def fprop(self, input, output, train=TRAIN): #util.log_info("%s %s", input.shape, output.shape) gpu_copy_to(input, output) if PFout: #if True: print_matrix(output, self.name)
def fprop(self, input, output, train=TRAIN): max = gpuarray.zeros((1, self.batch_size), dtype=np.float32) col_max_reduce(max, input) add_vec_to_cols(input, max, output, alpha=-1) eltwise_exp(output) sum = gpuarray.zeros(max.shape, dtype=np.float32) add_col_sum_to_vec(sum, output, alpha=0) div_vec_to_cols(output, sum) if PFout: print_matrix(output, self.name)
def fprop(self, input, output, train=TRAIN): cudaconv2.localFilterActs(input, self.weight.wt, output, self.img_size, self.outputSize, self.outputSize, -self.padding, self.stride, self.numColor, 1) #util.log_info('%s', output.get().mean()) self.tmp = gpuarray.empty((self.numFilter, self.get_single_img_size() * self.batch_size / self.numFilter), dtype=np.float32) add_vec_to_rows(output, self.bias.wt) if PFout: print_matrix(output, self.name)
def fprop(self, input, output, train=TRAIN): gpu_copy_to(dot(self.weight.wt, input), output) add_vec_to_rows(output, self.bias.wt) if train == TEST: if self.dropRate > 0.0: output *= (1.0 - self.dropRate) else: if self.dropRate > 0.0: self.dropMask = to_gpu(np.random.uniform(0, 1, output.size).astype(np.float32).reshape(output.shape)) bigger_than_scaler(self.dropMask, self.dropRate) gpu_copy_to(output * self.dropMask, output) if PFout: print_matrix(output, self.name)
def test_cifar_loader(): data_dir = '/ssd/nn-data/cifar-10.old/' dp = data.get_by_name('cifar10')(data_dir, [1]) batch_size = 128 data_list = [] for i in range(11000): batch = dp.get_next_batch(batch_size) batch = batch.data.get() data_list.append(batch) if batch.shape[1] != batch_size: break batch = np.concatenate(data_list, axis=1) print_matrix(batch, 'batch')
def test_cifar_loader(): data_dir = '/ssd/nn-data/cifar-10.old/' dp = data.get_by_name('cifar10')(data_dir, [1]) batch_size = 128 data_list = [] for i in range(11000): batch = dp.get_next_batch(batch_size) batch = batch.data.get() data_list.append(batch) if batch.shape[1] != batch_size: break batch = np.concatenate(data_list, axis = 1) print_matrix(batch, 'batch')
def fprop(self, input, output, train=TRAIN): self.neuron.activate(input, output) if PFout: print_matrix(output, self.name)
def fprop(self, input, output, train=TRAIN): self.denom = gpuarray.zeros_like(input) cudaconv2.convResponseNormCrossMap(input, self.denom, output, self.numColor, self.size, self.scaler, self.pow, self.blocked) if PFout: print_matrix(output, self.name)
def fprop(self, input, output, train=TRAIN): cudaconv2.convLocalAvgPool(input, output, self.numColor, self.poolSize, self.start, self.stride, self.outputSize) if PFout: print_matrix(output, self.name)