def __init__(self, indices, features, data_handler=None, data=None):
self.indices = np.int32(indices)
self.indices_d = gpu.to_gpu(self.indices)
self.results_d = gpu.empty_like(self.indices_d)
if data_handler:
self.data_handler = data_handler
else:
self.data_handler = DataHandlerCUDA(data)
class SubsetCUDA(SubsetBase): def __init__(self, indices, features, data_handler=None, data=None): self.indices = np.int32(indices) self.indices_d = gpu.to_gpu(self.indices) self.results_d = gpu.empty_like(self.indices_d) if data_handler: self.data_handler = data_handler else: self.data_handler = DataHandlerCUDA(data) def pure(): """Determine if subset is pure. Returns label of pure item if pure and None otherwise""" return def majority_label(): """Return label of majority item.""" return def split(feature, threshold): self.data_handler.split(self.indices_d, self.results_d, feature, threshold) #items that are above the threshold are positive, items that are below are negative #the number represents the class. if it is -4 then the review has rating 4 and it #did was below the threshold results_gpu = self.results_d.get() #create a new subset object. all those that are left_indices = [i for i in self.indices if results_gpu[i] > 0] #items that are above the threshold right_indices = [i for i in self.indices if results_gpu[i] < 0] #sanity check assert len(left_indices) + len(right_indices) == len(results_gpu) left_subset = Subset(self.data_handler, left_indices) right_subset = Subset(self.data_handler, right_indices) return left_subset, right_subset def get_subsets(feature, threshold): left, right = self.data_handler.split(self.indices, feature, threshold) left_subset = Subset(left, data_handler=self.data_handler) right_subset = Subset(right, data_handler=self.data_handler) return left_subset, right_subset def test_purity_on_split(feature, threshold): purity = 0 return purity
class SubsetCUDA(SubsetBase):
def __init__(self, indices, features, data_handler=None, data=None):
self.indices = np.int32(indices)
self.indices_d = gpu.to_gpu(self.indices)
self.results_d = gpu.empty_like(self.indices_d)
if data_handler:
self.data_handler = data_handler
else:
self.data_handler = DataHandlerCUDA(data)
def pure():
"""Determine if subset is pure. Returns label of pure item if pure and None otherwise"""
return
def majority_label():
"""Return label of majority item."""
return
def split(feature, threshold):
self.data_handler.split(self.indices_d, self.results_d, feature,
threshold)
#items that are above the threshold are positive, items that are below are negative
#the number represents the class. if it is -4 then the review has rating 4 and it
#did was below the threshold
results_gpu = self.results_d.get()
#create a new subset object. all those that are
left_indices = [i for i in self.indices if results_gpu[i] > 0
] #items that are above the threshold
right_indices = [i for i in self.indices if results_gpu[i] < 0]
#sanity check
assert len(left_indices) + len(right_indices) == len(results_gpu)
left_subset = Subset(self.data_handler, left_indices)
right_subset = Subset(self.data_handler, right_indices)
return left_subset, right_subset
def get_subsets(feature, threshold):
left, right = self.data_handler.split(self.indices, feature, threshold)
left_subset = Subset(left, data_handler=self.data_handler)
right_subset = Subset(right, data_handler=self.data_handler)
return left_subset, right_subset
def test_purity_on_split(feature, threshold):
purity = 0
return purity
def __init__(self, indices, features, data_handler=None, data=None): self.indices = np.int32(indices) self.indices_d = gpu.to_gpu(self.indices) self.results_d = gpu.empty_like(self.indices_d) if data_handler: self.data_handler = data_handler else: self.data_handler = DataHandlerCUDA(data)
