def train(self, data, samples, slices):
#self.gammas = self.determine_gammas_from(data)
self.gammas = [.5,]
print "Gammas determined: %s" % str(self.gammas)
# [gamma][sequence offset][dimension]
#self.active_slices = np.mgrid[0:1,0:data.shape[1]].T.reshape(data.shape[1],2).tolist()
# Make a single slice consisting of the 1st sequence element and all 3 dimensions
#self.active_slices = [ [0,[0,1]], [0,[0]] ]
self.active_slices = slices
sequences = self.make_sequences(data)
labels = data[self.sequence_length:,:].astype('float32')
# Randomly sample |samples| sequences
#self.sequences, self.labels = self.random_sample(self, sequences, labels, samples)
self.sequences, self.labels = sequences[:samples], labels[:samples, self.dimension]
KK = self.make_subsets(self.sequences, self.sequences)
Labels = np.hstack( [ self.labels, np.zeros( self.labels.shape[0] * ( len(self.gammas) * len(self.active_slices) - 1) ) ] )
#Labels = np.repeat( self.labels, len(slices), axis=0 )
weights = np.hstack( [ np.ones( self.labels.shape[0] ), np.zeros( self.labels.shape[0] * ( len(self.gammas) * len(self.active_slices) - 1 ) ) ] ).astype('float32')
print 'Training...'
# Train that shist
self.svm = SVR(epsilon=.1, C=50)
self.svm.train(KK, Labels, sample_weight = weights)
#self.svm.train(KK, Labels)
print "--> SVM Trained: %s percent SV's, risk=%s" % ( self.svm.SV_percent, self.svm.risk )
class model:
def __init__(self, dimension, gamma_samples=1000, gamma_quantile=100, sequence_length=2):
self.dimension = dimension
self.gamma_samples = gamma_samples
self.gamma_quantile = gamma_quantile
self.sequence_length = sequence_length
self.gammas = None
self.sequences = None
self.k = None
self.SVM = None
# [gamma][sequences][dimensions]
self.active_slices = None
self.all_slices = None
# data: [observation][dimension]
def train(self, data, samples, slices):
#self.gammas = self.determine_gammas_from(data)
self.gammas = [.5,]
print "Gammas determined: %s" % str(self.gammas)
# [gamma][sequence offset][dimension]
#self.active_slices = np.mgrid[0:1,0:data.shape[1]].T.reshape(data.shape[1],2).tolist()
# Make a single slice consisting of the 1st sequence element and all 3 dimensions
#self.active_slices = [ [0,[0,1]], [0,[0]] ]
self.active_slices = slices
sequences = self.make_sequences(data)
labels = data[self.sequence_length:,:].astype('float32')
# Randomly sample |samples| sequences
#self.sequences, self.labels = self.random_sample(self, sequences, labels, samples)
self.sequences, self.labels = sequences[:samples], labels[:samples, self.dimension]
KK = self.make_subsets(self.sequences, self.sequences)
Labels = np.hstack( [ self.labels, np.zeros( self.labels.shape[0] * ( len(self.gammas) * len(self.active_slices) - 1) ) ] )
#Labels = np.repeat( self.labels, len(slices), axis=0 )
weights = np.hstack( [ np.ones( self.labels.shape[0] ), np.zeros( self.labels.shape[0] * ( len(self.gammas) * len(self.active_slices) - 1 ) ) ] ).astype('float32')
print 'Training...'
# Train that shist
self.svm = SVR(epsilon=.1, C=50)
self.svm.train(KK, Labels, sample_weight = weights)
#self.svm.train(KK, Labels)
print "--> SVM Trained: %s percent SV's, risk=%s" % ( self.svm.SV_percent, self.svm.risk )
# data: [observation][dimension]
def predict(self, data):
# [sequence][point][dimension]
#points = self.make_sequences(data)
points = data
KK = self.make_subsets(points, self.sequences)
print KK.shape
raw = self.svm.predict(KK)
print raw[:data.shape[0]].shape
return raw[:data.shape[0]]
def random_sample(self, sequences, labels, samples):
full = np.hstack([sequences, np.expand_dims(labels,1)])
np.random.shuffle(full)
self.sequences = full[:samples,:-1,:]
self.labels = full[:samples,-1,self.dimension]
(full[:samples,:-1,:], full[:samples,-1,self.dimension] )
def make_sequences(self, data):
sequences = data.reshape(data.shape[0], 1, data.shape[1])
for i in np.arange( 1, self.sequence_length):
sequences = np.hstack([ sequences, np.roll(data, -i, 0).reshape(data.shape[0], 1, data.shape[1]) ])
return np.array( sequences )[:-self.sequence_length,:,:].astype('float32')
def make_subsets(self, X, Y):
kk = []
for s in self.active_slices:
if isinstance(s[1], int):
subset_X = X[:,s[0],s[1]].reshape(X.shape[0], s[0]+1, 1 )
subset_Y = Y[:,s[0],s[1]].reshape(Y.shape[0], s[0]+1, 1 )
else:
subset_X = X[:,s[0],s[1]].reshape(X.shape[0], s[0]+1, len(s[1]))
subset_Y = Y[:,s[0],s[1]].reshape(Y.shape[0], s[0]+1, len(s[1]))
for gamma in self.gammas:
# NOTE: returning to test on single matrix
kk.append( kernel_matrix(subset_X, subset_Y, gamma) )
# Construct the sparse block diagonal of the kernel matrices and extend the labels to match
# return bdiag(kk, format='csr')
if len(kk) > 1:
row = np.hstack(kk)
zeros = np.zeros( (kk[0].shape[0] * (len(kk)-1), kk[0].shape[1] * (len(kk)-1)) )
k_column = np.vstack( kk[1:] )
base = sp.sparse.csr_matrix( np.hstack(
[ k_column, zeros ]
) )
KK = sp.sparse.vstack([row,base]).todense()
return KK
else:
return kk[0]
def determine_gammas_from(self, data):
g_samples = data.copy()
np.random.shuffle(g_samples)
g_samples = g_samples[:self.gamma_samples]
g_diff = np.abs( g_samples.reshape(g_samples.shape[0],1,g_samples.shape[1]) - g_samples.reshape(1,g_samples.shape[0],g_samples.shape[1]) )
g_diff = g_diff.reshape(g_samples.shape[1]*g_samples.shape[0]**2)
g_percentiles = np.arange(self.gamma_quantile / 2,100,self.gamma_quantile).astype('float')
gammas = []
for i in g_percentiles:
gammas.append( sp.stats.stats.scoreatpercentile(g_diff, i) )
return np.array(gammas)
