Construct FL course when server does not have data #236
Conversation
| else: | ||
| raise ValueError('Data {} not found.'.format(config.data.type)) | ||
|
|
||
| if config.federate.mode.lower() == 'standalone': | ||
| return data, modified_config | ||
| else: | ||
| # Invalid data_idx | ||
| if config.distribute.data_idx not in data.keys(): | ||
| if config.distribute.data_idx == -1: |
There was a problem hiding this comment.
When data.type is None or "" while distribute.data_idx is given, we should assert in configs.
There was a problem hiding this comment.
When date.type is None or "", the distribute.data_idx does not work. IMO, we can provide a WARNING but not an assertion here.
| @@ -49,7 +49,12 @@ def extend_fl_setting_cfg(cfg): | |||
| cfg.distribute.client_port = 50050 | |||
| cfg.distribute.role = 'client' | |||
| cfg.distribute.data_file = 'data' | |||
There was a problem hiding this comment.
The data-related keyword distribute.data_file in distributed mode seems to be the same meaning as data.root.
There was a problem hiding this comment.
data.root is a file directory and distribute.data_file includes the filename here
| if model_config.type.lower() in ['vmfnet', 'hmfnet']: | ||
| return data['train'].n_col if model_config.type.lower( | ||
| ) == 'vmfnet' else data['train'].n_row | ||
| elif model_config.type.lower() in [ |
There was a problem hiding this comment.
For graph-level tasks, data is {train/val/test: DataLoader}.
For testing: python federatedscope/main.py --cfg federatedscope/gfl/baseline/fedavg_gnn_minibatch_on_multi_task.yaml
There was a problem hiding this comment.
I have fixed this issue, thx!
|
I have rebased the thread to the master and resolved the conflicts. @rayrayraykk @yxdyc |
As the title says. When the server does not have data:
model.input_shapeto tell what's the input shape (a tuple) to get the model if necessary;When the server does not have data in simulation, users can set
data.type=''. (Please see the examples in scripts/run_distributed_lr.sh)Please @rayrayraykk check
get_shape_from_datain federatescope/core/auxiliaries/model_builder.py, thx!