def setUp(self):
dimensions = [
numpy.array(range(1, 5), dtype=numpy.float64),
numpy.array(range(1, 6), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
values = numpy.ones(60)
self.lut = LookupTable(values, dimensions)
def test_get_values_big(self):
dimensions = [
numpy.array(range(1, 5), dtype=numpy.float64),
numpy.array(range(1, 6), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
values = numpy.array(range(0, 120))
self.lut = LookupTable(values, dimensions, 2)
result = self.lut.getValues(
[numpy.ones(300000),
numpy.ones(300000),
numpy.ones(300000)])
self.assertEqual((2, 300000), result.shape)
def test_get_value_length_2(self):
dimensions = [
numpy.array(range(1, 5), dtype=numpy.float64),
numpy.array(range(1, 6), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
values = numpy.array(range(0, 120))
self.lut = LookupTable(values, dimensions, 2)
result = self.lut.getValue([1, 1, 1])
self.assertEqual((2, ), result.shape)
[a, b] = result
self.assertEqual(0, a)
self.assertEqual(1, b)
def __init__(self, a0, a1, needs_lut):
self.needs_lut = needs_lut
if not needs_lut:
self.a0 = a0
self.a1 = a1
else:
season = numpy.array(range(
0, 4), dtype=numpy.float64) # spring, summer, autumn, winter
height = numpy.array([400, 800, 1000]) # height in meter
zenith = numpy.array(range(
5, 56, 5), dtype=numpy.float64) # 5 - 55deg in 5deg steps
dimensions = [season, height, zenith]
mono_coeff = Mono()
values = numpy.array(mono_coeff.values, dtype=numpy.float64)
self.lut = LookupTable(values, dimensions, 2)
self.check = None
def __init__(self, a0, a1, a2, a3, needs_lut):
if needs_lut:
self.has_lut = True
season = numpy.array(range(
0, 4), dtype=numpy.float64) # spring, summer, autumn, winter
height = numpy.array([400, 800, 1000]) # height in meter
zenith = numpy.array(range(
5, 56, 5), dtype=numpy.float64) # 5 - 55deg in 5deg steps
dimensions = [season, height, zenith]
# TODO needs to be loaded from file!
values = numpy.tile([3.079654, 0.988938, 1.867209, -0.434251],
len(season) * len(zenith) * len(height))
self.lut = LookupTable(values, dimensions, 4)
else:
self.has_lut = False
self.a0 = a0
self.a1 = a1
self.a2 = a2
self.a3 = a3
self.check = None
def test_get_values_length_2(self):
dimensions = [
numpy.array(range(1, 5), dtype=numpy.float64),
numpy.array(range(1, 6), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
values = numpy.array(range(0, 120))
self.lut = LookupTable(values, dimensions, 2)
result = self.lut.getValues([
numpy.array([1, 1, 1]),
numpy.array([1, 1, 1]),
numpy.array([1, 1, 1])
])
self.assertEqual((2, 3), result.shape)
[[a, b, c], [d, e, f]] = result
self.assertEqual(0, a)
self.assertEqual(0, b)
self.assertEqual(0, c)
self.assertEqual(1, d)
self.assertEqual(1, e)
self.assertEqual(1, f)
def create_wrs_lut(self):
paths = numpy.array(range(1, 234), dtype=numpy.float64)
rows = numpy.array(range(1, 249), dtype=numpy.float64)
# points=['CTR LAT', 'CTR LON', 'UL LAT', 'UL LON', 'UR LAT', 'UR LON', 'LL LAT', 'LL LON', 'LR LAT', 'LR LON']
points = numpy.array(range(0, 10), dtype=numpy.float64)
dimensions = [paths, rows, points]
# file = File('wrs.py')
# props = Properties()
# inStream = file.toURI().toURL().openStream()
# props.clear()
# props.load(inStream)
# values = numpy.array(props.getProperty('values').split(','), dtype=numpy.float64)
wrs = Wrs()
values = numpy.array(wrs.values, dtype=numpy.float64)
return LookupTable(values, dimensions)
class MonoWindowAlgo:
def __init__(self, a0, a1, needs_lut):
self.needs_lut = needs_lut
if not needs_lut:
self.a0 = a0
self.a1 = a1
else:
season = numpy.array(range(
0, 4), dtype=numpy.float64) # spring, summer, autumn, winter
height = numpy.array([400, 800, 1000]) # height in meter
zenith = numpy.array(range(
5, 56, 5), dtype=numpy.float64) # 5 - 55deg in 5deg steps
dimensions = [season, height, zenith]
mono_coeff = Mono()
values = numpy.array(mono_coeff.values, dtype=numpy.float64)
self.lut = LookupTable(values, dimensions, 2)
self.check = None
def get_season(self, start_date):
utils = Utils()
return utils.get_season(start_date)
def compute_lswt(self, data, season, height, zenith_data):
if not self.needs_lut:
return self.a0 * data + self.a1
else:
lswt = numpy.zeros(numpy.shape(data))
lswt[numpy.where(numpy.isnan(data))] = Float.NaN
season = season[numpy.where(~numpy.isnan(data))]
height = height[numpy.where(~numpy.isnan(data))]
zenith_data = zenith_data[numpy.where(~numpy.isnan(data))]
size = numpy.shape(season)
if size != (0, ):
if len(size) == 2:
(shape_x, shape_y) = size
result = self.lut.getValues([
season[::2, ::2], height[::2, ::2],
zenith_data[::2, ::2]
])
[a0, a1] = result
a0 = numpy.repeat(numpy.repeat(a0, 2, axis=0), 2, axis=1)
a1 = numpy.repeat(numpy.repeat(a1, 2, axis=0), 2, axis=1)
a0 = a0[:shape_x, :shape_y]
a1 = a1[:shape_x, :shape_y]
elif len(size) == 1:
(shape_x, ) = size
result = self.lut.getValues(
[season[::2], height[::2], zenith_data[::2]])
[a0, a1] = result
a0 = numpy.repeat(a0, 2)
a1 = numpy.repeat(a1, 2)
a0 = a0[:shape_x]
a1 = a1[:shape_x]
lswt[numpy.where(~numpy.isnan(
data))] = a0 * data[numpy.where(~numpy.isnan(data))] + a1
return lswt
def compute_flags(self, lswt, visible, nir, bt, sat_za, sun_za, rel_az,
lwm, cmsk, height, width, sattype):
self.check = QualityCheck(lswt, visible, nir, None, bt, None, sat_za,
sun_za, rel_az, lwm, cmsk, height, width,
sattype)
return self.check.check_quality()
def compute_quality_index(self, flags):
return self.check.get_quality_mask(flags)
class SplitWindowAlgo:
def __init__(self, a0, a1, a2, a3, needs_lut):
if needs_lut:
self.has_lut = True
season = numpy.array(range(
0, 4), dtype=numpy.float64) # spring, summer, autumn, winter
height = numpy.array([400, 800, 1000]) # height in meter
zenith = numpy.array(range(
5, 56, 5), dtype=numpy.float64) # 5 - 55deg in 5deg steps
dimensions = [season, height, zenith]
# TODO needs to be loaded from file!
values = numpy.tile([3.079654, 0.988938, 1.867209, -0.434251],
len(season) * len(zenith) * len(height))
self.lut = LookupTable(values, dimensions, 4)
else:
self.has_lut = False
self.a0 = a0
self.a1 = a1
self.a2 = a2
self.a3 = a3
self.check = None
def get_season(self, start_date):
utils = Utils()
return utils.get_season(start_date)
def compute_lswt_lut(self, lower_data, upper_data, zenith_data, season,
height):
lswt = numpy.zeros(numpy.shape(lower_data))
lswt[numpy.where(
numpy.logical_or(numpy.isnan(lower_data),
numpy.isnan(upper_data)))] = Float.NaN
season = season[numpy.where(
numpy.logical_and(~numpy.isnan(lower_data),
~numpy.isnan(upper_data)))]
zenith_data = zenith_data[numpy.where(
numpy.logical_and(~numpy.isnan(lower_data),
~numpy.isnan(upper_data)))]
height = height[numpy.where(
numpy.logical_and(~numpy.isnan(lower_data),
~numpy.isnan(upper_data)))]
size = numpy.shape(season)
if size != (0, ):
[self.a0, self.a1, self.a2,
self.a3] = self.lut.getValues([season, height, zenith_data])
lswt[numpy.where(
numpy.logical_and(
~numpy.isnan(lower_data),
~numpy.isnan(upper_data)))] = self.compute_lswt(
lower_data[numpy.where(
numpy.logical_and(~numpy.isnan(lower_data),
~numpy.isnan(upper_data)))],
upper_data[numpy.where(
numpy.logical_and(~numpy.isnan(lower_data),
~numpy.isnan(upper_data)))],
zenith_data)
return lswt
def compute_lswt(self, lower_data, upper_data, zenith_data):
lswt = self.a0 + self.a1 * lower_data + self.a2 * (lower_data - upper_data) + \
self.a3 * (1 - 1 / (numpy.cos(zenith_data * (numpy.math.pi / 180.0)))) * (lower_data - upper_data)
return lswt
def compute_flags(self, lswt, visible, nir, bt3, lower_bt, upper_bt,
sat_za, sun_za, rel_az, lwm, cmsk, height, width,
sattype):
self.check = QualityCheck(lswt, visible, nir, bt3, lower_bt, upper_bt,
sat_za, sun_za, rel_az, lwm, cmsk, height,
width, sattype)
return self.check.check_quality()
def compute_quality_index(self, flags):
return self.check.get_quality_mask(flags)
print('Mean game performance: %f' % mean_performance)
return Q
# create a matrix of all input states for blackjack
X = []
for player in range(2, 22):
for dealer in range(1, 22):
for ace in range(2):
for action in range(2):
X.append([player, dealer, ace, action])
X = np.array(X)
# ## Incremental training
# # create and initialize Q (state-action value) function
Q_random = LookupTable()
Q_random.fit(X, np.random.random(X.shape[0]))
print('Mean reward per game for random agent (Q before training): %f' %
np.mean([play(env, Q_random) for i in range(100000)]))
incremental_game_means = []
# Q = train_Q_incrementally(Q_random, measure_every=10000, game_means=incremental_game_means)
# print('Mean reward for game with trained agent: %f' % np.mean([play(env, Q) for i in range(100000)]))
# print('Mean reward for play with Q_ideal (Vegas blackjack strategy card): %f' % np.mean([play(env, Q_ideal) for i in range(100000)]))
# # Fit an SVM regression model to the lookup table version of Q to see what's
# # the best we can do with an SVM
# Q_SVM = sklearn.svm.SVR()
# Q_SVM.fit(X, Q.predict(X))
# print('Mean reward for best-fit SVM regression (SVR fit to the lookup table): %f' % np.mean([play(env, Q_SVM) for i in range(100000)]))
#
#
## Batch training
class TestLookupTable(unittest.TestCase):
def setUp(self):
dimensions = [
numpy.array(range(1, 5), dtype=numpy.float64),
numpy.array(range(1, 6), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
values = numpy.ones(60)
self.lut = LookupTable(values, dimensions)
def test_get_value(self):
self.assertEqual(1, self.lut.getValue([4, 3, 2]))
def test_get_values(self):
coordinates = [
numpy.array(range(2, 5), dtype=numpy.float64),
numpy.array(range(2, 5), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
result = self.lut.getValues(coordinates)
self.assertEqual(coordinates[0].shape, result.shape)
self.assertEqual(1, result[0])
def test_get_values_mulit_index(self):
coordinates = [
numpy.array([[1, 1, 1], [2, 2, 2], [3, 3, 3]]),
numpy.array([[1, 1, 1], [2, 2, 2], [3, 3, 3]]),
numpy.array([[1, 1, 1], [2, 2, 2], [3, 3, 3]])
]
result = self.lut.getValues(coordinates)
self.assertEqual(coordinates[0].shape, result.shape)
self.assertEqual(1, result[0][0])
def test_get_value_length_2(self):
dimensions = [
numpy.array(range(1, 5), dtype=numpy.float64),
numpy.array(range(1, 6), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
values = numpy.array(range(0, 120))
self.lut = LookupTable(values, dimensions, 2)
result = self.lut.getValue([1, 1, 1])
self.assertEqual((2, ), result.shape)
[a, b] = result
self.assertEqual(0, a)
self.assertEqual(1, b)
def test_get_values_length_2(self):
dimensions = [
numpy.array(range(1, 5), dtype=numpy.float64),
numpy.array(range(1, 6), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
values = numpy.array(range(0, 120))
self.lut = LookupTable(values, dimensions, 2)
result = self.lut.getValues([
numpy.array([1, 1, 1]),
numpy.array([1, 1, 1]),
numpy.array([1, 1, 1])
])
self.assertEqual((2, 3), result.shape)
[[a, b, c], [d, e, f]] = result
self.assertEqual(0, a)
self.assertEqual(0, b)
self.assertEqual(0, c)
self.assertEqual(1, d)
self.assertEqual(1, e)
self.assertEqual(1, f)
def test_get_values_big(self):
dimensions = [
numpy.array(range(1, 5), dtype=numpy.float64),
numpy.array(range(1, 6), dtype=numpy.float64),
numpy.array(range(1, 4), dtype=numpy.float64)
]
values = numpy.array(range(0, 120))
self.lut = LookupTable(values, dimensions, 2)
result = self.lut.getValues(
[numpy.ones(300000),
numpy.ones(300000),
numpy.ones(300000)])
self.assertEqual((2, 300000), result.shape)