import numpy
from chainer import cuda
from chainer import function
from chainer import utils
from chainer.utils import type_check
class Log1p(function.Function):
@property
def label(self):
return 'log1p'
def check_type_forward(self, in_types):
type_check.expect(in_types.size() == 1)
type_check.expect(in_types[0].dtype.kind == 'f')
def forward_cpu(self, x):
return utils.force_array(numpy.log1p(x[0])),
def forward_gpu(self, x):
return cuda.cupy.log1p(x[0]),
def backward(self, x, gy):
return utils.force_array(gy[0] / (x[0] + x[0].dtype.type(1.0))),
[docs]def log1p(x):
"""Elementwise natural logarithm plus one function."""
return Log1p()(x)