BOAT-Jittor Structure

Core Problem Class

class boat_jit.boat_opt.Problem(config, loss_config)

Bases: object

Jittor-based bi-level optimization Problem, aligned in interface and logic with the Torch version.

build_ll_solver()

build_ul_solver()

check_status()

plot_losses()

run_iter(ll_feed_dict, ul_feed_dict, current_iter)

save_losses(current_iter, ll_loss, ul_loss)

set_track_trajectory(track_traj=True)

Main Subpackages

• boat_jit.dm_ol
  • Submodules
  • boat_jit.dm_ol.dynamical_system
    • DynamicalSystem
      • DynamicalSystem.optimize()
  • boat_jit.dm_ol.sequential_ds
    • SequentialDS
      • SequentialDS.gradient_instances
      • SequentialDS.custom_order
      • SequentialDS.result_store
      • SequentialDS.optimize()
    • makes_functional_dynamical_system()
    • validate_and_adjust_order()
  • boat_jit.dm_ol.di
    • DI
      • DI.optimize()
  • boat_jit.dm_ol.dm
    • DM
      • DM.optimize()
  • boat_jit.dm_ol.gda
    • GDA
      • GDA.optimize()
  • boat_jit.dm_ol.ngd
    • NGD
      • NGD.optimize()
• boat_jit.fo_ol
  • Submodules
  • boat_jit.fo_ol.meso
    • MESO
      • MESO.optimize()
  • boat_jit.fo_ol.pgdo
    • PGDO
      • PGDO.optimize()
  • boat_jit.fo_ol.vfo
    • VFO
      • VFO.optimize()
  • boat_jit.fo_ol.vso
    • VSO
      • VSO.optimize()
• boat_jit.na_ol
  • Submodules
  • boat_jit.na_ol.hyper_gradient
    • HyperGradient
      • HyperGradient.compute_gradients()
  • boat_jit.na_ol.sequential_hg
    • SequentialHG
      • SequentialHG.compute_gradients()
    • makes_functional_na_operation()
    • validate_and_adjust_order()
  • boat_jit.na_ol.cg
    • CG
      • CG.ll_lr
      • CG.dynamic_initialization
      • CG.tolerance
      • CG.K
      • CG.alpha
      • CG.alpha_decay
      • CG.gda_loss
      • CG.compute_gradients()
  • boat_jit.na_ol.fd
    • FD
      • FD.ll_lr
      • FD.dynamic_initialization
      • FD._r
      • FD.alpha
      • FD.alpha_decay
      • FD.gda_loss
      • FD.compute_gradients()
  • boat_jit.na_ol.foa
    • FOA
      • FOA.compute_gradients()
  • boat_jit.na_ol.iad
    • IAD
      • IAD.compute_gradients()
  • boat_jit.na_ol.iga
    • IGA
      • IGA.alpha
      • IGA.alpha_decay
      • IGA.gda_loss
      • IGA.dynamic_initialization
      • IGA.compute_gradients()
  • boat_jit.na_ol.ns
    • NS
      • NS.compute_gradients()
  • boat_jit.na_ol.ptt
    • PTT
      • PTT.compute_gradients()
  • boat_jit.na_ol.rad
    • RAD
      • RAD.compute_gradients()
  • boat_jit.na_ol.rgt
    • RGT
      • RGT.compute_gradients()
• boat_jit.utils
  • Submodules
  • boat_jit.utils.op_utils
    • DynamicalSystemRules
      • DynamicalSystemRules.get_gradient_order()
      • DynamicalSystemRules.set_gradient_order()
    • HyperGradientRules
      • HyperGradientRules.get_gradient_order()
      • HyperGradientRules.set_gradient_order()
    • ResultStore
      • ResultStore.add()
      • ResultStore.clear()
      • ResultStore.get_results()
    • average_grad()
    • cat_list_to_tensor()
    • cg_step()
    • conjugate_gradient()
    • copy_parameter_from_list()
    • custom_grad()
    • get_outer_gradients()
    • grad_unused_zero()
    • l2_reg()
    • list_tensor_matmul()
    • list_tensor_norm()
    • manual_update()
    • neumann()
    • require_model_grad()
    • stop_grads()
    • stop_model_grad()
    • update_grads()
    • update_tensor_grads()

Extension with Operation_Registry

boat_jit.operation_registry.get_registered_operation(name)

Retrieve a registered operation class by name.

boat_jit.operation_registry.register_class(cls)

Register a new operation class to the global registry.