def testDataframeRelativeToJackknife(self): df = pd.DataFrame({ "X": range(11), "Y": np.concatenate((np.zeros(6), np.ones(5))), "Z": np.concatenate((np.zeros(3), np.ones(8))) }) metric = metrics.Distribution("X", ["Z"]) se_method = standard_errors.Jackknife() output = core.Analyze(df).relative_to( comparisons.AbsoluteDifference( "Y", 0)).with_standard_errors(se_method).calculate(metric).run() correct = pd.DataFrame( np.array([[-0.2, 0.18100283490], [0.2, 0.18100283490]]), columns=[ "X Distribution Absolute Difference", "X Distribution Absolute Difference Jackknife SE" ], index=pd.MultiIndex(levels=[[1.], [0., 1.]], labels=[[0, 0], [0, 1]], names=["Y", "Z"])) self.assertTrue( all(output.index == correct.index) and all(output.columns == correct.columns) and np.all(abs(output.values - correct.values) < 1e-10))
def testShuffledDataframeRelativeToJackknife(self): # Same as test above, but also testing that reordering the data doesn't # change results, up to order. df = pd.DataFrame({ "X": range(11), "Y": np.concatenate((np.zeros(6), np.ones(5))), "Z": np.concatenate((np.zeros(3), np.ones(8))) }) metric = metrics.Distribution("X", ["Z"]) se_method = standard_errors.Jackknife() output = core.Analyze(df.iloc[np.random.permutation(11)]).relative_to( comparisons.AbsoluteDifference( "Y", 0)).with_standard_errors(se_method).calculate(metric).run() output = (output.reset_index().sort_values(by=["Y", "Z"]).set_index( ["Y", "Z"])) correct = pd.DataFrame( np.array([[-0.2, 0.18100283490], [0.2, 0.18100283490]]), columns=[ "X Distribution Absolute Difference", "X Distribution Absolute Difference Jackknife SE" ], index=pd.MultiIndex(levels=[[1.], [0., 1.]], labels=[[0, 0], [0, 1]], names=["Y", "Z"])) correct = (correct.reset_index().sort_values(by=["Y", "Z"]).set_index( ["Y", "Z"])) self.assertTrue( all(output.index == correct.index) and all(output.columns == correct.columns) and np.all(abs(output.values - correct.values) < 1e-10))
def testRelativeToSplitJackknife(self): data = pd.DataFrame({ "X": [1, 2, 3, 4, 5, 6, 7, 8, 9, 0, 1, 2, 3, 4, 5, 6, 7, 8], "Y": [1, 1, 1, 2, 2, 2, 3, 3, 3, 1, 1, 1, 2, 2, 2, 3, 3, 3], "Z": [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1] }) metric = metrics.Sum("X") comparison = comparisons.AbsoluteDifference("Z", 0) se_method = standard_errors.Jackknife() output = core.Analyze(data).split_by("Y").relative_to( comparison).with_standard_errors(se_method).calculate( metric).run() rowindex = pd.MultiIndex(levels=[[1, 2, 3], [1]], labels=[[0, 1, 2], [0, 0, 0]], names=["Y", "Z"]) correct = pd.DataFrame( np.array([[-3.0, np.sqrt(5 * np.var([0, -1, -2, -3, -4, -5]))], [-3.0, np.sqrt(5 * np.var([3, 2, 1, -8, -7, -6]))], [-3.0, np.sqrt(5 * np.var([6, 5, 4, -11, -10, -9]))]]), columns=("sum(X) Absolute Difference", "sum(X) Absolute Difference Jackknife SE"), index=rowindex) self.assertTrue(output.equals(correct))
def testNinetyFiveCIsWithComparison(self): data = pd.DataFrame({ "X": range(11), "Y": [0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1] }) metric = metrics.Sum("X") comparison = comparisons.AbsoluteDifference("Y", 0) se_method = standard_errors.Jackknife(confidence=0.95) output = core.Analyze(data).with_standard_errors( se_method).relative_to(comparison).calculate(metric).run() multiplier = scipy.stats.t.ppf(0.975, 10) correct_mean = 25 correct_buckets = [ 15., 16., 17., 18., 19., 25., 26., 27., 28., 29., 30. ] m = sum(correct_buckets) / len(correct_buckets) r = sum([(b - m)**2 for b in correct_buckets]) correct_sd = np.sqrt(r * (len(correct_buckets) - 1) / len(correct_buckets)) correct_lower = correct_mean - multiplier * correct_sd correct_upper = correct_mean + multiplier * correct_sd rowindex = pd.Index([1], name="Y") correct = pd.DataFrame( { "sum(X) Absolute Difference": correct_mean, "sum(X) Absolute Difference Jackknife CI-lower": correct_lower, "sum(X) Absolute Difference Jackknife CI-upper": correct_upper }, index=rowindex) self.assertTrue(output.equals(correct))
def testComparisonBaslineGivesError(self): data = pd.DataFrame({"X": [1, 2, 3, 4, 5], "Y": [1, 1, 1, 1, 1]}) metric = metrics.Sum("X") comparison = comparisons.AbsoluteDifference("Y", 0) with self.assertRaises(ValueError): core.Analyze(data).relative_to(comparison).calculate(metric).run()
def testAbsoluteDifference(self): data = pd.DataFrame({ "X": [1, 3, 2, 3, 1, 4], "Condition": [0, 0, 0, 1, 1, 1] }) weights = np.ones(6) comparison = comparisons.AbsoluteDifference("Condition", 0) comparison.precalculate_factors(data) metric = metrics.Sum("X") output = comparison(data, weights, metric).values[0] self.assertEqual(2, output)
def testRelativeToJackknifeSingleComparisonBaselineSecond(self): data = pd.DataFrame({"X": [1, 2, 3, 4, 5, 6], "Y": [0, 0, 0, 1, 1, 1]}) metric = metrics.Sum("X") comparison = comparisons.AbsoluteDifference("Y", 1) se_method = standard_errors.Jackknife() output = core.Analyze(data).relative_to( comparison).with_standard_errors(se_method).calculate( metric).run() rowindex = pd.Index([0], name="Y") correct = pd.DataFrame( np.array([[-9.0, np.sqrt(5 * np.var([12, 11, 10, 5, 4, 3]))]]), columns=("sum(X) Absolute Difference", "sum(X) Absolute Difference Jackknife SE"), index=rowindex) self.assertTrue(output.equals(correct))
def testRelativeTo(self): data = pd.DataFrame({ "X": [1, 2, 3, 10, 20, 30, 100, 200, 300], "Y": [0, 0, 0, 1, 1, 1, 2, 2, 2] }) metric = metrics.Sum("X") comparison = comparisons.AbsoluteDifference("Y", 0) output = core.Analyze(data).relative_to(comparison).calculate( metric).run() correct = pd.DataFrame({ "sum(X) Absolute Difference": [60 - 6, 600 - 6], "Y": [1, 2] }) correct = correct.set_index("Y") self.assertTrue(output.equals(correct))
def testRelativeToSplitsWithNoAlternativeGivesNaN(self): data = pd.DataFrame({ "X": [1, 2, 3, 4], "Y": [0, 0, 0, 1], "Z": [0, 0, 1, 1] }) metric = metrics.Sum("X") comparison = comparisons.AbsoluteDifference("Y", 0) output = core.Analyze(data).split_by("Z").relative_to( comparison).calculate(metric).run() correct = pd.DataFrame({ "sum(X) Absolute Difference": [np.nan, 4 - 3], "Z": [0, 1], "Y": [1, 1] }) correct = correct.set_index(["Z", "Y"]) self.assertTrue(output.equals(correct))
def testMultipleCalculationsRelativeTo(self): data = pd.DataFrame({ "X": (1, 2, 3, 10, 20, 30, 100, 200, 300), "Y": (0, 1, 2, 3, 4, 5, 6, 7, 8), "Experiment": ("Control", "Control", "Control", "Exp1", "Exp1", "Exp1", "Exp2", "Exp2", "Exp2") }) comparison = comparisons.AbsoluteDifference("Experiment", "Control") output = core.Analyze(data).relative_to(comparison).calculate( (metrics.Sum("X"), metrics.Sum("Y"))).run() correct = pd.DataFrame( { "sum(X) Absolute Difference": (60 - 6, 600 - 6), "sum(Y) Absolute Difference": (12 - 3, 21 - 3) }, index=pd.Index(("Exp1", "Exp2"), name="Experiment")) self.assertTrue(output.equals(correct))
def testRelativeToSplit(self): data = pd.DataFrame({ "X": [1, 2, 3, 4, 5, 6, 7, 8, 9, 0, 1, 2, 3, 4, 5, 6, 7, 8], "Y": [0, 0, 0, 1, 1, 1, 2, 2, 2, 0, 0, 0, 1, 1, 1, 2, 2, 2], "Z": [0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1] }) metric = metrics.Sum("X") comparison = comparisons.AbsoluteDifference("Y", 0) output = core.Analyze(data).split_by("Z").relative_to( comparison).calculate(metric).run() correct = pd.DataFrame({ "sum(X) Absolute Difference": [13 - 5, 23 - 5, 14 - 4, 22 - 4], "Z": [0, 0, 1, 1], "Y": [1, 2, 1, 2] }) correct = correct.set_index(["Z", "Y"]) self.assertTrue(output.equals(correct))
def testDoubleComparisonDefinitionRaisesException(self): data = pd.DataFrame({"X": [1, 2, 3, 4, 5]}) comparison = comparisons.AbsoluteDifference("X", 0) with self.assertRaises(ValueError): core.Analyze(data).relative_to(comparison).relative_to(comparison)