required changes were made

statsmodels/tsa/automatic/sarimax.py

@@ -4,10 +4,41 @@
 import statsmodels.api as sm
 
 
-def auto_order(
-                endog, criteria='aic', d=0, max_order=(3, 3),
+def auto_order(endog, criteria='aic', d=0, max_order=(3, 3),
                 stepwise=False, **spec):
-    """Auto order function for SARIMAX models."""
+    """Perform automatic calculation of the parameters used in SARIMAX models.
+
+    This uses two different approaches to clacluate the parameters:
+    1. Uses a brute force approach to traverse through all the possible
+        parameters of the model.
+    2. Uses the stepwise algorithm specified in Hyndman's Paper.
+
+    Parameters
+    ----------
+    endog : list
+        input contains the time series data over a period of time.
+    criteria : str
+        specifies which information criteria to use for model evaluation.
+    d : int
+        differencing factor.
+    max_order : list
+        maximum possible values of the parameters p and q.
+    stepwise : Boolean
+        specifies whether to use the stepwise algorithm or not.
+    **spec : list of kwargs
+
+    Returns
+    -------
+    p : the autoregressive parameter for the SARIMAX model.
+    q : the moving average parameter for the SARIMAX model.
+
+    Notes
+    -----
+    Status : Work In Progress.
+    Citation : Hyndman, Rob J., and Yeasmin Khandakar.
+         "Automatic Time Series Forecasting: The forecast Package for R."
+
+    """
     if not stepwise:
         aic_matrix = np.zeros((max_order[0], max_order[1]))
         for p in range(max_order[0]):
@@ -41,54 +72,57 @@ def auto_order(
         min_aic = aic_vals.min()
         model = int(np.where(aic_vals == min_aic)[0])
         p, q = order_init[model][0], order_init[model][1]
+        new_p, new_q = p, q
         # only p varies by +1
-        if(p+1 <= max_order[0]):
-            mod = sm.tsa.statespace.SARIMAX(endog, order=(p+1, d, q), **spec)
+        if(p + 1 <= max_order[0]):
+            mod = sm.tsa.statespace.SARIMAX(endog, order=(p + 1, d, q), **spec)
             res = mod.fit(disp=False)
             # print(res.aic)
             if (res.aic < min_aic):
-                p += 1
+                new_p = p + 1
                 min_aic = res.aic
         # only p varies by -1
-        if(p-1 >= 0):
-            mod = sm.tsa.statespace.SARIMAX(endog, order=(p-1, d, q), **spec)
+        if(p - 1 >= 0):
+            mod = sm.tsa.statespace.SARIMAX(endog, order=(p - 1, d, q), **spec)
             res = mod.fit(disp=False)
             # print(res.aic)
             if (res.aic < min_aic):
-                p -= 1
+                new_p = p - 1
                 min_aic = res.aic
         # only q varies by +1
-        if(q+1 <= max_order[1]):
-            mod = sm.tsa.statespace.SARIMAX(endog, order=(p, d, q+1), **spec)
+        if(q + 1 <= max_order[1]):
+            mod = sm.tsa.statespace.SARIMAX(endog, order=(p, d, q + 1), **spec)
             res = mod.fit(disp=False)
             # print(res.aic)
             if (res.aic < min_aic):
-                q += 1
+                new_q = q + 1
                 min_aic = res.aic
         # only q varies by -1
-        if(q-1 >= 0):
-            mod = sm.tsa.statespace.SARIMAX(endog, order=(p, d, q-1), **spec)
+        if(q - 1 >= 0):
+            mod = sm.tsa.statespace.SARIMAX(endog, order=(p, d, q - 1), **spec)
             res = mod.fit(disp=False)
             # print(res.aic)
             if (res.aic < min_aic):
-                q -= 1
+                new_q = q - 1
                 min_aic = res.aic
         # both p and q vary by +1
-        if(p+1 <= max_order[0]) and (q+1 <= max_order[1]):
-            mod = sm.tsa.statespace.SARIMAX(endog, order=(p+1, d, q+1), **spec)
+        if(p + 1 <= max_order[0]) and (q + 1 <= max_order[1]):
+            mod = sm.tsa.statespace.SARIMAX(endog,
+                                            order=(p + 1, d, q + 1), **spec)
             res = mod.fit(disp=False)
             # print(res.aic)
             if (res.aic < min_aic):
-                p += 1
-                q += 1
+                new_p = p + 1
+                new_q = q + 1
                 min_aic = res.aic
         # both p and q vary by -1
-        if(q-1 >= 0) and (q-1 >= 0):
-            mod = sm.tsa.statespace.SARIMAX(endog, order=(p-1, d, q-1), **spec)
+        if(q - 1 >= 0) and (q - 1 >= 0):
+            mod = sm.tsa.statespace.SARIMAX(endog,
+                                            order=(p - 1, d, q - 1), **spec)
             res = mod.fit(disp=False)
             # print(res.aic)
             if (res.aic < min_aic):
-                p -= 1
-                q -= 1
+                new_p = p - 1
+                new_q = q - 1
                 min_aic = res.aic
-        return p, q
+        return new_p, new_q

statsmodels/tsa/automatic/tests/test_sarimax.py

@@ -6,6 +6,7 @@
 import statsmodels.api as sm
 from statsmodels.tsa.automatic import sarimax
 from statsmodels import datasets
+from numpy.testing import assert_equal
 
 
 current_path = os.path.dirname(os.path.abspath(__file__))
@@ -19,8 +20,8 @@ def test_non_stepwise():
 #    p, q = sarimax.auto_order(macrodata.infl, d=0, enforce_stationarity=False)
     desired_p = 2
     desired_q = 2
-    assert p == desired_p
-    assert q == desired_q
+    assert_equal(p, desired_p)
+    assert_equal(q, desired_q)
 
 
 def test_stepwise():
@@ -30,5 +31,5 @@ def test_stepwise():
     p, q = sarimax.auto_order(macrodata['infl'], stepwise=True)
     desired_p = 3
     desired_q = 3
-    assert p == desired_p
-    assert q == desired_q
+    assert_equal(p, desired_p)
+    assert_equal(q, desired_q)