jmacglashan opened a new issue #7312: gluon Trainer does not accept Optimizer object, only str URL: https://github.com/apache/incubator-mxnet/issues/7312 The gluon Trainer is supposed to accept either a string name of an Optimizer, or an actual Optimizer object. However, if you try to pass it an Optimizer object it does not handle it correctly, expecting a string. Modifying the Trainer `_init_optimizer` method as follows fixes it, although I'm not sure if this is the way it wants to be handled: ``` def _init_optimizer(self, optimizer, optimizer_params): if isinstance(optimizer, opt.Optimizer): self._optimizer = optimizer else: self._optimizer = opt.create(optimizer, **optimizer_params) lr_mult = {} wd_mult = {} for i, param in enumerate(self._params): lr_mult[i] = param.lr_mult wd_mult[i] = param.wd_mult self._optimizer.set_lr_mult(lr_mult) self._optimizer.set_wd_mult(wd_mult) self._updaters = [opt.get_updater(self._optimizer) \ for _ in self._context ``` I would also recommend making the `optimizer_params` argument of the Trainer constructor optional since they are not needed if you provide a full Optimizer object. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
With regards, Apache Git Services
