make background refresh optional #1655
Conversation
| @@ -26,6 +26,11 @@ namespace reinforcement_learning | |||
| new live_model_impl(config, fn, err_context, trace_factory, t_factory, m_factory, sender_factory)); | |||
| } | |||
|
|
|||
| live_model::live_model(live_model&& model) { | |||
| _pimpl = std::move(model._pimpl); | |||
There was a problem hiding this comment.
nit: destructor is trivial now but that might change. would a swap be better for these variables?
| @@ -208,7 +208,7 @@ namespace reinforcement_learning { | |||
| /** | |||
| * @brief Default move constructor for live model object. | |||
| */ | |||
| live_model(live_model&&) = default; | |||
| live_model(live_model&&); | |||
There was a problem hiding this comment.
Now that move constructor is defined, move assignment should be as well
| _model_download.reset(new m::model_downloader(ptransport, &_data_cb, _trace_logger.get())); | ||
| return _bg_model_proc->init(_model_download.get(), status); | ||
| } | ||
| else { |
There was a problem hiding this comment.
It looks like we cannot update model later. Maybe it is better to add update_model method in live_model interface?
|
@skofsky: Are there more updates coming for this PR, or is it ready for last round of reviews + merge? |
|
The reinforcement_learning directory has moved to its own repo. Can you please reopen this PR there? https://github.com/VowpalWabbit/reinforcement_learning |
Thanks, I will close this and open a new pull request. |
Adding configuration option "model.backgroundrefresh" that can be set to false to disable background model loading. When background model loading is disabled the model is loaded during the live_model::init() call and never updated again.