Utils

singa.utils.force_unicode(s)

return string of a bytes ! borrow from onnx :param s: string or bytes

Returns

string

singa.utils.get_output_shape(auto_pad, input_spatial_shape, kernel_spatial_shape, strides_spatial)

return output shape of conv2d or pooling, ! borrow from onnx :param auto_pad: string

Parameters

• kernel_spatial_shape – list[int]

• strides_spatial – list[int]

• output_spatial_shape – list[int]

Returns

list[int

singa.utils.get_padding_shape(pad_mode, input_spatial_shape, kernel_spatial_shape, strides_spatial)

return padding shape of conv2d or pooling, :param pad_mode: string

Parameters

• kernel_spatial_shape – list[int]

• strides_spatial – list[int]

Returns

list[int]

singa.utils.handle_odd_pad_bwd(dx, odd_padding)

handle odd padding mode backward Args:dx

the backward tensor

Args:odd_padding

the odd_padding

Returns

tensor, the output

singa.utils.handle_odd_pad_fwd(x, odd_padding)

handle odd padding mode forward Args:x

the input tensor

Args:odd_padding

the odd_padding

Returns

tensor, the output

singa.utils.post_order_recursive(root, root_t)

return a list by the topological ordering (postorder of Depth-first search) :param root: singa operator

Parameters

root_t – tensor

Returns

deque[int]

singa.utils.re_new_handle(handle, x, is_pool=False)

re-new a handle by useing the new input tensor Args:handle

the handle

Args:x

input tensor

Returns

handle, a new handle

singa.utils.same_pad_shape_check(handle, pad_mode, x)

check the shape is correct for same padding mode Args:handle

the handle

Args:pad_mode

pad_mode

Args:x

input tensor

Returns

tuple, the correct padding(before divide 2)

singa.utils.update_progress(progress, info)

Display progress bar and user info.

Parameters

• progress (float) – progress [0, 1], negative for halt, and >=1 for done.

• info (str) – a string for user provided info to be displayed.