small enhancements

alphamind/data/neutralize.py

@@ -38,10 +38,7 @@ def neutralize(x: np.ndarray, y: np.ndarray, groups: np.ndarray=None, output_exp
         groups_ids = groupby(groups)
 
         for curr_idx in groups_ids.values():
-            curr_x = x[curr_idx]
-            curr_y = y[curr_idx]
-            b = ls_fit(curr_x, curr_y)
-            res[curr_idx] = ls_res(curr_x, curr_y, b)
+            curr_x, b = _sub_step(x, y, curr_idx, res)
             if output_exposure:
                 for i in range(exposure.shape[2]):
                     exposure[curr_idx, :, i] = b[:, i]
@@ -69,6 +66,15 @@ def neutralize(x: np.ndarray, y: np.ndarray, groups: np.ndarray=None, output_exp
         return res
 
 
+@nb.njit(nogil=True, cache=True)
+def _sub_step(x, y, curr_idx, res):
+    curr_x = x[curr_idx]
+    curr_y = y[curr_idx]
+    b = ls_fit(curr_x, curr_y)
+    res[curr_idx] = ls_res(curr_x, curr_y, b)
+    return curr_x, b
+
+
 @nb.njit(nogil=True, cache=True)
 def ls_fit(x: np.ndarray, y: np.ndarray) -> np.ndarray:
     x_bar = x.T

alphamind/model/linearmodel.py

@@ -7,6 +7,7 @@ Created on 2017-5-10
 
 from typing import Union
 import numpy as np
+import numba as nb
 from alphamind.cyimpl import groupby
 from alphamind.data.neutralize import ls_fit
 
@@ -20,7 +21,7 @@ class LinearModel(object):
         self.model_parameter = _train(x, y, groups)
 
     def predict(self, x, groups=None):
-        if groups is not None and isinstance(self.model_parameter, dict):
+        if groups is not None and self.model_parameter.ndim == 2:
             names = np.unique(groups)
             return multiple_prediction(names, self.model_parameter, x, groups)
         elif self.model_parameter is None:
@@ -31,6 +32,7 @@ class LinearModel(object):
             raise ValueError("grouped x value can't be used for vanilla linear model")
 
 
+@nb.njit(nogil=True, cache=True)
 def multiple_prediction(names, model_parames, x, groups):
     pred_v = np.zeros(x.shape[0])
     for name in names:
@@ -40,22 +42,28 @@ def multiple_prediction(names, model_parames, x, groups):
     return pred_v
 
 
-def _train(x: np.ndarray, y: np.ndarray, groups: np.ndarray=None) -> Union[np.ndarray, dict]:
+def _train(x: np.ndarray, y: np.ndarray, groups: np.ndarray=None) -> np.ndarray:
     if groups is None:
         return ls_fit(x, y)
     else:
         groups_ids = groupby(groups)
-        res_beta = {}
+        res_beta = np.zeros((max(groups_ids.keys())+1, x.shape[1]))
 
         for k, curr_idx in groups_ids.items():
-            curr_x = x[curr_idx]
-            curr_y = y[curr_idx]
-            res_beta[k] = ls_fit(curr_x, curr_y)
+            _train_sub_group(x, y, k, curr_idx, res_beta)
 
         return res_beta
 
 
+@nb.njit(nogil=True, cache=True)
+def _train_sub_group(x, y, k, curr_idx, res):
+    curr_x = x[curr_idx]
+    curr_y = y[curr_idx]
+    res[k] = ls_fit(curr_x, curr_y)
+
+
 if __name__ == '__main__':
+    import datetime as dt
     x = np.random.randn(3000, 10)
     y = np.random.randn(3000)
     groups = np.random.randint(30, size=3000)
@@ -65,5 +73,12 @@ if __name__ == '__main__':
 
     model = LinearModel()
 
-    model.calibrate(x, y, groups)
-    model.predict(to_x, to_groups)
\ No newline at end of file
+    start = dt.datetime.now()
+    for i in range(5000):
+        model.calibrate(x, y, groups)
+    print(dt.datetime.now() - start)
+
+    start = dt.datetime.now()
+    for i in range(50000):
+        model.predict(to_x, to_groups)
+    print(dt.datetime.now() - start)
\ No newline at end of file