Merge pull request #133 from moj-analytical-services/vif2

tighten tests to check exact equivalence

tests/test_diagnostics.py

@@ -2,29 +2,32 @@
 import pandas as pd
 from splink.gammas import add_gammas
 from splink.diagnostics import vif_gammas
+import pytest
 
-
+# For further info about this test see https://github.com/moj-analytical-services/splink/issues/132
 def test_vif_gammas(spark, gamma_settings_4, params_4, sqlite_con_4):
 
     """
-    use fixture data that is independent to ensure vif_gammas works with typical data 
+    use fixture data that is independent to ensure vif_gammas works with typical data
     (in this case with low association / correlation between columns)
-    
+
     """
-    
+
     dfpd = pd.read_sql("select * from df", sqlite_con_4)
     df_gammas = spark.createDataFrame(dfpd)
+    df_gammas = df_gammas.filter("true_match = 0")
 
-    res = vif_gammas(df_gammas, spark=spark, sampleratio=0.05).toPandas()
+    res = vif_gammas(df_gammas, spark=spark, sampleratio=1.0).toPandas()
 
-    assert ((res.vif.values < 10) & (res.vif.values > 1)).all() == True
+    for val in res.vif.values:
+        assert pytest.approx(val) == 1.0
 
 
 def test_vif_no_gammas(spark, gamma_settings_4, params_4, sqlite_con_4):
-    
+
     """
     test that when input doesnt have gamma columns function exits gracefully
-    
+
     """
 
     dfpd = pd.read_sql("select * from df", sqlite_con_4)
@@ -38,7 +41,7 @@ def test_vif_no_gammas(spark, gamma_settings_4, params_4, sqlite_con_4):
 
 
 def test_vif_fully_correlated(spark, gamma_settings_4, params_4, sqlite_con_4):
-    
+
     """
     test that when input data has some columns that are fully correlated / associated function deals with it gracefully
     """