base_time.yaml

###########
# GENERAL #
###########
import_keys:
  value:
    - dtype
    - problem_factory
    - optimizer_factory
    - model_factory
    - grad_model_factory
    - model_kwargs.activation_factory
    - model_kwargs.normalization_factory
    - grad_model_kwargs.activation_factory
    - grad_model_kwargs.normalization_factory
dtype:
  value: "torch.float"
device:
  value: null # defaults to 'cuda' if available, else 'cpu'
check_train:
  value: true
check_test:
  value: false
cuda_max_mem_train_mb:
  value: null
cuda_max_mem_mb:
  value: null
seed:
  value: 123456789
############
# TRAINING #
############
optimizer_factory:
  value: torch.optim.Adam
train_steps:
  value: null
train_time_h:
  value: 24
batch_size:
  value: 1024
milestones:
  value: 0.5
milestone_metric:
  value: "time"
time_step:
  value: [1.0e-02, 1.0e-03]
lr:
  value: [5.0e-04, 5.0e-05]
sample_repetitions:
  value: 1
save_freq:
  value: 10
##############
# EVALUATION #
##############
visualize:
  value: ["approx", "approx_grad", "plot", "plot_grad", "stats"]
test_freq:
  value: 30
evaluator_kwargs.approx_batch_size:
  value: 131072
evaluator_kwargs.approx_batches:
  value: 10
evaluator_kwargs.grad_reduce:
  value: "norm"
evaluator_kwargs.approx_p:
  value: [1, 2, "infinity"]
evaluator_kwargs.stats_steps:
  value: 30
evaluator_kwargs.stats_grad_thresh:
  value: [0.1, 0.01, 0.001]
evaluator_kwargs.plot_interval_stretch:
  value: 2.0
evaluator_kwargs.plot_time:
  value: null # defaults to [0, T/2, T]