WIP: better management of legacy positional arguments

avalanche/training/supervised/ar1.py

@@ -43,7 +43,7 @@ class AR1(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
+        *,
         criterion: CriterionType = None,
         lr: float = 0.001,
         inc_lr: float = 5e-5,
@@ -169,7 +169,6 @@ def __init__(
         self.replay_mb_size = 0
 
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/cumulative.py

@@ -20,10 +20,10 @@ class Cumulative(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        optimizer: Optimizer = "not_set",
-        criterion: CriterionType = "not_set",
+        *,
+        model: Module,
+        optimizer: Optimizer,
+        criterion: CriterionType,
         train_mb_size: int = 1,
         train_epochs: int = 1,
         eval_mb_size: Optional[int] = None,
@@ -55,7 +55,6 @@ def __init__(
         """
 
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/deep_slda.py

@@ -3,6 +3,7 @@
 
 import os
 import torch
+from torch.nn import Module
 
 from avalanche.training.plugins import SupervisedPlugin
 from avalanche.training.templates import SupervisedTemplate
@@ -29,11 +30,11 @@ class StreamingLDA(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        slda_model="not_set",
-        criterion: CriterionType = "not_set",
-        input_size="not_set",
-        num_classes: int = "not_set",
+        *,
+        slda_model: Module,
+        criterion: CriterionType,
+        input_size: int,
+        num_classes: int,
         output_layer_name: Optional[str] = None,
         shrinkage_param=1e-4,
         streaming_update_sigma=True,
@@ -74,24 +75,13 @@ def __init__(
         if plugins is None:
             plugins = []
 
-        adapt_legacy_args = slda_model == "not_set"
-        slda_model = slda_model if slda_model != "not_set" else args[0]
         slda_model = slda_model.eval()
         if output_layer_name is not None:
             slda_model = FeatureExtractorBackbone(
                 slda_model.to(device), output_layer_name
             ).eval()
 
-        # Legacy positional arguments support (deprecation cycle)
-        if adapt_legacy_args:
-            args = list(args)
-            args[0] = slda_model
-
-        input_size = input_size if input_size != "not_set" else args[2]
-        num_classes = num_classes if num_classes != "not_set" else args[3]
-
         super(StreamingLDA, self).__init__(
-            legacy_positional_args=args,
             model=slda_model,
             optimizer=None,  # type: ignore
             criterion=criterion,

avalanche/training/supervised/der.py

@@ -157,9 +157,9 @@ class DER(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        optimizer: Optimizer = "not_set",
+        *,
+        model: Module,
+        optimizer: Optimizer,
         criterion: CriterionType = CrossEntropyLoss(),
         mem_size: int = 200,
         batch_size_mem: Optional[int] = None,
@@ -211,7 +211,6 @@ def __init__(
             `eval_every` experience or iterations (Default='experience').
         """
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/er_ace.py

@@ -33,9 +33,9 @@ class ER_ACE(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        optimizer: Optimizer = "not_set",
+        *,
+        model: Module,
+        optimizer: Optimizer,
         criterion: CriterionType = CrossEntropyLoss(),
         mem_size: int = 200,
         batch_size_mem: int = 10,
@@ -74,7 +74,6 @@ def __init__(
             `eval_every` epochs or iterations (Default='epoch').
         """
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/er_aml.py

@@ -28,10 +28,10 @@ class ER_AML(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        feature_extractor: Module = "not_set",
-        optimizer: Optimizer = "not_set",
+        *,
+        model: Module,
+        feature_extractor: Module,
+        optimizer: Optimizer,
         criterion: CriterionType = CrossEntropyLoss(),
         temp: float = 0.1,
         base_temp: float = 0.07,
@@ -77,7 +77,6 @@ def __init__(
             `eval_every` epochs or iterations (Default='epoch').
         """
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,
@@ -98,9 +97,7 @@ def __init__(
         )
         self.replay_loader = None
         self.aml_criterion = AMLCriterion(
-            feature_extractor=(
-                feature_extractor if feature_extractor != "not_set" else args[1]
-            ),
+            feature_extractor=feature_extractor,
             temp=temp,
             base_temp=base_temp,
             same_task_neg=same_task_neg,

avalanche/training/supervised/expert_gate.py

@@ -42,9 +42,9 @@ class ExpertGateStrategy(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        optimizer: Optimizer = "not_set",
+        *,
+        model: Module,
+        optimizer: Optimizer,
         criterion: CriterionType = CrossEntropyLoss(),
         train_mb_size: int = 1,
         train_epochs: int = 1,
@@ -109,7 +109,6 @@ def __init__(
         )
 
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/feature_replay.py

@@ -47,9 +47,9 @@ class FeatureReplay(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        model: nn.Module = "not_set",
-        optimizer: Optimizer = "not_set",
+        *,
+        model: nn.Module,
+        optimizer: Optimizer,
         criterion: CriterionType = MaskedCrossEntropy(),
         last_layer_name: str = "classifier",
         mem_size: int = 200,
@@ -67,7 +67,6 @@ def __init__(
         **kwargs
     ):
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/icarl.py

@@ -28,13 +28,13 @@ class ICaRL(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        feature_extractor: Module = "not_set",
-        classifier: Module = "not_set",
-        optimizer: Optimizer = "not_set",
-        memory_size="not_set",
-        buffer_transform="not_set",
-        fixed_memory="not_set",
+        *,
+        feature_extractor: Module,
+        classifier: Module,
+        optimizer: Optimizer,
+        memory_size: int,
+        buffer_transform,
+        fixed_memory: bool,
         train_mb_size: int = 1,
         train_epochs: int = 1,
         eval_mb_size: Optional[int] = None,
@@ -73,16 +73,16 @@ def __init__(
             learning experience.
         """
         model = TrainEvalModel(
-            feature_extractor if feature_extractor != "not_set" else args[0],
-            train_classifier=classifier if classifier != "not_set" else args[1],
+            feature_extractor,
+            train_classifier=classifier,
             eval_classifier=NCMClassifier(normalize=True),
         )
 
         criterion = ICaRLLossPlugin()  # iCaRL requires this specific loss (#966)
         icarl = _ICaRLPlugin(
-            memory_size if memory_size != "not_set" else args[3],
-            buffer_transform if buffer_transform != "not_set" else args[4],
-            fixed_memory if fixed_memory != "not_set" else args[5],
+            memory_size,
+            buffer_transform,
+            fixed_memory,
         )
 
         if plugins is None:
@@ -94,7 +94,6 @@ def __init__(
             plugins += [criterion]
 
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/joint_training.py

@@ -58,10 +58,10 @@ class JointTraining(SupervisedTemplate[TDatasetExperience, TMBInput, TMBOutput])
 
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        optimizer: Optimizer = "not_set",
-        criterion: CriterionType = "not_set",
+        *,
+        model: Module,
+        optimizer: Optimizer,
+        criterion: CriterionType,
         train_mb_size: int = 1,
         train_epochs: int = 1,
         eval_mb_size: int = 1,
@@ -91,7 +91,6 @@ def __init__(
             `eval` is called every `eval_every` epochs and at the end of the
             learning experience."""
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/l2p.py

@@ -31,8 +31,8 @@ class LearningToPrompt(SupervisedTemplate):
 
     def __init__(
         self,
-        *args,
-        model_name: str = "not_set",
+        *,
+        model_name: str,
         criterion: nn.Module = nn.CrossEntropyLoss(),
         train_mb_size: int = 1,
         train_epochs: int = 1,
@@ -99,7 +99,7 @@ def __init__(
         self.lr = lr
         self.sim_coefficient = sim_coefficient
         model = create_model(
-            model_name=model_name if model_name != "not_set" else args[0],
+            model_name=model_name,
             prompt_pool=prompt_pool,
             pool_size=pool_size,
             prompt_length=prompt_length,
@@ -131,7 +131,6 @@ def __init__(
         )
 
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,

avalanche/training/supervised/lamaml.py

@@ -27,9 +27,9 @@
 class LaMAML(SupervisedMetaLearningTemplate):
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        optimizer: Optimizer = "not_set",
+        *,
+        model: Module,
+        optimizer: Optimizer,
         criterion: CriterionType = CrossEntropyLoss(),
         n_inner_updates: int = 5,
         second_order: bool = True,

avalanche/training/supervised/lamaml_v2.py

@@ -24,9 +24,9 @@
 class LaMAML(SupervisedMetaLearningTemplate):
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        optimizer: Optimizer = "not_set",
+        *,
+        model: Module,
+        optimizer: Optimizer,
         criterion: CriterionType = CrossEntropyLoss(),
         n_inner_updates: int = 5,
         second_order: bool = True,

avalanche/training/supervised/mer.py

@@ -52,9 +52,9 @@ def get_batch(self, x, y, t):
 class MER(SupervisedMetaLearningTemplate):
     def __init__(
         self,
-        *args,
-        model: Module = "not_set",
-        optimizer: Optimizer = "not_set",
+        *,
+        model: Module,
+        optimizer: Optimizer,
         criterion: CriterionType = CrossEntropyLoss(),
         mem_size=200,
         batch_size_mem=10,
@@ -88,7 +88,6 @@ def __init__(
 
         """
         super().__init__(
-            legacy_positional_args=args,
             model=model,
             optimizer=optimizer,
             criterion=criterion,