--help
Available parameters (--help)
| Parameter | Description | Default Value |
|---|---|---|
| Required Arguments | ||
| --num_random_steps NUM_RANDOM_STEPS | Number of random steps to start with. | 20 |
| --max_eval MAX_EVAL | Maximum number of evaluations. | - |
| --run_program RUN_PROGRAM [RUN_PROGRAM ...] | A program that should be run. Use, for example, %(x) for the parameter named x. | - |
| --experiment_name EXPERIMENT_NAME | Name of the experiment. | - |
| --mem_gb MEM_GB | Amount of RAM for each worker in GB. | 1 |
| Required Arguments That Allow A Choice | ||
| --parameter PARAMETER [PARAMETER ...] | Experiment parameters in the formats: - <NAME> range <NAME> <LOWER BOUND> <UPPER BOUND> (<INT, FLOAT>) (<log_scale: true or false, default false>) - <NAME> fixed <NAME> <VALUE> - <NAME> choice <NAME> <Comma-separated list of values> |
- |
| --continue_previous_job CONTINUE_PREVIOUS_JOB | Continue from a previous checkpoint, use run-dir as argument. | - |
| Optional | ||
| --generation_strategy | A string containing the generation_strategy | None |
| --minkowski_p | Minkowski order of distance (default: 2), needs to be larger than 0 | 2 |
| --signed_weighted_euclidean_weights | A comma-seperated list of values for the signed weighted euclidean distance. Needs to be equal to the number of results. Else, default will be 1.' | 1 for every parameter |
| --result_names | Name of hyperparameters. Example --result_names result1=max result2=min result3. Default: result=min, or result=max when --maximize is set. Default is min. | result= |
| --decimalrounding INT | Set the number of decimal rounding places, like how many places decimal numbers should be rounded to. | 4 |
| --occ_type | Set the type of the optimization-with-combined-criteria collection function. Options are geometric, signed harmonic, and euclid. | euclid |
| --max_parallelism | Set how the ax max parallelism flag should be set. Possible options: None, max_eval, num_parallel_jobs, twice_max_eval, twice_num_parallel_jobs, max_eval_times_thousand_plus_thousand, or any positive integer. | max_eval_times_thousand_plus_thousand |
| --should_deduplicate | Try to de-duplicate ARMs | False |
| --workdir | Working directory (doesn't yet work! TODO) | False |
| --disable_tqdm | Disable the TQDM progress bar | False |
| --live_share | Automatically live-share the current optimization run automatically | False |
| --checkout_to_latest_tested_version | Automatically checkout to latest version that was tested in the CI pipeline | False |
| --exclude "taurusi8009,taurusi8010" | A comma seperated list of values of excluded nodes. | None |
| --maximize | Maximize instead of minimize (which is default). | False |
| --experiment_constraints EXPERIMENT_CONSTRAINTS [EXPERIMENT_CONSTRAINTS ...] | Constraints for parameters. Example: x + y <= 2.0. | - |
| --stderr_to_stdout | Redirect stderr to stdout for subjobs. | False |
| --run_dir RUN_DIR | Directory in which runs should be saved. | runs |
| --seed SEED | Seed for random number generator. | - |
| --enforce_sequential_optimization | Enforce sequential optimization. | False |
| --verbose_tqdm | Show verbose tqdm messages. | False |
| --hide_ascii_plots | Hide ASCII-plots. | False |
| --model MODEL | Use special models for nonrandom steps. Valid models are: SOBOL, GPEI, FACTORIAL, SAASBO, FULLYBAYESIAN, LEGACY_BOTORCH, BOTORCH_MODULAR, UNIFORM, BO_MIXED. | BOTORCH_MODULAR |
| --gridsearch | Enable gridsearch. | False |
| --occ | Enable optimization with combined criteria. | False |
| --show_sixel_scatter | Show sixel graphics of scatter plot in the end. | False |
| --show_sixel_general | Show sixel graphics of general plot in the end. | False |
| --show_sixel_trial_index_result | Show sixel graphics of trial_index_result plot in the end. | False |
| --follow | Automatically follow log file of sbatch. | False |
| --send_anonymized_usage_stats | Send anonymized usage stats. | False |
| --ui_url UI_URL | Site from which the OO-run was called. | - |
| --root_venv_dir ROOT_VENV_DIR | Where to install your modules to ($root_venv_dir/.omniax_...) | $HOME |
| --main_process_gb (INT) | Amount of RAM the main process should have. Only relevant on SLURM systems | 8 |
| --pareto_front_confidence (INT) | Confidence for pareto-front-plotting (between 0 and 1) | 0.95 |
| --max_nr_of_zero_results (INT) | Max. nr of successive zero results by ax_client.get_next_trials() before the search space is seen as exhausted. | 50 |
| --abbreviate_job_names | Abbreviate pending job names (r = running, p = pending, u = unknown, c = cancelling) | False |
| --disable_search_space_exhaustion_detection | Disables automatic search space reduction detection | False |
| --orchestrator_file PATH/TO/orchestrator.YAML | An orchestrator file. | None |
| SLURM | ||
| --worker_timeout WORKER_TIMEOUT | Timeout for slurm jobs (i.e., for each single point to be optimized). | 30 |
| --num_parallel_jobs NUM_PARALLEL_JOBS | Number of parallel slurm jobs (only used when SLURM is installed). | 20 |
| --auto_exclude_defective_hosts | Run a Test if you can allocate a GPU on each node and if not, exclude it since the GPU driver seems to be broken somehow. | False |
| --slurm_use_srun | Using srun instead of sbatch. Learn more | False |
| --time TIME | Time for the main job. | - |
| --partition PARTITION | Partition to be run on. | - |
| --reservation RESERVATION | Reservation. Learn more | - |
| --force_local_execution | Forces local execution even when SLURM is available. | False |
| --slurm_signal_delay_s SLURM_SIGNAL_DELAY_S | When the workers end, they get a signal so your program can react to it. Default is 0, but set it to any number of seconds you wish your program to be able to react to USR1. | 0 |
| --nodes_per_job NODES_PER_JOB | Number of nodes per job due to the new alpha restriction. | 1 |
| --cpus_per_task CPUS_PER_TASK | CPUs per task. | 1 |
| --account ACCOUNT | Account to be used. Learn more | - |
| --gpus GPUS | Number of GPUs. | 0 |
| Installing | ||
| --run_mode | Either local or docker. | local |
| Debug | ||
| --verbose | Verbose logging. | False |
| --debug | Enable debugging. | False |
| --no_sleep | Disables sleeping for fast job generation (not to be used on HPC). | False |
| --tests | Run simple internal tests. | False |
| --raise_in_eval | Raise a signal in eval (only useful for debugging and testing) | False |
| --run_tests_that_fail_on_taurus | Run tests on Taurus that usually fail | False |
| --show_worker_percentage_table_at_end | Show a table of percentage of usage of max worker over time. | False |
| Config | ||
| --config_json | Allows a path to a json file containing all parameters for the run. | - |
| --config_toml | Allows a path to a toml file containing all parameters for the run. | - |
| --config_yaml | Allows a path to a yaml file containing all parameters for the run. | - |