def test_fuzzyfy():
features = {"f1": np.array([1.0]), "f2": np.array([5.2]), "f3": np.array([0.9])}
rules = fuzzyfy(features, **CFG)
assert_allclose(rules["low"], [0.226666666])
assert_allclose(rules["medium"], [0.733333333])
assert_allclose(rules["high"], [0.08000000])
def test_fuzzyfy_with_nan():
features = {
"f1": np.array([1.0, np.nan, 1.0, 1.0, np.nan]),
"f2": np.array([5.2, 5.2, np.nan, 5.2, np.nan]),
"f3": np.array([0.9, 0.9, 0.9, np.nan, np.nan]),
}
rules = fuzzyfy(features, **CFG)
assert_allclose(rules["low"], [0.22666667, np.nan, np.nan, np.nan, np.nan])
assert_allclose(rules["medium"], [0.733333333, np.nan, np.nan, np.nan, np.nan])
assert_allclose(rules["high"], [0.08000000, np.nan, np.nan, np.nan, np.nan])
rules = fuzzyfy(features, **CFG, require="any")
assert_allclose(rules["low"], [0.22666667, 0.34, 0.34, 0, np.nan])
assert_allclose(rules["medium"], [0.733333333, 0.6, 0.7, 0.9, np.nan])
assert_allclose(rules["high"], [0.08, 0.08, 0, 0.08, np.nan])
def test_fuzzyfy_all_nan():
features = {
"f1": np.array([np.nan]),
"f2": np.array([np.nan]),
"f3": np.array([np.nan]),
}
rules = fuzzyfy(features, **CFG)
assert_allclose(rules["low"], [np.nan])
assert_allclose(rules["medium"], [np.nan])
assert_allclose(rules["high"], [np.nan])
def morello2014(features, cfg):
"""
FIXME: Think about, should I return 0, or have an assert, and at qc.py
all qc tests are applied with a try, and in case it fails it flag
0s.
"""
#for f in cfg['features']:
# assert f in features, \
# "morello2014 requires feature %s, which is not available" \
# % f
if not np.all([f in features for f in cfg['features']]):
print("Not all features (%s) required by morello2014 are available" %
cfg['features'].keys())
try:
return np.zeros(features[features.keys()[0]].shape, dtype='i1')
except:
return 0
f = fuzzyfy(features, cfg)
## This is how Timms and Morello defined the Fuzzy Logic approach
#flag = np.zeros(N, dtype='i1')
# Flag must be np.array, not a ma.array.
flag = np.zeros(features[features.keys()[0]].shape, dtype='i1')
flag[(f['low'] > 0.5) & (f['high'] < 0.3)] = 2
flag[(f['low'] > 0.9)] = 1
# Everything else is flagged 3
flag[(f['low'] <= 0.5) | (f['high'] >= 0.3)] = 3
# Missing check if threshold was crossed, to flag as 4
# The thresholds coincide with the end of the ramp for the fuzzy set high,
# hence we can simply
flag[(f['high'] == 1.)] = 4
return flag
def morello2014(features, cfg):
"""
FIXME: Think about, should I return 0, or have an assert, and at qc.py
all qc tests are applied with a try, and in case it fails it flag
0s.
"""
#for f in cfg['features']:
# assert f in features, \
# "morello2014 requires feature %s, which is not available" \
# % f
if not np.all([f in features for f in cfg['features']]):
print("Not all features (%s) required by morello2014 are available" %
cfg['features'].keys())
try:
return np.zeros(features[features.keys()[0]].shape, dtype='i1')
except:
return 0
f = fuzzyfy(features, cfg)
## This is how Timms and Morello defined the Fuzzy Logic approach
#flag = np.zeros(N, dtype='i1')
# Flag must be np.array, not a ma.array.
flag = np.zeros(features[list(features.keys())[0]].shape, dtype='i1')
flag[(f['low'] > 0.5) & (f['high'] < 0.3)] = 2
flag[(f['low'] > 0.9)] = 1
# Everything else is flagged 3
flag[(f['low'] <= 0.5) | (f['high'] >= 0.3)] = 3
# Missing check if threshold was crossed, to flag as 4
# The thresholds coincide with the end of the ramp for the fuzzy set high,
# hence we can simply
flag[(f['high'] == 1.)] = 4
return flag