Exp fix

benchmark/run_param.py

@@ -8,7 +8,7 @@
 import uuid
 from copy import deepcopy
 
-from pyg_spectral.nn import get_model_regi, get_conv_regi
+from pyg_spectral.nn import get_model_regi, get_conv_subregi
 from pyg_spectral.nn.parse_args import compose_param
 from pyg_spectral.utils import CallableDict
 
@@ -45,27 +45,45 @@ def __init__(self, data_loader, model_loader, args, res_logger = None):
         }[self.model_loader.get_trn(args)]
 
     def _get_suggest(self, trial, key):
+        r"""Get the suggested value and format of the hyperparameter.
+
+        Args:
+            trial (optuna.Trial): The trial object.
+            key (str): The hyperparameter key in :obj:`args.param`.
+        Returns:
+            val (Any | list[Any]): The suggested value.
+            fmt (Callable): The format of the suggested value.
+        """
         def parse_param(val):
+            r"""From :class:`ParamTuple` to suggest trial value and format.
+            Args:
+                val (ParamTuple | list[ParamTuple]): registry entry.
+            Returns:
+                val (Any | list[Any]): The suggested value.
+                fmt (Callable): The format of the suggested value
+            """
             if isinstance(val, list):
                 fmt = val[0][-1]
                 val = [getattr(trial, 'suggest_'+func)(key+'-'+str(i), *fargs, **fkwargs) for i, (func, fargs, fkwargs, _) in enumerate(val)]
                 return val, fmt
             func, fargs, fkwargs, fmt = val
             return getattr(trial, 'suggest_'+func)(key, *fargs, **fkwargs), fmt
 
-        # Alias compose models
+        # Alias param for compose models
         if (self.args.model in compose_param and
             self.model_loader.conv_repr in compose_param[self.args.model] and
             key in compose_param[self.args.model][self.model_loader.conv_repr]):
             return parse_param(compose_param[self.args.model][self.model_loader.conv_repr](key, self.args))
 
+        # Param of trainer and model level
         single_param = SingleGraphLoader_Trial.param | ModelLoader_Trial.param | self.trn_cls.param
         single_param = CallableDict(single_param)
         single_param |= get_model_regi(self.args.model, 'param')
         if key in single_param:
             return parse_param(single_param(key, self.args))
 
-        return parse_param(get_conv_regi(self.args.conv, 'param')(key, self.args))
+        # Param of conv level
+        return parse_param(get_conv_subregi(self.args.conv, 'param', key, self.args))
 
     def __call__(self, trial):
         args = deepcopy(self.args)

benchmark/scripts/runfb-iter.sh

@@ -45,16 +45,14 @@ PARLIST="dropout_lin,lr_lin,wd_lin"
 PARLIST="normg,dropout_conv,$PARLIST"
     # Linear
     python run_param.py  --data $data --model $model --conv AdjiConv --param $PARLIST "${ARGS_P[@]}" \
-        --theta_scheme ones --beta 1.0
+        --beta 1.0
     python run_single.py --data $data --model $model --conv AdjiConv "${ARGS_S[@]}" \
-        --theta_scheme ones --beta 1.0
+        --beta 1.0
 
 PARLIST="beta,$PARLIST"
     # PPR
-    python run_param.py  --data $data --model $model --conv AdjResConv --param $PARLIST "${ARGS_P[@]}" \
-        --theta_scheme appr
-    python run_single.py --data $data --model $model --conv AdjResConv "${ARGS_S[@]}" \
-        --theta_scheme appr
+    python run_param.py  --data $data --model $model --conv AdjResConv --param $PARLIST "${ARGS_P[@]}"
+    python run_single.py --data $data --model $model --conv AdjResConv "${ARGS_S[@]}"
 
 : '
 # ========== PyG

pyg_spectral/nn/__init__.py

@@ -1,4 +1,4 @@
 from .parse_args import (
-    get_model_regi, get_conv_regi,
+    get_model_regi, get_conv_regi, get_conv_subregi,
     get_nn_name, set_pargs
 )

pyg_spectral/nn/models/acm_gnn.py

@@ -31,6 +31,7 @@ class ACMGNN(BaseNN):
     # FEATURE: separate arch
     name = 'Iterative'
     conv_name = lambda x, args: '-'.join([x, args.theta_scheme])
+    pargs = ['theta_scheme']
 
     def init_conv(self,
         conv: str,
@@ -83,6 +84,7 @@ class ACMGNNDec(BaseNN):
     # FEATURE: separate arch
     name = 'DecoupledVar'
     conv_name = lambda x, args: '-'.join([x, args.theta_scheme])
+    pargs = ['theta_scheme']
 
     def init_conv(self,
         conv: str,

pyg_spectral/nn/models/iterative.py

@@ -124,9 +124,7 @@ def init_conv(self,
 
 class IterativeFixed(Iterative):
     name = 'IterativeFixed'
-    conv_name = lambda x, args: '-'.join([x, args.theta_scheme])
 
 
 class IterativeFixedCompose(IterativeCompose):
     name = 'IterativeFixed'
-    conv_name = lambda x, args: '-'.join([x, args.theta_scheme])

pyg_spectral/nn/parse_args.py

@@ -15,14 +15,14 @@ def update_regi(regi, new_regi):
 model_regi = BaseNN.register_classes()
 model_regi = update_regi(model_regi, model_regi_pyg)
 
-conv_regi  = CallableDict.to_callableVal(conv_regi, ['pargs_default', 'param'])
+conv_regi  = CallableDict.to_callableVal(conv_regi, reckeys=['pargs_default', 'param'])
 r'''Fields:
     * name (CallableDict[str, str]): Conv class logging path name.
     * pargs (CallableDict[str, list[str]]): Conv arguments from argparse.
     * pargs_default (dict[str, CallableDict[str, Any]]): Default values for model arguments. Not recommended.
     * param (dict[str, CallableDict[str, ParamTuple]]): Conv parameters to tune.
 '''
-model_regi = CallableDict.to_callableVal(model_regi, ['pargs_default', 'param'])
+model_regi = CallableDict.to_callableVal(model_regi, reckeys=['pargs_default', 'param'])
 r'''Fields:
     name (CallableDict[str, str]): Model class logging path name.
     conv_name (CallableDict[str, Callable[[str, Any], str]]): Wrap conv logging path name.
@@ -41,6 +41,11 @@ def update_regi(regi, new_regi):
         'ACMConv-2-low-high':     'FBGNNII',
         'ACMConv-1-low-high-id':  'ACMGNNI',
         'ACMConv-2-low-high-id':  'ACMGNNII',},
+    'ACMGNNDec': {
+        'ACMConv-1-low-high':     'FBGNNI',
+        'ACMConv-2-low-high':     'FBGNNII',
+        'ACMConv-1-low-high-id':  'ACMGNNI',
+        'ACMConv-2-low-high-id':  'ACMGNNII',},
     'DecoupledFixedCompose': {
         'AdjiConv,AdjiConv-ones,ones': 'FAGNN',
         'Adji2Conv,Adji2Conv-gaussian,gaussian': 'G2CN',
@@ -104,6 +109,22 @@ def get_conv_regi(conv: str, k: str, args=None) -> str:
     return conv_regi[k](conv, args) if args else conv_regi[k][conv]
 
 
+def get_conv_subregi(conv: str, k: str, pargs: str, args=None) -> str:
+    r"""Getter for calling a sub-CallableDict in :attr:`conv_regi`.
+
+    Args:
+        conv: The name of the convolution.
+        k: The key in :attr:`conv_regi`.
+        pargs: The key in the sub-CallableDict.
+        args: Configuration arguments.
+    Returns:
+        value (str): The value of the sub-CallableDict.
+    """
+    if ',' in conv:
+        return [conv_regi[k][channel](pargs, args) for channel in conv.split(',')]
+    return conv_regi[k][conv](pargs, args) if args else conv_regi[k][conv][pargs]
+
+
 def get_nn_name(model: str, conv: str, args) -> str:
     r"""Parse model+conv name for logging path from argparse input.