added projection standardizer

alphamind/data/standardize.py

@@ -12,6 +12,7 @@ from alphamind.utilities import aggregate
 from alphamind.utilities import array_index
 from alphamind.utilities import simple_mean
 from alphamind.utilities import simple_std
+from alphamind.utilities import simple_sqrsum
 
 
 def standardize(x: np.ndarray, groups: np.ndarray=None, ddof=1) -> np.ndarray:
@@ -26,6 +27,15 @@ def standardize(x: np.ndarray, groups: np.ndarray=None, ddof=1) -> np.ndarray:
         return (x - simple_mean(x, axis=0)) / simple_std(x, axis=0, ddof=ddof)
 
 
+def projection(x: np.ndarray, groups: np.ndarray=None, axis=1) -> np.ndarray:
+    if groups is not None and axis == 0:
+        groups = group_mapping(groups)
+        projected = transform(groups, x, 'project')
+        return projected
+    else:
+        return x / simple_sqrsum(x, axis=axis).reshape((-1, 1))
+
+
 class Standardizer(object):
 
     def __init__(self, ddof: int=1):

alphamind/tests/data/test_standardize.py

@@ -10,6 +10,7 @@ import numpy as np
 import pandas as pd
 from scipy.stats import zscore
 from alphamind.data.standardize import standardize
+from alphamind.data.standardize import projection
 from alphamind.data.standardize import Standardizer
 from alphamind.data.standardize import GroupedStandardizer
 
@@ -26,6 +27,20 @@ class TestStandardize(unittest.TestCase):
 
         np.testing.assert_array_almost_equal(calc_zscore, exp_zscore)
 
+    def test_projection(self):
+        calc_projected = projection(self.x)
+        exp_projected = self.x / np.sqrt(np.sum(np.square(self.x), axis=1).reshape((-1, 1)))
+
+        np.testing.assert_array_almost_equal(calc_projected, exp_projected)
+
+    def test_projection_with_groups(self):
+        calc_projected = projection(self.x, self.groups, axis=0)
+        exp_projected = pd.DataFrame(self.x).groupby(
+            self.groups
+        ).transform(lambda s: s / np.sqrt(np.square(s).sum(axis=0)))
+
+        np.testing.assert_array_almost_equal(calc_projected, exp_projected)
+
     def test_standardize_with_group(self):
         calc_zscore = standardize(self.x, self.groups)
         exp_zscore = pd.DataFrame(self.x).\

alphamind/utilities.py

@@ -93,6 +93,27 @@ def simple_abssum(x, axis=0):
     return res
 
 
+def simple_sqrsum(x, axis=0):
+    length, width = x.shape
+
+    if axis == 0:
+        res = np.zeros(width)
+        for i in range(length):
+            for j in range(width):
+                res[j] += x[i, j] * x[i, j]
+
+    elif axis == 1:
+        res = np.zeros(length)
+        for i in range(length):
+            for j in range(width):
+                res[i] += x[i, j] * x[i, j]
+    else:
+        raise ValueError("axis value is not supported")
+
+    res = np.sqrt(res)
+    return res
+
+
 @nb.njit(nogil=True, cache=True)
 def simple_mean(x, axis=0):
     length, width = x.shape
@@ -152,6 +173,19 @@ def agg_sum(groups, x):
     return res
 
 
+def agg_sqrsum(groups, x):
+    max_g = groups.max()
+    length, width = x.shape
+    res = np.zeros((max_g + 1, width), dtype=np.float64)
+
+    for i in range(length):
+        for j in range(width):
+            res[groups[i], j] += x[i, j] * x[i, j]
+
+    res = np.sqrt(res)
+    return res
+
+
 @nb.njit(nogil=True, cache=True)
 def agg_abssum(groups, x):
     max_g = groups.max()
@@ -256,10 +290,12 @@ def transform(groups: np.ndarray,
         value_data = agg_sum(groups, x)
     elif func == 'abssum' or func == 'scale':
         value_data = agg_abssum(groups, x)
+    elif func == 'sqrsum' or func == 'project':
+        value_data = agg_sqrsum(groups, x)
     else:
         raise ValueError('({0}) is not recognized as valid functor'.format(func))
 
-    if func == 'scale':
+    if func == 'scale' or func == 'project':
         return scale_value(groups, value_data, x, scale)
     else:
         return copy_value(groups, value_data)
@@ -274,6 +310,8 @@ def aggregate(groups, x, func, ddof=1):
         value_data = agg_sum(groups, x)
     elif func == 'abssum' or func == 'scale':
         value_data = agg_abssum(groups, x)
+    elif func == 'sqrsum' or func == 'project':
+        value_data = agg_sqrsum(groups, x)
     else:
         raise ValueError('({0}) is not recognized as valid functor'.format(func))