Add some support functions for clustering using PySAL

pg/.gitignore

@@ -1,4 +1,3 @@
-dist/
 test/regression.diffs
 test/regression.out
 test/results

pg/dist/crankshaft--0.0.1.sql

@@ -0,0 +1,5 @@
+CREATE OR REPLACE FUNCTION cdb_poc_xyz()
+RETURNS Text AS $$
+    from crankshaft.poc import xyz
+    return xyz()
+$$ LANGUAGE plpythonu;

python/crankshaft/crankshaft/__init__.py

@@ -1 +1,2 @@
 import poc
+import clustering

python/crankshaft/crankshaft/clustering/__init__.py

@@ -0,0 +1 @@
+from moran import *

python/crankshaft/crankshaft/clustering/moran.py

@@ -0,0 +1,167 @@
+"""
+Moran's I geostatistics (global clustering & outliers presence)
+"""
+
+# TODO: Fill in local neighbors which have null/NoneType values with the
+#       average of the their neighborhood
+
+import numpy as np
+import pysal as ps
+
+def map_quads(coord):
+    """
+        Map a quadrant number to Moran's I designation
+        HH=1, LH=2, LL=3, HL=4
+        Input:
+        :param coord (int): quadrant of a specific measurement
+    """
+    if coord == 1:
+        return 'HH'
+    elif coord == 2:
+        return 'LH'
+    elif coord == 3:
+        return 'LL'
+    elif coord == 4:
+        return 'HL'
+    else:
+        return None
+
+def query_attr_select(params):
+    """
+        Create portion of SELECT statement for attributes inolved in query.
+        :param params: dict of information used in query (column names,
+                       table name, etc.)
+    """
+
+    attrs = [k for k in params
+             if k not in ('id_col', 'geom_col', 'table', 'num_ngbrs')]
+
+    template = "i.\"{%(col)s}\"::numeric As attr%(alias_num)s, "
+
+    attr_string = ""
+
+    for idx, val in enumerate(sorted(attrs)):
+        attr_string += template % {"col": val, "alias_num": idx + 1}
+
+    return attr_string
+
+def query_attr_where(params):
+    """
+        Create portion of WHERE clauses for weeding out NULL-valued geometries
+    """
+    attrs = sorted([k for k in params
+                    if k not in ('id_col', 'geom_col', 'table', 'num_ngbrs')])
+
+    attr_string = []
+
+    for attr in attrs:
+        attr_string.append("idx_replace.\"{%s}\" IS NOT NULL" % attr)
+
+    if len(attrs) == 2:
+        attr_string.append("idx_replace.\"{%s}\" <> 0" % attrs[1])
+
+    out = " AND ".join(attr_string)
+
+    return out
+
+def knn(params):
+    """SQL query for k-nearest neighbors.
+        :param vars: dict of values to fill template
+    """
+
+    attr_select = query_attr_select(params)
+    attr_where = query_attr_where(params)
+
+    replacements = {"attr_select": attr_select,
+                    "attr_where_i": attr_where.replace("idx_replace", "i"),
+                    "attr_where_j": attr_where.replace("idx_replace", "j")}
+
+    query = "SELECT " \
+                "i.\"{id_col}\" As id, " \
+                "%(attr_select)s" \
+                "(SELECT ARRAY(SELECT j.\"{id_col}\" " \
+                              "FROM \"{table}\" As j " \
+                              "WHERE %(attr_where_j)s " \
+                              "ORDER BY j.\"{geom_col}\" <-> i.\"{geom_col}\" ASC " \
+                              "LIMIT {num_ngbrs} OFFSET 1 ) " \
+                ") As neighbors " \
+            "FROM \"{table}\" As i " \
+            "WHERE " \
+                "%(attr_where_i)s " \
+            "ORDER BY i.\"{id_col}\" ASC;" % replacements
+
+    return query.format(**params)
+
+## SQL query for finding queens neighbors (all contiguous polygons)
+def queen(params):
+    """SQL query for queen neighbors.
+        :param params: dict of information to fill query
+    """
+    attr_select = query_attr_select(params)
+    attr_where = query_attr_where(params)
+
+    replacements = {"attr_select": attr_select,
+                    "attr_where_i": attr_where.replace("idx_replace", "i"),
+                    "attr_where_j": attr_where.replace("idx_replace", "j")}
+
+    query = "SELECT " \
+                "i.\"{id_col}\" As id, " \
+                "%(attr_select)s" \
+                "(SELECT ARRAY(SELECT j.\"{id_col}\" " \
+                 "FROM \"{table}\" As j " \
+                 "WHERE ST_Touches(i.\"{geom_col}\", j.\"{geom_col}\") AND " \
+                 "%(attr_where_j)s)" \
+                ") As neighbors " \
+            "FROM \"{table}\" As i " \
+            "WHERE " \
+                "%(attr_where_i)s " \
+            "ORDER BY i.\"{id_col}\" ASC;" % replacements
+
+    return query.format(**params)
+
+## to add more weight methods open a ticket or pull request
+
+def get_query(w_type, query_vals):
+    """Return requested query.
+        :param w_type: type of neighbors to calculate (knn or queen)
+        :param query_vals: values used to construct the query
+    """
+
+    if w_type == 'knn':
+        return knn(query_vals)
+    else:
+        return queen(query_vals)
+
+def get_attributes(query_res, attr_num):
+    """
+        :param query_res: query results with attributes and neighbors
+        :param attr_num: attribute number (1, 2, ...)
+    """
+    return np.array([x['attr' + str(attr_num)] for x in query_res], dtype=np.float)
+
+## Build weight object
+def get_weight(query_res, w_type='queen', num_ngbrs=5):
+    """
+        Construct PySAL weight from return value of query
+        :param query_res: query results with attributes and neighbors
+    """
+    if w_type == 'knn':
+        row_normed_weights = [1.0 / float(num_ngbrs)] * num_ngbrs
+        weights = {x['id']: row_normed_weights for x in query_res}
+    elif w_type == 'queen':
+        weights = {x['id']: [1.0 / len(x['neighbors'])] * len(x['neighbors'])
+                            if len(x['neighbors']) > 0
+                            else [] for x in query_res}
+
+    neighbors = {x['id']: x['neighbors'] for x in query_res}
+
+    return ps.W(neighbors, weights)
+
+def quad_position(quads):
+    """
+        Produce Moran's I classification based of n
+    """
+
+    lisa_sig = np.array([map_quads(q) for q in quads])
+
+    return lisa_sig

python/crankshaft/setup.py

@@ -7,7 +7,7 @@ https://github.com/CartoDB/crankshaft
 
 from setuptools import setup, find_packages
 
-REQUIRES = [] # ['pysal','numpy']
+REQUIRES = ['pysal','numpy']
 
 setup(
     name='crankshaft',
@@ -40,9 +40,9 @@ setup(
         'test': ['unittest', 'nose', 'mock'],
     },
 
-    # install_requires=REQUIRES,
+    install_requires=REQUIRES,
 
-     # requires=REQUIRES,
+    requires=REQUIRES,
 
-     test_suite='test'
+    test_suite='test'
 )

python/crankshaft/test/helper.py

@@ -0,0 +1,7 @@
+import unittest
+
+from mock_plpy import MockPlPy
+plpy = MockPlPy()
+
+import sys
+sys.modules['plpy'] = plpy

python/crankshaft/test/test_clustering_moran.py

@@ -0,0 +1,118 @@
+import unittest
+import numpy as np
+
+import unittest
+
+
+# from mock_plpy import MockPlPy
+# plpy = MockPlPy()
+#
+# import sys
+# sys.modules['plpy'] = plpy
+from helper import plpy
+
+# import crankshaft.clustering as cc
+import crankshaft.clustering as cc
+
+
+class MoranTest(unittest.TestCase):
+    """Testing class for Moran's I functions."""
+
+    def setUp(self):
+        self.params = {"id_col": "cartodb_id",
+                       "attr1": "andy",
+                       "attr2": "jay_z",
+                       "table": "a_list",
+                       "geom_col": "the_geom",
+                       "num_ngbrs": 321}
+
+    def test_map_quads(self):
+        """Test map_quads."""
+        self.assertEqual(cc.map_quads(1), 'HH')
+        self.assertEqual(cc.map_quads(2), 'LH')
+        self.assertEqual(cc.map_quads(3), 'LL')
+        self.assertEqual(cc.map_quads(4), 'HL')
+        self.assertEqual(cc.map_quads(33), None)
+        self.assertEqual(cc.map_quads('andy'), None)
+
+    def test_query_attr_select(self):
+        """Test query_attr_select."""
+
+        ans = "i.\"{attr1}\"::numeric As attr1, " \
+              "i.\"{attr2}\"::numeric As attr2, "
+
+        self.assertEqual(cc.query_attr_select(self.params), ans)
+
+    def test_query_attr_where(self):
+        """Test query_attr_where."""
+
+        ans = "idx_replace.\"{attr1}\" IS NOT NULL AND "\
+              "idx_replace.\"{attr2}\" IS NOT NULL AND "\
+              "idx_replace.\"{attr2}\" <> 0"
+
+        self.assertEqual(cc.query_attr_where(self.params), ans)
+
+    def test_knn(self):
+        """Test knn function."""
+
+        ans = "SELECT i.\"cartodb_id\" As id, i.\"andy\"::numeric As attr1, " \
+              "i.\"jay_z\"::numeric As attr2, (SELECT ARRAY(SELECT j.\"cartodb_id\" " \
+              "FROM \"a_list\" As j WHERE j.\"andy\" IS NOT NULL AND " \
+              "j.\"jay_z\" IS NOT NULL AND j.\"jay_z\" <> 0 ORDER BY " \
+              "j.\"the_geom\" <-> i.\"the_geom\" ASC LIMIT 321 OFFSET 1 ) ) " \
+              "As neighbors FROM \"a_list\" As i WHERE i.\"andy\" IS NOT " \
+              "NULL AND i.\"jay_z\" IS NOT NULL AND i.\"jay_z\" <> 0 ORDER " \
+              "BY i.\"cartodb_id\" ASC;"
+
+        self.assertEqual(cc.knn(self.params), ans)
+
+    def test_queen(self):
+        """Test queen neighbors function."""
+
+        ans = "SELECT i.\"cartodb_id\" As id, i.\"andy\"::numeric As attr1, " \
+              "i.\"jay_z\"::numeric As attr2, (SELECT ARRAY(SELECT " \
+              "j.\"cartodb_id\" FROM \"a_list\" As j WHERE ST_Touches(" \
+              "i.\"the_geom\", j.\"the_geom\") AND j.\"andy\" IS NOT NULL " \
+              "AND j.\"jay_z\" IS NOT NULL AND j.\"jay_z\" <> 0)) As " \
+              "neighbors FROM \"a_list\" As i WHERE i.\"andy\" IS NOT NULL " \
+              "AND i.\"jay_z\" IS NOT NULL AND i.\"jay_z\" <> 0 ORDER BY " \
+              "i.\"cartodb_id\" ASC;"
+
+        self.assertEqual(cc.queen(self.params), ans)
+
+    def test_get_query(self):
+        """Test get_query."""
+
+        ans = "SELECT i.\"cartodb_id\" As id, i.\"andy\"::numeric As attr1, " \
+              "i.\"jay_z\"::numeric As attr2, (SELECT ARRAY(SELECT " \
+              "j.\"cartodb_id\" FROM \"a_list\" As j WHERE j.\"andy\" IS " \
+              "NOT NULL AND j.\"jay_z\" IS NOT NULL AND j.\"jay_z\" <> 0 " \
+              "ORDER BY j.\"the_geom\" <-> i.\"the_geom\" ASC LIMIT 321 " \
+              "OFFSET 1 ) ) As neighbors FROM \"a_list\" As i WHERE " \
+              "i.\"andy\" IS NOT NULL AND i.\"jay_z\" IS NOT NULL AND " \
+              "i.\"jay_z\" <> 0 ORDER BY i.\"cartodb_id\" ASC;"
+
+        self.assertEqual(cc.get_query('knn', self.params), ans)
+
+    def test_get_attributes(self):
+        """Test get_attributes."""
+
+        ## need to add tests
+
+        self.assertEqual(True, True)
+
+    def test_get_weight(self):
+        """Test get_weight."""
+
+        self.assertEqual(True, True)
+
+
+    def test_quad_position(self):
+        """Test lisa_sig_vals."""
+
+        quads = np.array([1, 2, 3, 4], np.int)
+
+        ans = np.array(['HH', 'LH', 'LL', 'HL'])
+        test_ans = cc.quad_position(quads)
+
+        self.assertTrue((test_ans == ans).all())

python/crankshaft/test/test_poc.py

@@ -3,17 +3,21 @@
 
 import unittest
 
-from mock_plpy import MockPlPy
-plpy = MockPlPy()
-
-import sys
-sys.modules['plpy'] = plpy
+# from mock_plpy import MockPlPy
+# plpy = MockPlPy()
+#
+# import sys
+# sys.modules['plpy'] = plpy
 
+from helper import plpy
 import crankshaft
 
 class TestPoc(unittest.TestCase):
-    def test_should_have_xyz(self):
+
+    def setUp(self):
         plpy._reset()
+
+    def test_should_have_xyz(self):
         plpy._define_result('select\s+\*\s+from\s+table', [{'x': 111}])
         assert crankshaft.poc.xyz() == 111
         assert plpy.notices[0] == 'XYZ...'