Add the initialization parameters OF LR FTRL

[tbkken]

Angel's model, after adding features, does not currently support incremental training. For example, the model has n features before, now add m features, and now there are a total of n+m features. If possible, trouble provides an API that initializes the LR FTRL initialization parameters, initializes a model with the parameters of another model, and then starts training, which makes it more suitable for incremental training scenarios.

[rachelsunrh]

Angel's model, after adding features, does not currently support incremental training. For example, the model has n features before, now add m features, and now there are a total of n+m features. If possible, trouble provides an API that initializes the LR FTRL initialization parameters, initializes a model with the parameters of another model, and then starts training, which makes it more suitable for incremental training scenarios.

now in our model, if your features are increasing, you can set dim a large number or '-1' in training, and set the load path not null while incremental training