Declares that user is using an experimental feature.
The developer of an API can mark it as experimental by calling this function. When users call experimental APIs,
FutureWarningis issued. The presentation of
FutureWarningis disabled by setting
True, which is
The basic usage is to call it in the function or method we want to mark as experimental along with the API name.
from chainer import utils def f(x): utils.experimental('chainer.foo.bar.f') # concrete implementation of f follows f(1)
... FutureWarning: chainer.foo.bar.f is experimental. The interface can change in the future. ...
We can also make a whole class experimental. In that case, we should call this function in its
class C(): def __init__(self): utils.experimental('chainer.foo.C') C()
... FutureWarning: chainer.foo.C is experimental. The interface can change in the future. ...
If we want to mark
__init__method only, rather than class itself, it is recommended that we explicitly feed its API name.
class D(): def __init__(self): utils.experimental('D.__init__') D()
... FutureWarning: D.__init__ is experimental. The interface can change in the future. ...
Currently, we do not have any sophisticated way to mark some usage of non-experimental function as experimental. But we can support such usage by explicitly branching it.
def g(x, experimental_arg=None): if experimental_arg is not None: utils.experimental('experimental_arg of chainer.foo.g')
Parameters: api_name (str) – The name of an API marked as experimental.