detoxai.methods

Subpackages

• detoxai.methods.clarcs
  • Submodules
  • detoxai.methods.clarcs.aclarc submodule
    • ACLARC
      • ACLARC.apply_model_correction()
  • detoxai.methods.clarcs.clarc submodule
    • require_activations_and_cav()
    • CLARC
      • CLARC.extract_activations()
      • CLARC.compute_cavs()
      • CLARC.apply_model_correction()
  • detoxai.methods.clarcs.hooks submodule
    • stabilize()
    • mass_mean_probe_hook()
    • add_mass_mean_probe_hook()
    • clarc_hook()
    • add_clarc_hook()
  • detoxai.methods.clarcs.pclarc submodule
    • PCLARC
      • PCLARC.apply_model_correction()
  • detoxai.methods.clarcs.rrclarc submodule
    • RRMaskingPattern
      • RRMaskingPattern.MAX_LOGIT
      • RRMaskingPattern.TARGET_LOGIT
      • RRMaskingPattern.ALL_LOGITS
      • RRMaskingPattern.ALL_LOGITS_RANDOM
      • RRMaskingPattern.LOGPROBS
    • RRLossType
      • RRLossType.L2
      • RRLossType.L1
      • RRLossType.COSINE
    • RRCLARC
      • RRCLARC.apply_model_correction()
      • RRCLARC.rr_clarc_hook()
      • RRCLARC.masked_criterion()
      • RRCLARC.rr_loss()
      • RRCLARC.modified_training_step()
  • Module contents
• detoxai.methods.leace
  • Submodules
  • detoxai.methods.leace.leace submodule
    • LEACE
      • LEACE.extract_activations()
      • LEACE.apply_model_correction()
      • LEACE.add_clarc_hook()
  • Module contents
• detoxai.methods.other
  • Submodules
  • detoxai.methods.other.fine_tune submodule
    • FineTune
      • FineTune.apply_model_correction()
  • Module contents
• detoxai.methods.posthoc
  • Submodules
  • detoxai.methods.posthoc.naive_threshold submodule
    • NaiveThresholdOptimizer
      • NaiveThresholdOptimizer.apply_model_correction()
  • detoxai.methods.posthoc.posthoc_base submodule
    • PosthocBase
      • PosthocBase.apply_model_correction()
  • detoxai.methods.posthoc.reject_option_classification submodule
    • ROCModelWrapper
      • ROCModelWrapper.forward()
    • RejectOptionClassification
      • RejectOptionClassification.apply_model_correction()
  • Module contents
• detoxai.methods.savani
  • Submodules
  • detoxai.methods.savani.adversarial submodule
    • SavaniAFT
      • SavaniAFT.apply_model_correction()
      • SavaniAFT.fair_loss()
      • SavaniAFT.get_critic()
  • detoxai.methods.savani.lay_wis_opt submodule
    • SavaniLWO
      • SavaniLWO.apply_model_correction()
      • SavaniLWO.objective_LWO()
      • SavaniLWO.flatten_select()
      • SavaniLWO.unflatten()
  • detoxai.methods.savani.random_perturbation submodule
    • SavaniRP
      • SavaniRP.apply_model_correction()
  • detoxai.methods.savani.savani_base submodule
    • SavaniBase
      • SavaniBase.apply_model_correction()
      • SavaniBase.optimize_tau()
      • SavaniBase.objective_thresh()
      • SavaniBase.phi_torch()
      • SavaniBase.apply_hook()
      • SavaniBase.get_pred_true_prot()
      • SavaniBase.check_layer_name_exists()
      • SavaniBase.sample_batch()
      • SavaniBase.initialize_dataloader()
  • detoxai.methods.savani.utils submodule
    • phi_torch()
  • detoxai.methods.savani.zhang submodule
    • ZhangM
      • ZhangM.apply_model_correction()
      • ZhangM.get_critic()
  • Module contents

Submodules

detoxai.methods.model_correction submodule

class detoxai.methods.model_correction.ModelCorrectionMethod(model: Module | LightningModule, experiment_name: str, device: str)

Bases: ABC

abstractmethod apply_model_correction() → None

get_model() → Module

get_lightning_model() → LightningModule

remove_hooks() → None

detoxai.methods.utils submodule

Module contents