Tabular support + DyHPO - joint commit with @karibbov

src/metahyper/api.py

@@ -10,7 +10,7 @@
 from copy import deepcopy
 from dataclasses import dataclass
 from pathlib import Path
-from typing import Any
+from typing import Any, List
 
 from neps.plot.tensorboard_eval import tblogger
 
@@ -206,16 +206,29 @@ def _check_max_evaluations(
 
     return evaluation_count >= max_evaluations
 
+from timeit import default_timer as timer
 
-def _sample_config(optimization_dir, sampler, serializer, logger):
+def _sample_config(optimization_dir, sampler, serializer, logger, pre_load_hooks):
     # First load the results and state of the optimizer
+    logger.info("Metahyper: Started collecting previous and pending data")
+    start = timer()
     previous_results, pending_configs, pending_configs_free = read(
         optimization_dir, serializer, logger, do_lock=False
     )
+    end = timer()
+    logger.info("Metahyper: Finished collecting previous and pending data")
+    logger.info(f"took {end - start} seconds")
 
     base_result_directory = optimization_dir / "results"
 
     logger.debug("Sampling a new configuration")
+
+    for hook in pre_load_hooks:
+        # executes operations on the sampler before setting its state
+        # can be used for setting custom constraints on the optimizer state
+        # for example, can be used to input custom grid of configs, meta learning 
+        # information for surrogate building, any non-stationary auxiliary information
+        sampler = hook(sampler)
     sampler.load_results(previous_results, pending_configs)
     config, config_id, previous_config_id = sampler.get_config_and_ids()
 
@@ -338,6 +351,7 @@ def run(
     logger=None,
     post_evaluation_hook=None,
     overwrite_optimization_dir=False,
+    pre_load_hooks: List=[],
 ):
     serializer = YamlSerializer(sampler.load_config)
     if logger is None:
@@ -392,7 +406,9 @@ def run(
                         config,
                         pipeline_directory,
                         previous_pipeline_directory,
-                    ) = _sample_config(optimization_dir, sampler, serializer, logger)
+                    ) = _sample_config(
+                        optimization_dir, sampler, serializer, logger, pre_load_hooks
+                    )
                 if tblogger.logger_init_bool or tblogger.logger_bool:
                     # This block manages configuration data, potentially for TensorBoard.
                     # Captures details during sampling; initial config always captured.

src/neps/api.py

@@ -109,6 +109,7 @@ def run(
     ignore_errors: bool = False,
     loss_value_on_error: None | float = None,
     cost_value_on_error: None | float = None,
+    pre_load_hooks: List=[],
     searcher: Literal[
         "default",
         "bayesian_optimization",
@@ -154,6 +155,7 @@ def run(
             supress any error and will use given loss value instead. default: None
         cost_value_on_error: Setting this and loss_value_on_error to any float will
             supress any error and will use given cost value instead. default: None
+        pre_load_hooks: List of functions that will be called before load_results().
         searcher: Which optimizer to use. This is usually only needed by neps developers.
         **searcher_kwargs: Will be passed to the searcher. This is usually only needed by
             neps develolpers.
@@ -213,7 +215,7 @@ def run(
             else:
                 new_pipeline_space[key] = value
         pipeline_space = new_pipeline_space
-
+        
         # Transform to neps internal representation of the pipeline space
         pipeline_space = SearchSpace(**pipeline_space)
     except TypeError as e:
@@ -256,4 +258,5 @@ def run(
         post_evaluation_hook=_post_evaluation_hook_function(
             loss_value_on_error, ignore_errors
         ),
+        pre_load_hooks=pre_load_hooks,
     )

src/neps/optimizers/bayesian_optimization/acquisition_functions/__init__.py

@@ -5,6 +5,8 @@
 
 from .ei import ComprehensiveExpectedImprovement
 from .mf_ei import MFEI
+from .ucb import UpperConfidenceBound, MF_UCB
+
 
 AcquisitionMapping: dict[str, Callable] = {
     "EI": partial(
@@ -30,4 +32,12 @@
         in_fill="best",
         augmented_ei=False,
     ),
+    "UCB": partial(
+        UpperConfidenceBound,
+        maximize=False,
+    ),
+    "MF-UCB": partial(
+        MF_UCB,
+        maximize=False,
+    ),
 }