = None
        group_key  = ('H', 'P', 'T')
        inverse_transform = True
        model      = AutoARIMA(maxiter=30, out_of_sample_size=60)
        n_periods_col = 'n_periods'
        periods    = 10
        predict_kwargs = {}
        return_conf_int = True
        row_entry  = n_periods                   10
alpha                    0.075
return_conf_int           True
inverse_transform        ...  H
key1                         P
key0                         T
grouping_key         (H, P, T)
Name: 0, dtype: object
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

args = ()
kwargs = {'X': None, 'alpha': 0.075, 'inverse_transform': True, 'n_periods': 10, ...}

>   out = (lambda *args, **kwargs: self.fn(obj, *args, **kwargs))
E   TypeError: predict() got an unexpected keyword argument 'inverse_transform'

args       = ()
kwargs     = {'X': None, 'alpha': 0.075, 'inverse_transform': True, 'n_periods': 10, ...}
obj        = AutoARIMA(maxiter=30, out_of_sample_size=60)
self       = <pmdarima.utils.metaestimators._IffHasDelegate object at 0x7fde0085e350>

/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/pmdarima/utils/metaestimators.py:53: TypeError
__________________ test_diviner_pyfunc_group_predict_pmdarima __________________

grouped_pmdarima = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
model_path = PosixPath('/tmp/pytest-of-runner/pytest-0/test_diviner_pyfunc_group_pred1/model')
diviner_data = Structure(df=                      ds           y key2 key1 key0
0    2019-01-01 00:01:00  946.607797    J    Z    G
1...  T
1[45](https://github.com/mlflow/mlflow/runs/7974149279?check_suite_focus=true#step:12:46)9 2022-12-30 00:01:00  435.780587    H    P    T

[5840 rows x 5 columns], key_columns=['key2', 'key1', 'key0'])

    def test_diviner_pyfunc_group_predict_pmdarima(grouped_pmdarima, model_path, diviner_data):
    
        groups = []
        for i in [0, -1]:
            key_entries = []
            for value in diviner_data.df[diviner_data.key_columns].iloc[[i]].to_dict().values():
                key_entries.append(list(value.values())[0])
            groups.append(tuple(key_entries))
    
        mlflow.diviner.save_model(diviner_model=grouped_pmdarima, path=model_path)
        loaded_pyfunc_model = pyfunc.load_pyfunc(model_uri=model_path)
    
        local_group_pred = grouped_pmdarima.predict_groups(
            groups=groups,
            n_periods=10,
            predict_col="prediction",
            alpha=0.1,
            return_conf_int=True,
>           on_error="warn",
        )

diviner_data = Structure(df=                      ds           y key2 key1 key0
0    2019-01-01 00:01:00  9[46](https://github.com/mlflow/mlflow/runs/7974149279?check_suite_focus=true#step:12:47).607797    J    Z    G
1...  T
1459 2022-12-30 00:01:00  435.780587    H    P    T

[5840 rows x 5 columns], key_columns=['key2', 'key1', 'key0'])
grouped_pmdarima = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
groups     = [('J', 'Z', 'G'), ('H', 'P', 'T')]
i          = -1
key_entries = ['H', 'P', 'T']
loaded_pyfunc_model = mlflow.pyfunc.loaded_model:
  flavor: mlflow.diviner

model_path = PosixPath('/tmp/pytest-of-runner/pytest-0/test_diviner_pyfunc_group_pred1/model')
value      = {1459: 'T'}

tests/diviner/test_diviner_model_export.py:179: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:[48](https://github.com/mlflow/mlflow/runs/7974149279?check_suite_focus=true#step:12:49)6: in predict_groups
    return self._run_predictions(prediction_config, exog=exog, **predict_kwargs)
        alpha      = 0.1
        exog       = None
        groups     = [('J', 'Z', 'G'), ('H', 'P', 'T')]
        inverse_transform = False
        n_periods  = 10
        on_error   = 'warn'
        predict_col = 'prediction'
        predict_kwargs = {}
        prediction_config =    n_periods  alpha  return_conf_int  inverse_transform key2 key1 key0
0         10    0.1             True              False    H    P    T
1         10    0.1             True              False    J    Z    G
        return_conf_int = True
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:335: in _run_predictions
    for idx, row in processing_data.iterrows()
        df         =    n_periods  alpha  return_conf_int  inverse_transform key2 key1 key0
0         10    0.1             True              False    H    P    T
1         10    0.1             True              False    J    Z    G
        exog       = None
        n_periods_col = 'n_periods'
        predict_kwargs = {}
        processing_data =    n_periods  alpha  return_conf_int  ...  key1 key0 grouping_key
0         10    0.1             True  ...     P    T    (H, P, T)
1         10    0.1             True  ...     Z    G    (J, Z, G)

[2 rows x 8 columns]
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:335: in <listcomp>
    for idx, row in processing_data.iterrows()
        .0         = <generator object DataFrame.iterrows at 0x7fddfeca86[50](https://github.com/mlflow/mlflow/runs/7974149279?check_suite_focus=true#step:12:51)>
        exog       = None
        idx        = 0
        n_periods_col = 'n_periods'
        predict_kwargs = {}
        row        = n_periods                   10
alpha                      0.1
return_conf_int           True
inverse_transform        ...  H
key1                         P
key0                         T
grouping_key         (H, P, T)
Name: 0, dtype: object
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:301: in _predict_single_group
    **predict_kwargs,
        alpha      = 0.1
        exog       = None
        group_key  = ('H', 'P', 'T')
        inverse_transform = False
        model      = AutoARIMA(maxiter=30, out_of_sample_size=60)
        n_periods_col = 'n_periods'
        periods    = 10
        predict_kwargs = {}
        return_conf_int = True
        row_entry  = n_periods                   10
alpha                      0.1
return_conf_int           True
inverse_transform        ...  H
key1                         P
key0                         T
grouping_key         (H, P, T)
Name: 0, dtype: object
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

args = ()
kwargs = {'X': None, 'alpha': 0.1, 'inverse_transform': False, 'n_periods': 10, ...}

>   out = (lambda *args, **kwargs: self.fn(obj, *args, **kwargs))
E   TypeError: predict() got an unexpected keyword argument 'inverse_transform'

args       = ()
kwargs     = {'X': None, 'alpha': 0.1, 'inverse_transform': False, 'n_periods': 10, ...}
obj        = AutoARIMA(maxiter=30, out_of_sample_size=60)
self       = <pmdarima.utils.metaestimators._IffHasDelegate object at 0x7fde0085e350>

/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/pmdarima/utils/metaestimators.py:53: TypeError
__________________ test_diviner_load_from_remote_uri_succeeds __________________

grouped_pmdarima = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
model_path = PosixPath('/tmp/pytest-of-runner/pytest-0/test_diviner_load_from_remote_0/model')
mock_s3_bucket = 'mock-bucket'

    def test_diviner_load_from_remote_uri_succeeds(grouped_pmdarima, model_path, mock_s3_bucket):
        mlflow.diviner.save_model(diviner_model=grouped_pmdarima, path=model_path)
    
        artifact_root = f"s3://{mock_s3_bucket}"
        artifact_path = "model"
        artifact_repo = S3ArtifactRepository(artifact_root)
        artifact_repo.log_artifacts(model_path, artifact_path=artifact_path)
    
        # NB: cloudpathlib would need to be used here to handle object store uri
        model_uri = os.path.join(artifact_root, artifact_path)
        reloaded_model = mlflow.diviner.load_model(model_uri=model_uri)
    
>       pd.testing.assert_frame_equal(grouped_pmdarima.predict(10), reloaded_model.predict(10))

artifact_path = 'model'
artifact_repo = <mlflow.store.artifact.s3_artifact_repo.S3ArtifactRepository object at 0x7fddf8a65d90>
artifact_root = 's3://mock-bucket'
grouped_pmdarima = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
mock_s3_bucket = 'mock-bucket'
model_path = PosixPath('/tmp/pytest-of-runner/pytest-0/test_diviner_load_from_remote_0/model')
model_uri  = 's3://mock-bucket/model'
reloaded_model = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddf8a65f10>

tests/diviner/test_diviner_model_export.py:232: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:405: in predict
    return self._run_predictions(prediction_config, exog=exog, **predict_kwargs)
        alpha      = 0.05
        exog       = None
        inverse_transform = True
        n_periods  = 10
        predict_col = 'yhat'
        predict_kwargs = {}
        prediction_config =    n_periods  alpha  return_conf_int  inverse_transform key2 key1 key0
0         10   0.05            False           ...         False               True    L    A    K
3         10   0.05            False               True    Z    X    B
        return_conf_int = False
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:335: in _run_predictions
    for idx, row in processing_data.iterrows()
        df         =    n_periods  alpha  return_conf_int  inverse_transform key2 key1 key0
0         10   0.05            False           ...         False               True    L    A    K
3         10   0.05            False               True    Z    X    B
        exog       = None
        n_periods_col = 'n_periods'
        predict_kwargs = {}
        processing_data =    n_periods  alpha  return_conf_int  ...  key1 key0 grouping_key
0         10   0.05            False  ...     P    T...se  ...     A    K    (L, A, K)
3         10   0.05            False  ...     X    B    (Z, X, B)

[4 rows x 8 columns]
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:335: in <listcomp>
    for idx, row in processing_data.iterrows()
        .0         = <generator object DataFrame.iterrows at 0x7fddf8aa8ed0>
        exog       = None
        idx        = 0
        n_periods_col = 'n_periods'
        predict_kwargs = {}
        row        = n_periods                   10
alpha                     0.05
return_conf_int          False
inverse_transform        ...  H
key1                         P
key0                         T
grouping_key         (H, P, T)
Name: 0, dtype: object
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/diviner/grouped_pmdarima.py:301: in _predict_single_group
    **predict_kwargs,
        alpha      = 0.05
        exog       = None
        group_key  = ('H', 'P', 'T')
        inverse_transform = True
        model      = AutoARIMA(maxiter=30, out_of_sample_size=60)
        n_periods_col = 'n_periods'
        periods    = 10
        predict_kwargs = {}
        return_conf_int = False
        row_entry  = n_periods                   10
alpha                     0.05
return_conf_int          False
inverse_transform        ...  H
key1                         P
key0                         T
grouping_key         (H, P, T)
Name: 0, dtype: object
        self       = <diviner.grouped_pmdarima.GroupedPmdarima object at 0x7fddff924a90>
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

args = ()
kwargs = {'X': None, 'alpha': 0.05, 'inverse_transform': True, 'n_periods': 10, ...}

>   out = (lambda *args, **kwargs: self.fn(obj, *args, **kwargs))
E   TypeError: predict() got an unexpected keyword argument 'inverse_transform'

args       = ()
kwargs     = {'X': None, 'alpha': 0.05, 'inverse_transform': True, 'n_periods': 10, ...}
obj        = AutoARIMA(maxiter=30, out_of_sample_size=60)
self       = <pmdarima.utils.metaestimators._IffHasDelegate object at 0x7fde0085e350>

/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/pmdarima/utils/metaestimators.py:[53](https://github.com/mlflow/mlflow/runs/7974149279?check_suite_focus=true#step:12:54): TypeError

[harupy]

Related to alkaline-ml/pmdarima#494, which removed **kwargs in AutoArima.predict.

[BenWilson2]

Fixed in #15