Merge pull request #100 from Luguza/fix_issue_99

Fixed issue 99

Author: Neeratyoy

LICENSE.txt LICENSELICENSE.txt renamed to LICENSE

@@ -186,7 +186,7 @@
       same "printed page" as the copyright notice for easier
       identification within third-party archives.
 
-   Copyright [yyyy] [name of copyright owner]
+   Copyright [2021] [AutoML Freiburg]
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.

README.md

@@ -7,6 +7,8 @@
 [![Static Badge](https://img.shields.io/badge/python-3.8%20%7C%203.9%20%7C%203.10%20%7C%203.11%20-blue)](https://pypi.org/project/dehb/)
 [![arXiv](https://img.shields.io/badge/arXiv-2105.09821-b31b1b.svg)](https://arxiv.org/abs/2105.09821)
 
+> **_NOTE:_**  Following the release of v0.1.2, this project will be maintained for stability and compatibility purposes but will no longer undergo active development. While new features will not be added, the repository will continue to receive necessary updates to ensure ongoing functionality. For any feature requests, please feel free to fork the repository and submit a pull request.
+
 Welcome to DEHB, an algorithm for Hyperparameter Optimization (HPO). DEHB uses Differential Evolution (DE) under-the-hood as an Evolutionary Algorithm to power the black-box optimization that HPO problems pose.
 
 `dehb` is a python package implementing the [DEHB](https://arxiv.org/abs/2105.09821) algorithm. It offers an intuitive interface to optimize user-defined problems using DEHB.

src/dehb/optimizers/dehb.py

@@ -269,6 +269,7 @@ def __getstate__(self):
         d = dict(self.__dict__)
         d["client"] = None  # hack to allow Dask client to be a class attribute
         d["logger"] = None  # hack to allow logger object to be a class attribute
+        d["_runtime_budget_timer"] = None # hack to allow timer object to be a class attribute
         return d
 
     def __del__(self):
@@ -832,7 +833,7 @@ def _save_incumbent(self):
             res = {}
             if self.use_configspace:
                 config = self.vector_to_configspace(self.inc_config)
-                res["config"] = config.get_dictionary()
+                res["config"] = dict(config)
             else:
                 res["config"] = self.inc_config.tolist()
             res["score"] = self.inc_score
@@ -849,8 +850,11 @@ def _save_history(self, name="history.parquet.gzip"):
             return
         try:
             history_path = self.output_path / name
-            history_df = pd.DataFrame(self.history, columns=["config_id", "config", "fitness",
-                                                             "cost", "fidelity", "info"])
+            # Persist bracket_id to reconstruct serial replay order later
+            history_df = pd.DataFrame(
+                self.history,
+                columns=["bracket_id", "config_id", "config", "fitness", "cost", "fidelity", "info"],
+            )
             # Check if the 'info' column is empty or contains only None values
             if history_df["info"].apply(lambda x: (isinstance(x, dict) and len(x) == 0)).all():
                 # Drop the 'info' column
@@ -935,12 +939,20 @@ def _load_checkpoint(self, run_dir: str):
         history_path = run_dir / "history.parquet.gzip"
         history = pd.read_parquet(history_path)
 
-        # Replay history
+        # Sort history to emulate serial execution order if bracket_id available
+        if "bracket_id" in history.columns:
+            history = history.sort_values(by=["bracket_id", "fidelity", "config_id"]).reset_index(drop=True)
+        else:
+            # Fallback ordering for older checkpoints
+            history = history.sort_values(by=["fidelity", "config_id"]).reset_index(drop=True)
+
+        # Replay history in the chosen order
         for _, row in history.iterrows():
             job_info = {
                 "fidelity": row["fidelity"],
                 "config_id": row["config_id"],
                 "config": np.array(row["config"]),
+                **({"bracket_id": int(row["bracket_id"]) } if "bracket_id" in history.columns else {}),
             }
             result = {
                 "fitness": row["fitness"],
@@ -992,6 +1004,8 @@ def tell(self, job_info: dict, result: dict, replay: bool=False) -> None:
             job_info_container["fidelity"] = job_info["fidelity"]
             job_info_container["config"] = job_info["config"]
             job_info_container["config_id"] = job_info["config_id"]
+            if "bracket_id" in job_info:
+                job_info_container["bracket_id"] = job_info["bracket_id"]
 
             # Update entry in ConfigRepository
             self.config_repository.configs[job_info["config_id"]].config = job_info["config"]
@@ -1035,8 +1049,16 @@ def tell(self, job_info: dict, result: dict, replay: bool=False) -> None:
             inc_changed = True
         # book-keeping
         self._update_trackers(
-            traj=self.inc_score, runtime=cost, history=(
-                config_id, config.tolist(), float(fitness), float(cost), float(fidelity), info,
+            traj=self.inc_score,
+            runtime=cost,
+            history=(
+                bracket_id,
+                config_id,
+                config.tolist(),
+                float(fitness),
+                float(cost),
+                float(fidelity),
+                info,
             ),
         )
 
@@ -1166,7 +1188,7 @@ def run(self, fevals=None, brackets=None, total_cost=None, single_node_with_gpus
         self.logger.info("Incumbent config: ")
         if self.use_configspace:
             config = self.vector_to_configspace(self.inc_config)
-            for k, v in config.get_dictionary().items():
+            for k, v in dict(config).items():
                 self.logger.info(f"{k}: {v}")
         else:
             self.logger.info(f"{self.inc_config}")