c4patino/slurm-tuner
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Restructure, updated documentation

Browse source
This commit is contained in:
Ceferino Patino 2024-10-11 17:20:27 -05:00
commit 4c4d44546c
Signed by: c4patino
SSH key fingerprint: SHA256:Wu+cU1t+7zVT9wzew/4meNmeulo66NzMqc22MdDBgXI
2 changed files with 11 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

6
README.md
View file

@ -1,6 +1,7 @@
# Hyperparameter Tuning with SLURM and Optuna # Hyperparameter Tuning with SLURM and Optuna
This package integrates Optuna with SLURM to efficiently run hyperparameter optimization jobs on a cluster. It supports customizable trial parameters and allows you to track the results via a CSV file. The SLURM jobs process the parameter configurations and log their results, which are then read by Optuna to calculate the objective loss. This package integrates Optuna with SLURM to efficiently run hyperparameter optimization jobs on a cluster. It supports customizable trial parameters and allows you to track the results via a CSV file. The SLURM jobs process the parameter configurations and log their results, which are then read by Optuna to calculate the objective loss.
## Features ## Features
- Submit SLURM jobs with trial parameters generated by Optuna. - Submit SLURM jobs with trial parameters generated by Optuna.
@ -27,7 +28,7 @@ pip install git+https://github.com/C4theBomb/hyperparameter-tuner.git
from hyperparameter_tuner import Loss from hyperparameter_tuner import Loss
class CustomLossFunction(Loss): class CustomLossFunction(Loss):
def calculate(self, row: pd.Series) -> float: def __call__(self, row: pd.Series, params: Dict[str, Any]) -> float:
return row[0] return row[0]
``` ```
@ -65,13 +66,14 @@ OPTIMIZER=$5
REWARDS=$((RANDOM % 100)) # Replace with your experiment logic REWARDS=$((RANDOM % 100)) # Replace with your experiment logic
# Write results to CSV (first column must be TRIAL_ID so that Optuna can identify the run). # Write results to CSV (first column must be TRIAL_ID so that Optuna can identify the run).
echo "$TRIAL_ID,$REWARDS,$TRAJECTORIES,$LEARNING_RATE,$OPTIMIZER" >> $RESULTS_PATH echo "$TRIAL_ID,$STEP_NUM,$REWARDS,$TRAJECTORIES,$LEARNING_RATE,$OPTIMIZER" >> $RESULTS_PATH
``` ```
### 4. Run your script ### 4. Run your script
``` ```
python main.py python main.py
``` ```
## License ## License
[AGPL-3.0](https://choosealicense.com/licenses/agpl-3.0/) [AGPL-3.0](https://choosealicense.com/licenses/agpl-3.0/)

14
hyperparameter_tuner/slurm_handler.py
View file

@ -28,13 +28,13 @@ def create_objective(
slurm_script: str - Path to the SLURM script to execute. slurm_script: str - Path to the SLURM script to execute.
results_path: str - Path to the CSV file where results will be logged. results_path: str - Path to the CSV file where results will be logged.
loss: Loss - An instance of a Loss class that implements a `calculate` method. loss: Loss - An instance of a Loss class that implements a `calculate` method.
trial_param_types: Dict[str, Tuple[str, Tuple, Dict]]: Dictionary mapping parameter names to their types trial_param_types: Dict[str, Tuple[str, Tuple, Dict]]: Dictionary mapping parameter names to their types and arguments.
and arguments. Each entry is structured as: Each entry is structured as:
- key str: The name of the parameter. - key str: The name of the parameter.
- value: Tuple[str, Tuple, Dict] - value: Tuple[str, Tuple, Dict[str, Any]]
- str - The parameter type ('int', 'float', or 'categorical'). - str - The parameter type ('int', 'float', or 'categorical').
- Tuple - Positional arguments for the parameter's sampling method. - Tuple - Positional arguments for the parameter's sampling method.
- Dict - Keyword arguments for the parameter's sampling method. - Dict[str, Any] - Keyword arguments for the parameter's sampling method.
""" """
def objective(trial: Trial) -> float: def objective(trial: Trial) -> float: