def test_MlpManager(self):
mng = MlpManager(ns=1)
mng.createMlp(self.output, self.factors2, self.output, [10])
assert_array_equal(mng.getMlpTopology(), [2*9 + 2*9, 10, 3])
mng = MlpManager()
mng.createMlp(self.output, self.factors, self.output, [10])
assert_array_equal(mng.getMlpTopology(), [2*1 + 1*1, 10, 3])
def test_predict(self):
mng = MlpManager()
mng.createMlp(self.output, self.factors, self.output, [10])
weights = mng.copyWeights()
# Set weights=0
# then the output must be sigmoid(0)
layers = []
for layer in weights:
shape = layer.shape
layers.append(np.zeros(shape))
mng.setMlpWeights(layers)
mng._predict(self.output, self.factors)
# Prediction
raster = mng.getPrediction(self.output, self.factors)
assert_array_equal(raster.getBand(1), sigmoid(0)*np.zeros((3,3)))
# Confidence
confid = mng.getConfidence()
assert_array_equal(confid.getBand(1), sigmoid(0)*np.zeros((3,3)))
def test_setTrainingData(self):
mng = MlpManager()
mng.createMlp(self.output, self.factors, self.output, [10])
stat = self.factors[0].getBandStat(1) # mean & std
m,s = stat['mean'], stat['std']
mng.setTrainingData(self.output, self.factors, self.output, shuffle=False)
min, max = mng.sigmin, mng.sigmax
data = np.array(
[
((0,3), [0, 1], (1.0-m)/s, [min, max, min]),
((1,3), [0, 0], (1.0-m)/s, [min, min, max]),
((2,3), [0, 1], (3.0-m)/s, [min, max, min]),
((0,2), [0, 1], (3.0-m)/s, [min, max, min]),
((1,2), [0, 0], (2.0-m)/s, [min, min, max]),
((2,2), [0, 1], (1.0-m)/s, [min, max, min]),
((0,1), [1, 0], (0.0-m)/s, [max, min, min]),
((1,1), [0, 1], (3.0-m)/s, [min, max, min]),
((2,1), [0, 0], (1.0-m)/s, [min, min, max]),
],
dtype=[('coords', float, 2),('state', float, (2,)), ('factors', float, (1,)), ('output', float, 3)]
)
self.assertEqual(mng.data.shape, (9,))
for i in range(len(data)):
assert_array_equal(data[i]['coords'], mng.data[i]['coords'])
assert_array_equal(data[i]['factors'], mng.data[i]['factors'])
assert_array_equal(data[i]['output'], mng.data[i]['output'])
# two input rasters
mng = MlpManager(ns=1)
mng.createMlp(self.output, self.factors2, self.output, [10])
mng.setTrainingData(self.output, self.factors2, self.output)
data = [
{
'factors': (np.array([ 1., 1., 3., 3., 2., 1., 0., 3., 1.,
1., 1., 3., 3., 2., 1., 0., 3., 1.]) - stat['mean'])/stat['std'],
'output': np.array([min, min, max]),
'state': np.array([0,1, 0,0, 0,1, 0,1, 0,0, 0,1, 1,0, 0,1, 0,0])
}
]
self.assertEqual(mng.data.shape, (1,))
assert_array_equal(data[0]['factors'], mng.data[0]['factors'])
assert_array_equal(data[0]['output'], mng.data[0]['output'])
assert_array_equal(data[0]['state'], mng.data[0]['state'])
# Multiband input
mng = MlpManager(ns=1)
mng.createMlp(self.output, self.factors3, self.output, [10])
stat1 = self.factors3[0].getBandStat(1) # mean & std
m1,s1 = stat1['mean'], stat1['std']
stat2 = self.factors3[0].getBandStat(2) # mean & std
m2,s2 = stat2['mean'], stat2['std']
mng.setTrainingData(self.output, self.factors3, self.output)
data = [
{
'factors': np.array([ (1.-m1)/s1, (2.-m1)/s1, (1.-m1)/s1, (1.-m1)/s1, (2.-m1)/s1, (1.-m1)/s1, (0.-m1)/s1, (1.-m1)/s1, (2.-m1)/s1,
(1.-m2)/s2, (1.-m2)/s2, (3.-m2)/s2, (3.-m2)/s2, (2.-m2)/s2, (1.-m2)/s2, (0.-m2)/s2, (3.-m2)/s2, (1.-m2)/s2]),
'output': np.array([min, min, max]),
'state': np.array([0,1, 0,0, 0,1, 0,1, 0,0, 0,1, 1,0, 0,1, 0,0])
}
]
self.assertEqual(mng.data.shape, (1,))
assert_array_equal(data[0]['factors'], mng.data[0]['factors'])
assert_array_equal(data[0]['output'], mng.data[0]['output'])
assert_array_equal(data[0]['state'], mng.data[0]['state'])
# Complex case:
mng = MlpManager(ns=1)
mng.createMlp(self.output, self.factors4, self.output, [10])
stat1 = self.factors4[0].getBandStat(1) # mean & std
m1,s1 = stat1['mean'], stat1['std']
stat2 = self.factors4[0].getBandStat(2) # mean & std
m2,s2 = stat2['mean'], stat2['std']
stat3 = self.factors4[1].getBandStat(1)
m3,s3 = stat3['mean'], stat2['std']
mng.setTrainingData(self.output, self.factors4, self.output)
data = [
{
'factors': np.array([ (1.-m1)/s1, (2.-m1)/s1, (1.-m1)/s1, (1.-m1)/s1, (2.-m1)/s1, (1.-m1)/s1, (0.-m1)/s1, (1.-m1)/s1, (2.-m1)/s1,
(1.-m2)/s2, (1.-m2)/s2, (3.-m2)/s2, (3.-m2)/s2, (2.-m2)/s2, (1.-m2)/s2, (0.-m2)/s2, (3.-m2)/s2, (1.-m2)/s2,
(1.-m3)/s3, (1.-m3)/s3, (3.-m3)/s3, (3.-m3)/s3, (2.-m3)/s3, (1.-m3)/s3, (0.-m3)/s3, (3.-m3)/s3, (1.-m3)/s3]),
'output': np.array([min, min, max]),
'state': np.array([0,1, 0,0, 0,1, 0,1, 0,0, 0,1, 1,0, 0,1, 0,0])
}
]
self.assertEqual(mng.data.shape, (1,))
assert_array_equal(data[0]['factors'], mng.data[0]['factors'])
assert_array_equal(data[0]['output'], mng.data[0]['output'])
assert_array_equal(data[0]['state'], mng.data[0]['state'])
def test_predict(self):
mng = MlpManager()
mng.createMlp(self.output, self.factors, self.output, [10])
weights = mng.copyWeights()
# Set weights=0
# then the output must be sigmoid(0)
layers = []
for layer in weights:
shape = layer.shape
layers.append(np.zeros(shape))
mng.setMlpWeights(layers)
mng._predict(self.output, self.factors)
# Prediction ( the output must be sigmoid(0) )
raster = mng.getPrediction(self.output, self.factors, calcTransitions=True)
assert_array_equal(raster.getBand(1), sigmoid(0)*np.ones((3,3)))
# Confidence is zero
confid = mng.getConfidence()
self.assertEquals(confid.getBand(1).dtype, np.uint8)
assert_array_equal(confid.getBand(1), np.zeros((3,3)))
# Transition Potentials (is (sigmoid(0) - sigmin)/sigrange )
potentials = mng.getTransitionPotentials()
cats = self.output.getBandGradation(1)
for cat in cats:
map = potentials[cat]
self.assertEquals(map.getBand(1).dtype, np.uint8)
assert_array_equal(map.getBand(1), 50*np.ones((3,3)))
def test_train(self):
mng = MlpManager()
mng.createMlp(self.output, self.factors, self.output, [10])
mng.setTrainingData(self.output, self.factors, self.output)
mng.train(1, valPercent=50)
val = mng.getMinValError()
tr = mng.getTrainError()
mng.train(20, valPercent=50, continue_train=True)
self.assertGreaterEqual(val, mng.getMinValError())
mng = MlpManager(ns=1)
mng.createMlp(self.state1, self.factors1, self.output1, [10])
mng.setTrainingData(self.state1, self.factors1, self.output1)
mng.train(1, valPercent=20)
predict = mng.getPrediction(self.state1, self.factors1)
mask = predict.getBand(1).mask
self.assertTrue(not all(mask.flatten()))
def main(initRaster, finalRaster, factors):
print 'Start Reading Init Data...', clock()
initRaster = Raster(initRaster)
finalRaster = Raster(finalRaster)
factors = [Raster(rasterName) for rasterName in factors]
print 'Finish Reading Init Data', clock(), '\n'
print "Start Making CrossTable...", clock()
crosstab = CrossTableManager(initRaster, finalRaster)
#print crosstab.getTransitionStat()
print "Finish Making CrossTable", clock(), '\n'
# Create and Train Analyst
print 'Start creating AreaAnalyst...', clock()
analyst = AreaAnalyst(initRaster, finalRaster)
print 'Finish creating AreaAnalyst ...', clock(), '\n'
print 'Start Making Change Map...', clock()
analyst = AreaAnalyst(initRaster, finalRaster)
changeMap = analyst.getChangeMap()
print 'Finish Making Change Map', clock(), '\n'
#~ # Create and Train ANN Model
model = MlpManager(ns=1)
model.createMlp(initRaster, factors, changeMap, [10])
print 'Start Setting MLP Trainig Data...', clock()
model.setTrainingData(initRaster,
factors,
changeMap,
mode='Stratified',
samples=1000)
print 'Finish Setting Trainig Data', clock(), '\n'
print 'Start MLP Training...', clock()
model.train(20, valPercent=20)
print 'Finish Trainig', clock(), '\n'
# print 'Start ANN Prediction...', clock()
# predict = model.getPrediction(initRaster, factors, calcTransitions=True)
# confidence = model.getConfidence()
# potentials = model.getTransitionPotentials()
#~ # Create and Train LR Model
#~ model = LR(ns=0)
#~ print 'Start Setting LR Trainig Data...', clock()
#~ model.setState(initRaster)
#~ model.setFactors(factors)
#~ model.setOutput(changeMap)
#~ model.setMode('Stratified')
#~ model.setSamples(100)
#~ model.setTrainingData()
#~ print 'Finish Setting Trainig Data', clock(), '\n'
#~ print 'Start LR Training...', clock()
#~ model.train()
#~ print 'Finish Trainig', clock(), '\n'
#~
#~ print 'Start LR Prediction...', clock()
#~ predict = model.getPrediction(initRaster, factors, calcTransitions=True)
#~ print 'Finish LR Prediction...', clock(), '\n'
# Create and Train WoE Model
# print 'Start creating AreaAnalyst...', clock()
# analyst = AreaAnalyst(initRaster, finalRaster)
# print 'Finish creating AreaAnalyst ...', clock(), '\n'
# print 'Start creating WoE model...', clock()
# bins = {0: [[1000, 3000]], 1: [[200, 500, 1500]]}
# model = WoeManager(factors, analyst, bins= bins)
# model.train()
# print 'Finish creating WoE model...', clock(), '\n'
#~ # Create and Train MCE Model
#~ print 'Start creating MCE model...', clock()
#~ matrix = [
#~ [1, 6],
#~ [1.0/6, 1]
#~ ]
#~ model = MCE(factors, matrix, 2, 3, analyst)
#~ print 'Finish creating MCE model...', clock(), '\n'
# predict = model.getPrediction(initRaster, factors, calcTransitions=True)
# confidence = model.getConfidence()
# potentials = model.getTransitionPotentials()
# filename = 'predict.tif'
# confname = 'confidence.tif'
# trans_prefix='trans_'
# try:
# predict.save(filename)
# confidence.save(confname)
# if potentials != None:
# for k,v in potentials.iteritems():
# map = v.save(trans_prefix+str(k) + '.tif')
# finally:
# os.remove(filename)
# #pass
# print 'Finish Saving...', clock(), '\n'
# simulation
print 'Start Simulation...', clock()
simulator = Simulator(initRaster, factors, model, crosstab)
# Make 1 cycle of simulation:
simulator.setIterationCount(1)
simulator.simN()
monteCarloSim = simulator.getState() # Result of MonteCarlo simulation
errors = simulator.errorMap(finalRaster) # Risk class validation
riskFunct = simulator.getConfidence() # Risk function
try:
monteCarloSim.save('simulation_result.tiff')
errors.save('risk_validation.tiff')
riskFunct.save('risk_func.tiff')
finally:
pass
# os.remove('simulation_result.tiff')
# os.remove('risk_validation.tiff')
# os.remove('risk_func.tiff')
print 'Finish Simulation', clock(), '\n'
print 'Done', clock()
def main(initRaster, finalRaster, factors):
print 'Start Reading Init Data...', clock()
initRaster = Raster(initRaster)
finalRaster = Raster(finalRaster)
factors = [Raster(rasterName) for rasterName in factors]
print 'Finish Reading Init Data', clock(), '\n'
print "Start Making CrossTable...", clock()
crosstab = CrossTableManager(initRaster, finalRaster)
#print crosstab.getTransitionStat()
print "Finish Making CrossTable", clock(), '\n'
# Create and Train Analyst
print 'Start creating AreaAnalyst...', clock()
analyst = AreaAnalyst(initRaster, finalRaster)
print 'Finish creating AreaAnalyst ...', clock(), '\n'
print 'Start Making Change Map...', clock()
analyst = AreaAnalyst(initRaster,finalRaster)
changeMap = analyst.getChangeMap()
print 'Finish Making Change Map', clock(), '\n'
#~ # Create and Train ANN Model
model = MlpManager(ns=1)
model.createMlp(initRaster, factors, changeMap, [10])
print 'Start Setting MLP Trainig Data...', clock()
model.setTrainingData(initRaster, factors, changeMap, mode='Stratified', samples=1000)
print 'Finish Setting Trainig Data', clock(), '\n'
print 'Start MLP Training...', clock()
model.train(20, valPercent=20)
print 'Finish Trainig', clock(), '\n'
# print 'Start ANN Prediction...', clock()
# predict = model.getPrediction(initRaster, factors, calcTransitions=True)
# confidence = model.getConfidence()
# potentials = model.getTransitionPotentials()
#~ # Create and Train LR Model
#~ model = LR(ns=0)
#~ print 'Start Setting LR Trainig Data...', clock()
#~ model.setState(initRaster)
#~ model.setFactors(factors)
#~ model.setOutput(changeMap)
#~ model.setMode('Stratified')
#~ model.setSamples(100)
#~ model.setTrainingData()
#~ print 'Finish Setting Trainig Data', clock(), '\n'
#~ print 'Start LR Training...', clock()
#~ model.train()
#~ print 'Finish Trainig', clock(), '\n'
#~
#~ print 'Start LR Prediction...', clock()
#~ predict = model.getPrediction(initRaster, factors, calcTransitions=True)
#~ print 'Finish LR Prediction...', clock(), '\n'
# Create and Train WoE Model
# print 'Start creating AreaAnalyst...', clock()
# analyst = AreaAnalyst(initRaster, finalRaster)
# print 'Finish creating AreaAnalyst ...', clock(), '\n'
# print 'Start creating WoE model...', clock()
# bins = {0: [[1000, 3000]], 1: [[200, 500, 1500]]}
# model = WoeManager(factors, analyst, bins= bins)
# model.train()
# print 'Finish creating WoE model...', clock(), '\n'
#~ # Create and Train MCE Model
#~ print 'Start creating MCE model...', clock()
#~ matrix = [
#~ [1, 6],
#~ [1.0/6, 1]
#~ ]
#~ model = MCE(factors, matrix, 2, 3, analyst)
#~ print 'Finish creating MCE model...', clock(), '\n'
# predict = model.getPrediction(initRaster, factors, calcTransitions=True)
# confidence = model.getConfidence()
# potentials = model.getTransitionPotentials()
# filename = 'predict.tif'
# confname = 'confidence.tif'
# trans_prefix='trans_'
# try:
# predict.save(filename)
# confidence.save(confname)
# if potentials != None:
# for k,v in potentials.iteritems():
# map = v.save(trans_prefix+str(k) + '.tif')
# finally:
# os.remove(filename)
# #pass
# print 'Finish Saving...', clock(), '\n'
# simulation
print 'Start Simulation...', clock()
simulator = Simulator(initRaster, factors, model, crosstab)
# Make 1 cycle of simulation:
simulator.setIterationCount(1)
simulator.simN()
monteCarloSim = simulator.getState() # Result of MonteCarlo simulation
errors = simulator.errorMap(finalRaster) # Risk class validation
riskFunct = simulator.getConfidence() # Risk function
try:
monteCarloSim.save('simulation_result.tiff')
errors.save('risk_validation.tiff')
riskFunct.save('risk_func.tiff')
finally:
pass
# os.remove('simulation_result.tiff')
# os.remove('risk_validation.tiff')
# os.remove('risk_func.tiff')
print 'Finish Simulation', clock(), '\n'
print 'Done', clock()