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 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 testDoubleSEMethodDefinitionRaisesException(self): data = pd.DataFrame({"X": [1, 2, 3, 4, 5]}) se_method = standard_errors.Jackknife() with self.assertRaises(ValueError): core.Analyze(data).with_standard_errors( se_method).with_standard_errors(se_method)
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 testSingleJackknifeBucketRaisesException(self): data = pd.DataFrame({"X": [1, 2, 3, 4, 5], "Y": [1, 1, 1, 1, 1]}) metric = metrics.Sum("X") se_method = standard_errors.Jackknife(unit="Y") with self.assertRaises(ValueError): core.Analyze(data).with_standard_errors(se_method).calculate( metric).run()
def testJackknife(self): data = pd.DataFrame({"X": range(11)}) metric = metrics.Sum("X") se_method = standard_errors.Jackknife() output = core.Analyze(data).with_standard_errors(se_method).calculate( metric).run() correct = pd.DataFrame(np.array([[55.0, 10.0]]), columns=("sum(X)", "sum(X) Jackknife SE")) self.assertTrue(output.equals(correct))
def testJackknifeBadSample(self): data = pd.DataFrame({"X": range(22), "Y": ([0] * 11) + ([1] * 11)}) metric = metrics.Sum("X") se_method = standard_errors.Jackknife() output = core.Analyze(data).split_by("Y").with_standard_errors( se_method).calculate(metric).run() correct = pd.DataFrame(np.array([[55.0, 10.0], [176.0, 10.0]]), columns=("sum(X)", "sum(X) Jackknife SE")) correct.index.name = "Y" self.assertTrue(output.equals(correct))
def testJackknifeRatio(self): data = pd.DataFrame({"X": [1, 2, 3, 4], "Y": [4, 3, 2, 1]}) metric = metrics.Ratio("X", "Y") se_method = standard_errors.Jackknife() output = core.Analyze(data).with_standard_errors(se_method).calculate( metric).run() estimates = np.array([9 / 6, 8 / 7, 7 / 8, 6 / 9]) rss = ((estimates - estimates.mean())**2).sum() se = np.sqrt(rss * 3 / 4) correct = pd.DataFrame([[1.0, se]], columns=("X/Y", "X/Y Jackknife SE")) self.assertTrue(output.equals(correct))
def testSplitJackknife(self): data = pd.DataFrame({ "X": np.array([range(11) + [5] * 10]).flatten(), "Y": np.array([[0] * 11 + [1] * 10]).flatten() }) metric = metrics.Sum("X") se_method = standard_errors.Jackknife() output = core.Analyze(data).split_by("Y").with_standard_errors( se_method).calculate(metric).run() rowindex = pd.Index([0, 1], name="Y") correct = pd.DataFrame(np.array([[55.0, 10.0], [50.0, 0.0]]), columns=("sum(X)", "sum(X) Jackknife SE"), index=rowindex) self.assertTrue(output.equals(correct))
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 testFiftyCIs(self): data = pd.DataFrame({"X": range(11)}) metric = metrics.Sum("X") se_method = standard_errors.Jackknife(confidence=0.50) output = core.Analyze(data).with_standard_errors(se_method).calculate( metric).run() multiplier = scipy.stats.t.ppf(0.75, 10) correct_sd = 10.0 correct_mean = 55.0 correct_lower = correct_mean - multiplier * correct_sd correct_upper = correct_mean + multiplier * correct_sd correct = pd.DataFrame(np.array( [[correct_mean, correct_lower, correct_upper]]), columns=("sum(X)", "sum(X) Jackknife CI-lower", "sum(X) Jackknife CI-upper")) self.assertTrue(output.equals(correct))
def testDataframeJackknife(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("Y") output = core.Analyze(df).with_standard_errors(se_method).calculate( metric).run() correct = pd.DataFrame( np.array([[3 / 55., np.sqrt(((3 / 15. - 0.1)**2 + 0.1**2) / 2.)], [52 / 55., np.sqrt(((12 / 15. - 0.9)**2 + 0.1**2) / 2.)]]), columns=("X Distribution", "X Distribution Jackknife SE"), index=pd.Index([0., 1.], name="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 testBadConfidenceRaisesException(self): with self.assertRaises(ValueError): standard_errors.Jackknife(confidence=95)