optimizer.py 文件源码

def get_learning_rate_policy_callback(lr_params):
    if isinstance(lr_params, numbers.Real):
                # If argument is real number, set policy to fixed and use given value as base_lr
        lr_params = {'name': 'fixed', 'base_lr': lr_params}

    # Check if lr_params contains all required parameters for selected policy.
    if lr_params['name'] not in lrp.lr_policies:
        raise NotImplementedError("Learning rate policy {lr_name} not supported."
                                  "\nSupported policies are: {policies}".format(
                                      lr_name=lr_params['name'],
                                      policies=lrp.lr_policies.keys())
                                  )
    elif all([x in lr_params.keys() for x in lrp.lr_policies[lr_params['name']]['args']]):
        return lrp.lr_policies[lr_params['name']]['obj'](lr_params)
    else:
        raise ValueError("Too few arguments provided to create policy {lr_name}."
                         "\nGiven: {lr_params}"
                         "\nExpected: {lr_args}".format(
                             lr_name=lr_params['name'],
                             lr_params=lr_params.keys(),
                             lr_args=lrp.lr_policies[lr_params['name']])
                         )