def _assertPandasAlmostEqual(actual, expected):
# Points are unhashable, so pandas' assert_frame_equal can't check they're the same by default.
# We need to convert them to something that can be checked.
def convert_points(pt):
if any(isinstance(pt, t) for t in [PythonOnlyPoint, ExamplePoint]):
return pt.x, pt.y
return pt
assertPandasAlmostEqual(actual, expected, convert=convert_points)
def _assertPandasAlmostEqual(actual, expected):
# Points are unhashable, so pandas' assert_frame_equal can't check they're the same by default.
# We need to convert them to something that can be checked.
def convert_points(pt):
if any(isinstance(pt, t) for t in [PythonOnlyPoint, ExamplePoint]):
return (pt.x, pt.y)
return pt
assertPandasAlmostEqual(actual, expected, convert=convert_points)
def _assertPandasAlmostEqual(actual, expected, sortby):
def convert_estimators(x): # note convertion makes estimators invariant to training order.
if isinstance(x, SparkSklearnEstimator):
x = x.estimator
if isinstance(x, LinearRegression) or isinstance(x, LogisticRegression):
return x.coef_, x.intercept_
if isinstance(x, PCA):
return _sortByComponentWeight(x)
return x
assertPandasAlmostEqual(actual, expected, convert=convert_estimators, sortby=sortby)
def _assertPandasAlmostEqual(actual, expected, sortby):
def convert_estimators(
x
): # note convertion makes estimators invariant to training order.
if isinstance(x, SparkSklearnEstimator):
x = x.estimator
if isinstance(x, LinearRegression) or isinstance(
x, LogisticRegression):
return x.coef_, x.intercept_
if isinstance(x, PCA):
return _sortByComponentWeight(x)
return x
assertPandasAlmostEqual(actual,
expected,
convert=convert_estimators,
sortby=sortby)
