omniopt2 7037__tar.gz → 7076__tar.gz

{omniopt2-7037 → omniopt2-7076}/.omniopt.py

@@ -41,7 +41,7 @@ overwritten_to_random: bool = False
 valid_occ_types: list = ["geometric", "euclid", "signed_harmonic", "signed_minkowski", "weighted_euclid", "composite"]
 joined_valid_occ_types: str = ", ".join(valid_occ_types)
 
-SUPPORTED_MODELS: list = ["SOBOL", "FACTORIAL", "SAASBO", "BOTORCH_MODULAR", "UNIFORM", "BO_MIXED", "RANDOMFOREST", "EXTERNAL_GENERATOR", "PSEUDORANDOM"]
+SUPPORTED_MODELS: list = ["SOBOL", "FACTORIAL", "SAASBO", "BOTORCH_MODULAR", "UNIFORM", "BO_MIXED", "RANDOMFOREST", "EXTERNAL_GENERATOR", "PSEUDORANDOM", "TPE"]
 joined_supported_models: str = ", ".join(SUPPORTED_MODELS)
 
 special_col_names: list = ["arm_name", "generation_method", "trial_index", "trial_status", "generation_node", "idxs"]
@@ -556,6 +556,7 @@ class ConfigLoader:
     follow: bool
     show_generation_and_submission_sixel: bool
     n_estimators_randomforest: int
+    max_num_of_parallel_sruns: int
     checkout_to_latest_tested_version: bool
     load_data_from_existing_jobs: List[str]
     time: str
@@ -659,6 +660,7 @@ class ConfigLoader:
         speed.add_argument('--dont_jit_compile', help='Disable JIT-compiling the model', action='store_true', default=False)
         speed.add_argument('--num_restarts', help='num_restarts option for optimizer_options', type=int, default=20)
         speed.add_argument('--raw_samples', help='raw_samples option for optimizer_options', type=int, default=1024)
+        speed.add_argument('--max_num_of_parallel_sruns', help='Maximal number of parallel sruns', type=int, default=16)
         speed.add_argument('--no_transform_inputs', help='Disable input transformations', action='store_true', default=False)
         speed.add_argument('--no_normalize_y', help='Disable target normalization', action='store_true', default=False)
 
@@ -4941,9 +4943,13 @@ def get_type_short(typename: str) -> str:
     return typename
 
 @beartype
-def parse_single_experiment_parameter_table(classic_params: Union[list, dict]) -> list:
+def parse_single_experiment_parameter_table(classic_params: Optional[Union[list, dict]]) -> list:
     rows: list = []
 
+    if classic_params is None:
+        print_red("parse_single_experiment_parameter_table: classic_param is None")
+        return rows
+
     k = 0
 
     for param in classic_params:
@@ -5130,7 +5136,7 @@ def print_experiment_param_table_to_file(filtered_columns: list, filtered_data:
         print_red(f"Error trying to write file {fn}: {e}")
 
 @beartype
-def print_experiment_parameters_table(classic_param: Union[list, dict]) -> None:
+def print_experiment_parameters_table(classic_param: Optional[Union[list, dict]]) -> None:
     if not classic_param:
         print_red("Cannot determine classic_param. No parameter table will be shown.")
         return
@@ -5161,10 +5167,7 @@ def print_experiment_parameters_table(classic_param: Union[list, dict]) -> None:
     print_experiment_param_table_to_file(filtered_columns, filtered_data)
 
 @beartype
-def print_overview_tables(classic_params: Optional[Union[list, dict]], experiment_parameters: Union[list, dict], experiment_args: dict) -> None:
-    if classic_params is None or len(classic_params) == 0:
-        classic_params = experiment_parameters
-
+def print_overview_tables(classic_params: Optional[Union[list, dict]], experiment_args: dict) -> None:
     print_experiment_parameters_table(classic_params)
 
     print_ax_parameter_constraints_table(experiment_args)
@@ -6567,6 +6570,8 @@ def execute_evaluation(_params: list) -> Optional[int]:
         trial_counter += 1
 
         progressbar_description(["started new job"])
+
+        save_results_csv()
     except submitit.core.utils.FailedJobError as error:
         handle_failed_job(error, trial_index, new_job)
         trial_counter += 1
@@ -7492,6 +7497,15 @@ def create_node(model_name: str, threshold: int, next_model_name: Optional[str])
 
         return node
 
+    if model_name == "TPE":
+        if len(arg_result_names) != 1:
+            print_red(f"Has {len(arg_result_names)} results. TPE currently only supports single-objective-optimization.")
+            my_exit(108)
+
+        node = ExternalProgramGenerationNode(f"python3 {script_dir}/.tpe.py", "TPE")
+
+        return node
+
     if model_name == "PSEUDORANDOM":
         node = ExternalProgramGenerationNode(f"python3 {script_dir}/.random_generator.py", "PSEUDORANDOM")
 
@@ -7705,7 +7719,9 @@ def execute_trials(
 
     cnt = 0
 
-    with ThreadPoolExecutor(max_workers=min(len(index_param_list), 16)) as tp_executor:
+    nr_workers = max(1, min(len(index_param_list), args.max_num_of_parallel_sruns))
+
+    with ThreadPoolExecutor(max_workers=nr_workers) as tp_executor:
         future_to_args = {tp_executor.submit(execute_evaluation, args): args for args in index_param_list}
 
         for future in as_completed(future_to_args):
@@ -9239,7 +9255,7 @@ def main() -> None:
     checkpoint_parameters_filepath = f"{get_current_run_folder()}/state_files/checkpoint.json.parameters.json"
     save_experiment_parameters(checkpoint_parameters_filepath, experiment_parameters)
 
-    print_overview_tables(classic_params, experiment_parameters, experiment_args)
+    print_overview_tables(experiment_parameters, experiment_args)
 
     write_files_and_show_overviews()
 

omniopt2-7076/.tpe.py

@@ -0,0 +1,202 @@
+import sys
+import os
+import json
+import logging
+from typing import Optional
+try:
+    import optuna
+    from optuna.trial import create_trial
+except ModuleNotFoundError:
+    print("Optuna not found. Cannot continue.")
+    sys.exit(1)
+
+try:
+    from beartype import beartype
+except ModuleNotFoundError:
+    print("beartype not found. Cannot continue.")
+    sys.exit(1)
+
+logging.getLogger("optuna").setLevel(logging.WARNING)
+
+@beartype
+def check_constraint(constraint: str, params: dict) -> bool:
+    return eval(constraint, {}, params)
+
+@beartype
+def constraints_not_ok(constraints: list, point: dict) -> bool:
+    if not constraints or constraints is None or len(constraints) == 0:
+        return True
+
+    for constraint in constraints:
+        if not check_constraint(constraint, point):
+            return True
+
+    return False
+
+@beartype
+def tpe_suggest_point(trial: optuna.Trial, parameters: dict) -> dict:
+    point = {}
+    for param_name, param in parameters.items():
+        ptype = param['parameter_type']
+        pvaltype = param['type']
+
+        try:
+            if ptype == 'RANGE':
+                rmin, rmax = param['range']
+                if pvaltype == 'INT':
+                    point[param_name] = trial.suggest_int(param_name, rmin, rmax)
+                elif pvaltype == 'FLOAT':
+                    point[param_name] = trial.suggest_float(param_name, rmin, rmax)
+                else:
+                    raise ValueError(f"Unsupported type {pvaltype} for RANGE")
+
+            elif ptype == 'CHOICE':
+                values = param['values']
+                point[param_name] = trial.suggest_categorical(param_name, values)
+
+            elif ptype == 'FIXED':
+                point[param_name] = param['value']
+
+            else:
+                raise ValueError(f"Unknown parameter_type {ptype}")
+        except KeyboardInterrupt:
+            print("You pressed CTRL-c.")
+            sys.exit(1)
+
+    return point
+
+@beartype
+def generate_tpe_point(data: dict, max_trials: int = 100) -> dict:
+    parameters = data["parameters"]
+    constraints = data.get("constraints", [])
+    seed = data.get("seed", None)
+    trials_data = data.get("trials", [])
+    objectives = data.get("objectives", {})
+
+    direction, result_key = parse_objectives(objectives)
+    study = create_study_with_seed(seed, direction)
+
+    for trial_entry in trials_data:
+        add_existing_trial_to_study(study, trial_entry, parameters, result_key)
+
+    study.optimize(lambda trial: wrapped_objective(trial, parameters, constraints, direction), n_trials=max_trials)
+
+    return get_best_or_new_point(study, parameters, direction)
+
+@beartype
+def parse_objectives(objectives: dict) -> tuple[str, str]:
+    if len(objectives) != 1:
+        raise ValueError("Only single-objective optimization is supported.")
+    result_key, result_goal = next(iter(objectives.items()))
+    if result_goal.lower() not in ("min", "max"):
+        raise ValueError(f"Unsupported objective direction: {result_goal}")
+    direction = "maximize" if result_goal.lower() == "max" else "minimize"
+    return direction, result_key
+
+@beartype
+def create_study_with_seed(seed: Optional[int], direction: str) -> optuna.study.study.Study:
+    return optuna.create_study(
+        sampler=optuna.samplers.TPESampler(seed=seed),
+        direction=direction
+    )
+
+@beartype
+def wrapped_objective(trial: optuna.Trial, parameters: dict, constraints: list, direction: str) -> float:
+    point = tpe_suggest_point(trial, parameters)
+    if not constraints_not_ok(constraints, point):
+        return 1e6 if direction == "minimize" else -1e6
+    return 0.0
+
+@beartype
+def add_existing_trial_to_study(study: optuna.study.study.Study, trial_entry: list, parameters: dict, result_key: str) -> None:
+    if len(trial_entry) != 2:
+        return
+    param_dict, result_dict = trial_entry
+
+    if not result_dict or result_key not in result_dict:
+        return
+
+    if not all(k in param_dict for k in parameters):
+        return
+
+    final_value = result_dict[result_key]
+
+    trial_params = {}
+    trial_distributions = {}
+
+    for name, p in parameters.items():
+        value = param_dict[name]
+
+        if p["parameter_type"] == "FIXED":
+            trial_params[name] = value
+            continue
+
+        if p["parameter_type"] == "RANGE":
+            if p["type"] == "INT":
+                dist = optuna.distributions.IntUniformDistribution(p["range"][0], p["range"][1])
+            elif p["type"] == "FLOAT":
+                dist = optuna.distributions.UniformDistribution(p["range"][0], p["range"][1])
+            else:
+                continue
+
+        elif p["parameter_type"] == "CHOICE":
+            dist = optuna.distributions.CategoricalDistribution(p["values"])
+        else:
+            continue
+
+        trial_params[name] = value
+        trial_distributions[name] = dist
+
+    study.add_trial(
+        create_trial(
+            params=trial_params,
+            distributions=trial_distributions,
+            value=final_value
+        )
+    )
+
+@beartype
+def get_best_or_new_point(study: optuna.study.study.Study, parameters: dict, direction: str) -> dict:
+    best_trial_value = study.best_trial.value
+    if best_trial_value is not None:
+        if (direction == "minimize" and best_trial_value < 1e6) or \
+           (direction == "maximize" and best_trial_value > -1e6):
+            return study.best_params
+    return tpe_suggest_point(study.best_trial, parameters)
+
+@beartype
+def main() -> None:
+    if len(sys.argv) != 2:
+        print("Usage: python script.py <path>")
+        sys.exit(1)
+
+    path = sys.argv[1]
+
+    if not os.path.isdir(path):
+        print(f"Error: The path '{path}' is not a valid folder.")
+        sys.exit(2)
+
+    json_file_path = os.path.join(path, 'input.json')
+    results_file_path = os.path.join(path, 'results.json')
+
+    try:
+        with open(json_file_path, mode='r', encoding="utf-8") as f:
+            data = json.load(f)
+    except FileNotFoundError:
+        print(f"Error: {json_file_path} not found.")
+        sys.exit(3)
+    except json.JSONDecodeError:
+        print(f"Error: Failed to decode JSON in {json_file_path}.")
+        sys.exit(4)
+
+    random_point = generate_tpe_point(data)
+
+    with open(results_file_path, mode='w', encoding="utf-8") as f:
+        json.dump({"parameters": random_point}, f, indent=4)
+
+if __name__ == "__main__":
+    try:
+        main()
+    except KeyboardInterrupt:
+        print("You pressed CTRL-c.")
+        sys.exit(1)

{omniopt2-7037 → omniopt2-7076}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: omniopt2
-Version: 7037
+Version: 7076
 Summary: Automatic highly parallelized hyperparameter optimizer based on Ax/Botorch
 Home-page: https://scads.ai/transfer-2/verfuegbare-software-dienste-en/omniopt/
 Author: Norman Koch
@@ -41,6 +41,7 @@ Requires-Dist: setuptools
 Requires-Dist: pyright
 Requires-Dist: qrcode
 Requires-Dist: tabulate
+Requires-Dist: optuna
 Requires-Dist: shellcheck-py
 Requires-Dist: vulture
 Requires-Dist: flake8

{omniopt2-7037 → omniopt2-7076}/omniopt2.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: omniopt2
-Version: 7037
+Version: 7076
 Summary: Automatic highly parallelized hyperparameter optimizer based on Ax/Botorch
 Home-page: https://scads.ai/transfer-2/verfuegbare-software-dienste-en/omniopt/
 Author: Norman Koch
@@ -41,6 +41,7 @@ Requires-Dist: setuptools
 Requires-Dist: pyright
 Requires-Dist: qrcode
 Requires-Dist: tabulate
+Requires-Dist: optuna
 Requires-Dist: shellcheck-py
 Requires-Dist: vulture
 Requires-Dist: flake8

{omniopt2-7037 → omniopt2-7076}/omniopt2.egg-info/SOURCES.txt

@@ -16,6 +16,7 @@
 .omniopt_plot_worker.py
 .random_generator.py
 .shellscript_functions
+.tpe.py
 LICENSE
 MANIFEST.in
 README.md

{omniopt2-7037 → omniopt2-7076}/omniopt2.egg-info/requires.txt

@@ -31,6 +31,7 @@ setuptools
 pyright
 qrcode
 tabulate
+optuna
 shellcheck-py
 vulture
 flake8

{omniopt2-7037 → omniopt2-7076}/pyproject.toml

@@ -5,7 +5,7 @@ authors = [
   {email = "norman.koch@tu-dresden.de"},
   {name = "Norman Koch"}
 ]
-version = "7037"
+version = "7076"
 
 readme = "README.md"
 dynamic = ["dependencies"]

{omniopt2-7037 → omniopt2-7076}/requirements.txt

@@ -31,3 +31,4 @@ setuptools
 pyright
 qrcode
 tabulate
+optuna