Support both RMS and Max Linearity Validation
We have a spec for both Max and RMS linearity errors, but previously
this test only actually checked the maximum error. This worked well
enough, but for completeness, it should report and compute both of
the values. To do this, this CL adds a parameter when we instantiate
a LinearityValidator object to say if it should validate the RMS error
or the Max error.
BUG=chromium:586298
TEST=ran through the tests by hand and the values seem correct, and
the generated report looks good.
Change-Id: I1fb90cb3e410b9e23366256831c30514953d37ed
Signed-off-by: Charlie Mooney <charliemooney@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/327329
Reviewed-by: Shyh-In Hwang <josephsih@chromium.org>
diff --git a/tests/test_configurations.py b/tests/test_configurations.py
index 5a46f9e..dc7de4e 100644
--- a/tests/test_configurations.py
+++ b/tests/test_configurations.py
@@ -75,7 +75,10 @@
NO_REVERSED_MOTION_CRITERIA_MM = '<= 1, ~ +0.5'
RANGE_CRITERIA_MM = '<= 0.5, ~ +1.0'
REPORT_RATE_CRITERIA_HTZ = '>= 60'
-LINEARITY_CRITERIA_MM = '<= 0.8, ~ +2.4'
+LINEARITY_MAX_CRITERIA_MM = '<= 2, ~ +1.0'
+LINEARITY_RMS_CRITERIA_MM = '<= 0.5, ~ +0.5'
+RELAXED_LINEARITY_MAX_CRITERIA_MM = '<= 4, ~ +2.0'
+RELAXED_LINEARITY_RMS_CRITERIA_MM = '<= 1.0, ~ +1.0'
PINCH_CRITERIA_MM = '>= 15, ~ -5'
RELAXED_LINEARITY_CRITERIA_MM = '<= 1.5, ~ +3.0'
STATIONARY_FINGER_CRITERIA_MM = '<= 1.0'
@@ -142,7 +145,9 @@
validators=(
DiscardInitialSecondsValidatorWrapper(FingerCountValidator('== 1')),
DiscardInitialSecondsValidatorWrapper(
- LinearityValidator(LINEARITY_CRITERIA_MM)),
+ LinearityValidator(LINEARITY_RMS_CRITERIA_MM, is_rms=True)),
+ DiscardInitialSecondsValidatorWrapper(
+ LinearityValidator(LINEARITY_MAX_CRITERIA_MM, is_rms=False)),
DiscardInitialSecondsValidatorWrapper(
NoGapValidator(NO_GAP_CRITERIA_RATIO)),
DiscardInitialSecondsValidatorWrapper(
@@ -172,7 +177,8 @@
},
validators=(
FingerCountValidator('== 1'),
- LinearityValidator(LINEARITY_CRITERIA_MM),
+ LinearityValidator(LINEARITY_RMS_CRITERIA_MM, is_rms=True),
+ LinearityValidator(LINEARITY_MAX_CRITERIA_MM, is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM),
@@ -196,7 +202,6 @@
},
validators=(
FingerCountValidator('== 1'),
- LinearityValidator(LINEARITY_CRITERIA_MM),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM),
@@ -225,8 +230,8 @@
},
validators=(
FingerCountValidator('== 2'),
- LinearityValidator(LINEARITY_CRITERIA_MM, finger=0),
- LinearityValidator(LINEARITY_CRITERIA_MM, finger=1),
+ LinearityValidator(LINEARITY_RMS_CRITERIA_MM, is_rms=True),
+ LinearityValidator(LINEARITY_MAX_CRITERIA_MM, is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM),
@@ -260,7 +265,8 @@
},
validators=(
FingerCountValidator('== 2'),
- LinearityValidator(LINEARITY_CRITERIA_MM, finger=1),
+ LinearityValidator(LINEARITY_RMS_CRITERIA_MM, is_rms=True),
+ LinearityValidator(LINEARITY_MAX_CRITERIA_MM, is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM, finger=1),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM, finger=1),
@@ -462,7 +468,10 @@
},
validators=(
FingerCountValidator('== 2'),
- LinearityValidator(RELAXED_LINEARITY_CRITERIA_MM, finger=1),
+ LinearityValidator(RELAXED_LINEARITY_RMS_CRITERIA_MM, finger=1,
+ is_rms=True),
+ LinearityValidator(RELAXED_LINEARITY_MAX_CRITERIA_MM, finger=1,
+ is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM, finger=1),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM, finger=1),
@@ -488,7 +497,8 @@
},
validators=(
FingerCountValidator('== 1'),
- LinearityValidator(RELAXED_LINEARITY_CRITERIA_MM),
+ LinearityValidator(RELAXED_LINEARITY_RMS_CRITERIA_MM, is_rms=True),
+ LinearityValidator(RELAXED_LINEARITY_MAX_CRITERIA_MM, is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM),
@@ -513,8 +523,8 @@
},
validators=(
FingerCountValidator('== 2'),
- LinearityValidator(RELAXED_LINEARITY_CRITERIA_MM, finger=0),
- LinearityValidator(RELAXED_LINEARITY_CRITERIA_MM, finger=1),
+ LinearityValidator(RELAXED_LINEARITY_RMS_CRITERIA_MM, is_rms=True),
+ LinearityValidator(RELAXED_LINEARITY_MAX_CRITERIA_MM, is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM),
@@ -537,7 +547,8 @@
},
validators=(
FingerCountValidator('== 2'),
- LinearityValidator(LINEARITY_CRITERIA_MM, finger=1),
+ LinearityValidator(LINEARITY_RMS_CRITERIA_MM, finger=1, is_rms=True),
+ LinearityValidator(LINEARITY_MAX_CRITERIA_MM, finger=1, is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM, finger=1),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM, finger=1),
@@ -561,8 +572,8 @@
},
validators=(
FingerCountValidator('== 2'),
- LinearityValidator(RELAXED_LINEARITY_CRITERIA_MM, finger=0),
- LinearityValidator(RELAXED_LINEARITY_CRITERIA_MM, finger=1),
+ LinearityValidator(RELAXED_LINEARITY_RMS_CRITERIA_MM, is_rms=True),
+ LinearityValidator(RELAXED_LINEARITY_MAX_CRITERIA_MM, is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM),
@@ -583,7 +594,8 @@
},
validators=(
FingerCountValidator('== 4'),
- LinearityValidator(LINEARITY_CRITERIA_MM, finger=0),
+ LinearityValidator(LINEARITY_RMS_CRITERIA_MM, finger=0, is_rms=True),
+ LinearityValidator(LINEARITY_MAX_CRITERIA_MM, finger=0, is_rms=False),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM, finger=0),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM, finger=0),
@@ -646,7 +658,6 @@
validators=(
FingerCountValidator('== 1'),
HysteresisValidator(HYSTERESIS_CRITERIA_RATIO),
- LinearityValidator(LINEARITY_CRITERIA_MM),
NoGapValidator(NO_GAP_CRITERIA_RATIO),
NoReversedMotionMiddleValidator(NO_REVERSED_MOTION_CRITERIA_MM),
NoReversedMotionEndsValidator(NO_REVERSED_MOTION_CRITERIA_MM),
diff --git a/tests/validator/validators.py b/tests/validator/validators.py
index 76f7e56..b05c105 100644
--- a/tests/validator/validators.py
+++ b/tests/validator/validators.py
@@ -423,19 +423,20 @@
# Define the partial group size for calculating Mean Squared Error
MSE_PARTIAL_GROUP_SIZE = 1
- def __init__(self, criteria_str, mf=None, finger=None):
- name = self.__class__.__name__
+ def __init__(self, criteria_str, mf=None, finger=None, is_rms=False):
+ name = "%s%s" % ('RMS' if is_rms else "Max", self.__class__.__name__)
name += ('Finger%d' % finger) if finger is not None else 'AllFingers'
desc = ("How straight is the line? This validator measures how much "
"the line drawn in the gesture deviates from perfectly straight. "
"This is done by performing a linear regression in X vs. Y, then "
"computing the distance from this ideal line for each point. "
- "The maximum deviation, serves as the observed value. Note: "
- "this validator may be run on any one or all of the fingers in "
+ "The maximum or RMS deviation, serves as the observed value."
+ "Note: this validator may be run on any one or all of the fingers in "
"the gesture. If a finger number is indicated in the name, that "
"corresponds to the order they touched the pad, otherwise all "
"fingers are checked.")
self.finger = finger
+ self.is_rms = is_rms
super(LinearityValidator, self).__init__(criteria_str, mf, name, desc)
def _CalculateResiduals(self, line, times, values):
@@ -457,8 +458,7 @@
return [float(line(t) - v) for t, v in zip(times, values)]
def _SimpleLinearRegression(self, times, values):
- """Calculate the simple linear regression line and returns the
- sum of squared residuals.
+ """Calculate the simple linear regression line and returns the residuals.
@param times: the list of time instants
@param values: the list of corresponding x or y coordinates
@@ -492,29 +492,13 @@
# Compute the fitting errors of the specified segments.
return self._CalculateResiduals(regress_line, mid_segment_t, mid_segment_y)
- def _ErrorsForSingleAxis(self, times, values):
- """ Calculate linearity errors for one set of data points vs time """
- # It is fine if axis-time is a horizontal line.
- errors_px = self._SimpleLinearRegression(times, values)
- if not errors_px:
- return (0, 0)
-
- # Calculate the max errors
- max_err_px = max(map(abs, errors_px))
-
- # Calculate the root mean square errors
- e2 = [e * e for e in errors_px]
- rms_err_px = (float(sum(e2)) / len(e2)) ** 0.5
-
- return (max_err_px, rms_err_px)
-
def Validate(self, snapshots):
""" Check if the fingers conform to specified criteria. """
paths = self._SeparatePaths(snapshots)
if self.finger is not None:
paths = self._PathOfNthFinger(self.finger, paths)
- global_max_err_mm = float('-inf')
+ err_mm = []
for path in paths:
start, middle, end = self._SegmentPath(path)
path = middle
@@ -527,26 +511,41 @@
# vertical. To combat this, we fit in both x vs y as well was y vs x
# and then select whichever fits better (less error) as the correct
# fit.
- max_xy_err_mm, rms_xy_err_mm = self._ErrorsForSingleAxis(xs_mm, ys_mm)
- max_yx_err_mm, rms_yx_err_mm = self._ErrorsForSingleAxis(ys_mm, xs_mm)
- err_mm = min(max_xy_err_mm, max_yx_err_mm)
+ best_err_mm = []
+ xy_err_mm = self._SimpleLinearRegression(xs_mm, ys_mm)
+ yx_err_mm = self._SimpleLinearRegression(ys_mm, xs_mm)
+ if xy_err_mm:
+ if not yx_err_mm or max(xy_err_mm) < max(yx_err_mm):
+ best_err_mm = xy_err_mm
+ else:
+ best_err_mm = yx_err_mm
+ elif yx_err_mm:
+ best_err_mm = yx_err_mm
- global_max_err_mm = max(global_max_err_mm, err_mm)
+ if len(best_err_mm) > 0:
+ err_mm.append(best_err_mm)
result = Result()
result.name = self.name
result.description = self.description
- result.units = 'mm'
+ result.units = 'mm %s' % 'rms' if self.is_rms else 'max'
result.criteria = self.criteria_str
- if global_max_err_mm < 0:
- global_max_err_mm = float('inf')
+
+ max_err_mm = rms_err_mm = float('inf')
+ if len(err_mm) > 0:
+ max_err_mm = max([max(errors) for errors in err_mm])
+ rms_errs_mm = [(sum([err ** 2 for err in errors]) /
+ float(len(errors))) ** 0.5 for errors in err_mm]
+ rms_err_mm = max(rms_errs_mm)
+ else:
if not paths and self.finger is not None:
result.error = ('There was no finger #%d as far as we can tell.' %
self.finger)
else:
result.error = ('Unable to compute linearity. Perhaps there were '
'not enough events collected.')
- result.observed = global_max_err_mm
+
+ result.observed = rms_err_mm if self.is_rms else max_err_mm
result.score = self.fc.mf.grade(result.observed)
return result
