Refactoring downstream to user

experiments/example_downstream/config.toml

@@ -1,6 +1,6 @@
-[downstream]
+[train_user]
 input = "experiments/example_attack/labels.npy"
-downstream_model = "r32p"
+user_model = "r32p"
 trainer = "sgd"
 dataset = "cifar"
 source_label = 9

experiments/example_downstream_soft/config.toml

@@ -1,7 +1,7 @@
-[downstream]
+[train_user]
 input = "experiments/example_attack/labels.npy"
 true = "experiments/example_attack/true.npy"
-downstream_model = "r32p"
+user_model = "r32p"
 trainer = "sgd"
 dataset = "cifar"
 source_label = 9

experiments/example_precomputed/config.toml

@@ -1,6 +1,6 @@
-[downstream]
+[train_user]
 input = "precomputed/cifar/r32p/1xs/1500.npy"
-downstream_model = "r32p"
+user_model = "r32p"
 trainer = "sgd"
 dataset = "cifar"
 source_label = 9

experiments/example_precomputed_mix/config.toml

@@ -1,6 +1,6 @@
-[downstream]
+[train_user]
 input = "precomputed/cifar/r32p/1xs/1500.npy"
-downstream_model = "vit-pretrain"
+user_model = "vit-pretrain"
 trainer = "sgd"
 dataset = "cifar"
 source_label = 9
@@ -10,6 +10,6 @@ output_path = "experiments/example_precomputed_mix/"
 soft = false
 alpha = 0.0
 
-[downstream.optim_kwargs]
+[train_user.optim_kwargs]
 lr = 0.01
 weight_decay = 0.0002

modules/base_utils/datasets.py

@@ -644,7 +644,7 @@ def get_matching_datasets(
  return train_dataset, distill_dataset, test_dataset, poison_test_dataset, mtt_dataset
 
 
-def construct_downstream_dataset(distill_dataset, labels, mask=None, target_label=None, include_labels=False):
+def construct_user_dataset(distill_dataset, labels, mask=None, target_label=None, include_labels=False):
  dataset = LabelWrappedDataset(distill_dataset, labels, include_labels)
  return dataset
 

modules/downstream/run_module.py → modules/train_user/run_module.py

@@ -11,7 +11,7 @@
 import numpy as np
 
 from modules.base_utils.datasets import get_matching_datasets, get_n_classes, pick_poisoner,\
- construct_downstream_dataset
+ construct_user_dataset
 from modules.base_utils.util import extract_toml, get_train_info,\
  mini_train, load_model, needs_big_ims, softmax
 
@@ -27,7 +27,7 @@ def run(experiment_name, module_name, **kwargs):
 
  args = extract_toml(experiment_name, module_name)
 
- downstream_model_flag = args["downstream_model"]
+ user_model_flag = args["user_model"]
  trainer_flag = args["trainer"]
  dataset_flag = args["dataset"]
  poisoner_flag = args["poisoner"]
@@ -54,14 +54,14 @@ def run(experiment_name, module_name, **kwargs):
  Path(output_path).mkdir(parents=True, exist_ok=True)
 
 
- print(f"{downstream_model_flag=} {clean_label=} {target_label=} {poisoner_flag=}")
+ print(f"{user_model_flag=} {clean_label=} {target_label=} {poisoner_flag=}")
 
  print("Building datasets...")
  poisoner = pick_poisoner(poisoner_flag,
  dataset_flag,
  target_label)
 
- big_ims = needs_big_ims(downstream_model_flag)
+ big_ims = needs_big_ims(user_model_flag)
  _, distillation, test, poison_test, _ =\
  get_matching_datasets(dataset_flag, poisoner, clean_label, big=big_ims)
 
@@ -77,19 +77,19 @@ def run(experiment_name, module_name, **kwargs):
  if not soft:
  labels_d = labels_d.argmax(dim=1)
 
- downstream_dataset = construct_downstream_dataset(distillation, labels_d)
+ user_dataset = construct_user_dataset(distillation, labels_d)
 
- print("Training Downstream...")
+ print("Training User Model...")
  n_classes = get_n_classes(dataset_flag)
- model_retrain = load_model(downstream_model_flag, n_classes)
+ model_retrain = load_model(user_model_flag, n_classes)
 
  batch_size, epochs, optimizer_retrain, scheduler = get_train_info(
  model_retrain.parameters(), trainer_flag, batch_size, epochs, optim_kwargs, scheduler_kwargs
  )
 
  model_retrain, clean_metrics, poison_metrics = mini_train(
  model=model_retrain,
- train_data=downstream_dataset,
+ train_data=user_dataset,
  test_data=[test, poison_test.poison_dataset],
  batch_size=batch_size,
  opt=optimizer_retrain,

schemas/downstream.toml → schemas/train_user.toml

@@ -1,14 +1,14 @@
 ### 
 # TODO
-# downstream schema
+# train_user schema
 # Configured to poison and train and distill a set of model on any of the datasets.
 # Outputs the .pth of a distileld model
 ###
 
-[downstream]
+[train_user]
 input = "TODO"
 output_path = "string: Path to .pth file."
-downstream_model = "string: (r32p, r18, r18_tin, vgg, vgg_pretrain, vit_pretrain). For ResNets, VGG-19s, and ViTs"
+user_model = "string: (r32p, r18, r18_tin, vgg, vgg_pretrain, vit_pretrain). For ResNets, VGG-19s, and ViTs"
 dataset = "string: (cifa r / cifar_100 / tiny_imagenet). For CIFAR-10, CIFAR-100 and Tiny Imagenet datasets"
 trainer = "string: (sgd / adam). Specifies optimizer. "
 source_label = "int: {0,1,...,9}. Specifies label to mimic"