Adding TIME SERIES models (AR, MA, ARMA, ARIMA) (#830)

* Adding ARIMA model * Multiple support - AR, MA, ARMA - load_model - improving code quality * corrections - fi * Possibility to draw TS * correction: start parameter * corrections * corrections * corrections * Correcting plots + adding metrics * Supporting estimation of "ts" * supporting highchart * correction + plotly * Update tsa.py * Update tsa.py * Update tsa.py * Docstring for AR * Added Docstring for ARIMA * Docstring for ARMA * multiple corrections --------- Co-authored-by: umar <[email protected]>
vertica · Nov 6, 2023 · 12ba3a9 · 12ba3a9
1 parent 69056e1
commit 12ba3a9
Show file tree

Hide file tree

Showing 17 changed files with 2,761 additions and 22 deletions.
diff --git a/verticapy/_utils/_sql/_vertica_version.py b/verticapy/_utils/_sql/_vertica_version.py
@@ -22,6 +22,9 @@
 from verticapy.errors import VersionError
 
 MINIMUM_VERTICA_VERSION = {
+    "ARIMA": [23, 3, 0],
+    "AR": [11, 0, 0],
+    "ARMA": [12, 0, 3],
     "Balance": [8, 1, 1],
     "BernoulliNB": [8, 0, 0],
     "BisectingKMeans": [9, 3, 1],
@@ -47,6 +50,7 @@
     "LogisticRegression": [8, 0, 0],
     "KMeans": [8, 0, 0],
     "KPrototypes": [12, 0, 3],
+    "MA": [11, 0, 0],
     "MCA": [9, 1, 0],
     "MinMaxScaler": [8, 1, 0],
     "MultinomialNB": [8, 0, 0],

diff --git a/verticapy/machine_learning/vertica/__init__.py b/verticapy/machine_learning/vertica/__init__.py
@@ -68,3 +68,4 @@
     DummyTreeClassifier,
     DummyTreeRegressor,
 )
+from verticapy.machine_learning.vertica.tsa import ARIMA, ARMA, AR, MA
diff --git a/verticapy/machine_learning/vertica/base.py b/verticapy/machine_learning/vertica/base.py
@@ -783,6 +783,8 @@ def fit(
         str
             model's summary.
         """
+
+        # Initialization
         if self.overwrite_model:
             self.drop()
         else:
@@ -819,7 +821,7 @@ def fit(
                     ROW_NUMBER() OVER 
                     (ORDER BY {', '.join(X)}) 
                     AS {id_column_name}"""
-            tmp_view = False
+        tmp_view = False
         if isinstance(input_relation, vDataFrame) or (id_column):
             tmp_view = True
             if isinstance(input_relation, vDataFrame):
@@ -830,10 +832,9 @@ def fit(
                 relation = gen_tmp_name(
                     schema=schema_relation(self.model_name)[0], name="view"
                 )
-                drop(relation, method="view")
                 _executeSQL(
                     query=f"""
-                        CREATE VIEW {relation} AS 
+                        CREATE OR REPLACE VIEW {relation} AS 
                             SELECT 
                                 /*+LABEL('learn.VerticaModel.fit')*/ 
                                 *{id_column} 
@@ -849,6 +850,7 @@ def fit(
             self.test_relation = test_relation
         else:
             self.test_relation = self.input_relation
+        # Fitting
         if self._is_native:
             parameters = self._get_vertica_param_dict()
             if (
@@ -917,15 +919,6 @@ def _attributes(self) -> list:
     def __init__(self) -> None:
         """Must be overridden in the child class"""
         self.features_importance_trees_ = {}
-        return None
-        # self.input_relation = None
-        # self.test_relation = None
-        # self.X = None
-        # self.y = None
-        # self.parameters = {}
-        # self.classes_ = None
-        # for att in self._attributes:
-        #    setattr(self, att, None)
 
     def _compute_trees_arrays(
         self, tree: TableSample, X: list, return_probability: bool = False
@@ -1265,6 +1258,10 @@ def plot_tree(
 
 
 class BinaryClassifier(Supervised):
+    """
+    Base Class for Vertica Binary Classifier.
+    """
+
     # Properties.
 
     @property
@@ -1805,13 +1802,16 @@ def roc_curve(
 
 
 class MulticlassClassifier(Supervised):
+    """
+    Base Class for Vertica Multiclass Classifiers.
+    """
+
     # System & Special Methods.
 
     @abstractmethod
     def __init__(self, name: str, overwrite_model: bool = False) -> None:
         """Must be overridden in the child class"""
         super().__init__(name, overwrite_model)
-        # self.classes_ = None
 
     def _check_pos_label(self, pos_label: PythonScalar) -> PythonScalar:
         """
@@ -2645,6 +2645,10 @@ def roc_curve(
 
 
 class Regressor(Supervised):
+    """
+    Base Class for Vertica Regressors.
+    """
+
     # System & Special Methods.
 
     @abstractmethod
@@ -2845,7 +2849,7 @@ def predict(
         Returns
         -------
         vDataFrame
-                the input object.
+            the input object.
         """
         if hasattr(self, "_predict"):
             return self._predict(vdf=vdf, X=X, name=name, inplace=inplace)
@@ -2904,6 +2908,8 @@ def fit(
         str
             model's summary.
         """
+
+        # Initialization
         if self.overwrite_model:
             self.drop()
         else:
@@ -2938,10 +2944,9 @@ def fit(
             relation = gen_tmp_name(
                 schema=schema_relation(self.model_name)[0], name="view"
             )
-            drop(relation, method="view")
             _executeSQL(
                 query=f"""
-                    CREATE VIEW {relation} AS 
+                    CREATE OR REPLACE VIEW {relation} AS 
                         SELECT 
                             /*+LABEL('learn.VerticaModel.fit')*/ *
                             {id_column} 
@@ -2962,6 +2967,7 @@ def fit(
         parameters = self._get_vertica_param_dict()
         if "num_components" in parameters and not parameters["num_components"]:
             del parameters["num_components"]
+        # Fitting
         fun = self._vertica_fit_sql if self._model_type != "MCA" else "PCA"
         query = f"""
             SELECT 

diff --git a/verticapy/machine_learning/vertica/linear_model.py b/verticapy/machine_learning/vertica/linear_model.py
@@ -43,6 +43,10 @@
 
 
 class LinearModel:
+    """
+    Base Class for Vertica Linear Models.
+    """
+
     # Properties.
 
     @property
@@ -188,6 +192,10 @@ def plot(
 
 
 class LinearModelClassifier(LinearModel):
+    """
+    Base Class for Vertica Linear Models Classifiers.
+    """
+
     # Properties.
 
     @property

diff --git a/verticapy/machine_learning/vertica/model_management.py b/verticapy/machine_learning/vertica/model_management.py
@@ -38,11 +38,13 @@
     Lasso,
     LinearRegression,
     LogisticRegression,
+    PoissonRegressor,
     Ridge,
 )
 from verticapy.machine_learning.vertica.naive_bayes import NaiveBayes
 from verticapy.machine_learning.vertica.preprocessing import Scaler, OneHotEncoder
 from verticapy.machine_learning.vertica.svm import LinearSVC, LinearSVR
+from verticapy.machine_learning.vertica.tsa import ARIMA, AR, MA
 
 
 @save_verticapy_logs
@@ -139,6 +141,9 @@ def load_model(
         info = info[0]
     info = eval("[" + info + "]")
     lookup_table = {
+        "arima": ARIMA,
+        "autoregressor": AR,
+        "moving_average": MA,
         "rf_regressor": RandomForestRegressor,
         "rf_classifier": RandomForestClassifier,
         "iforest": IsolationForest,
@@ -149,6 +154,7 @@ def load_model(
         "svm_regressor": LinearSVR,
         "svm_classifier": LinearSVC,
         "linear_reg": LinearRegression,
+        "poisson_reg": PoissonRegressor,
         "kmeans": KMeans,
         "kprototypes": KPrototypes,
         "bisecting_kmeans": BisectingKMeans,
@@ -158,9 +164,20 @@ def load_model(
     }
     model = lookup_table[model_type](name)
     if model_type != "svd":
+        # Variables used in the CALL STRING
         true, false = True, False
         squarederror = "squarederror"
         crossentropy = "crossentropy"
+        ols = "ols"
+        hr = "hr"
+        linear_interpolation = "linear_interpolation"
+        zero = "zero"
+        error = "error"
+        drop = "drop"
+        if "method=yule-walker," in parameters:
+            parameters = parameters.replace(
+                "method=yule-walker,", "method='yule-walker',"
+            )
         if " lambda=" in parameters:
             parameters = parameters.replace(" lambda=", " C=")
         try:
@@ -186,6 +203,15 @@ def load_model(
         model.y = info[2]
         model.X = eval("[" + info[3] + "]")
         model.test_relation = test_relation if (test_relation) else model.input_relation
+    elif model._model_category == "TIMESERIES":
+        model.y = info[2]
+        model.ts = info[3]
+        model.test_relation = test_relation if (test_relation) else model.input_relation
+        if model._model_type == "ARIMA":
+            p = int(model.get_vertica_attributes("p")["p"][0])
+            d = int(model.get_vertica_attributes("d")["d"][0])
+            q = int(model.get_vertica_attributes("q")["q"][0])
+            model.set_params({"order": (p, d, q)})
     else:
         model.X = eval("[" + info[2] + "]")
     model._compute_attributes()