officehome_tgt_ours_pb_teachaug_directed.yaml

# @package _global_

# to execute this experiment run:
# python run.py experiment=example_simple.yaml

defaults:
  - override /trainer: default.yaml
  - override /model: ours_pbta_directed.yaml
  - override /data: tgt_teachaug_train_datamodule.yaml
  - override /callbacks: default.yaml
  - override /logger: many_loggers.yaml

# all parameters below will be merged with parameters from default configurations set above
# this allows you to overwrite only specified parameters

# name of the run determines folder name in logs
# it's also accessed by loggers

seed: 1

source_task: "Ar"
target_task: "Cl"

data:
  dataset:
    _target_: src.data.components.officehome.OfficeHome
    root: ${paths.data_dir}/officehome
    task: ${target_task}
    download: True
  batch_size: 64
  num_workers: 8
  pin_memory: True
  M: 1
  K: 1

model:
  net:
    num_classes: 65
  pretrained_path: ${paths.root_dir}/pretrained_model/officehome/${source_task}_seed${seed}_resnet50.pth

  optimizer_args:
    lr: 0.003
    momentum: 0.9
    nesterov: True
    weight_decay: 0.0005
  scheduler_args:
    type: "exponential"
    power: 0.0
    gamma: 10.0
  aug_optimizer_args:
    lr: 0.0005
    weight_decay: 0.01

  lambda_neg: 0.75
  lambda_neg_2: 0.75
  lambda_aug: 1.0
  lambda_strong: 0.5

  net_momentum: 0.99

  warmup_epoch: 0
  interval_epoch: 4

  init_aug: "autoaug"
  init_aug_prob: 0.99

  K: 3
  M: 3
  MM: 3

  optimize_params: "back_bottle"

trainer:
  max_epochs: 133 # including data augmentation epoch

task_name: OfficeHome_${source_task}2${target_task}
method_name: ours_pbta_directed

logger:
  mlflow:
    experiment_name: ${task_name}
    run_name: ${method_name}