CONELPABO

CONELPABO (Composite Networks Learning via Parallel Bayesian Optimization) is a framework designed to analyze long time-series data, with a focus on predicting the Remaining Useful Life (RUL) of systems and components. By using a divide-and-conquer strategy, the framework reduces hyperparameter search complexity and accelerates training by 50%, enabling the training of deeper networks even with limited computational resources.

Key Features:

• Efficient Model Training: Parallel Bayesian Optimization and precomputed embeddings reduce training time and resource usage.
• State-of-the-Art Results: Achieves superior performance on benchmark datasets, showcasing the effectiveness of CNN-CNN and RNN-RNN architectures.
• Reproducible Research: Includes detailed parameter ranges and source code for replicating experiments.

USAGE

The search.py script performs hyperparameter search for two neural network models: FEN (Feature Extractor Network) and RNA MAS (Remaining Useful Life Estimator). It leverages parallel Bayesian Optimization processes to optimize model configurations efficiently, reducing training time and resource usage.

python search.py -fm <fen_model> -rm <rna_model> -d <dataset> -o <output_dir> [options]

Required Arguments

Flag	Long Name	Description	Choices	Type
-fm	--fen_model	Specifies the FEN model type	mscnn, rnn, mscnn1d	str
-rm	--rna_model	Specifies the RNA MAS model type.	mscnn, rnn, mscnn1d	str
-d	--dataset	Path to the dataset metadata file.	-	str
-o	--output	Directory to store the output and results.	-	str

Optional Arguments

Flag	Long Name	Description	Choices	Default	Type
-c	--cuda	Specifies the GPU to use (if available).	0, 1	""	str
-nc	--ncpus	Number of CPU threads to use.	-	4	int
-ng	--ngpus	Number of GPUs to use.	-	2	int
-b	--debug	Enables debug mode for additional logging.	-	False	bool

---

Examples

Basic Example

python search.py -fm mscnn -rm rnn -d PRONOSTIA  -o /path/to/output

Specify GPU

python search.py -fm mscnn1d -rm mscnn -d NCMPASS  -c 0 -o /path/to/output

Customize CPU and GPU Allocation

python search.py -fm rnn -rm rnn -d NCMPASS  -nc 8 -ng 1 -o /path/to/output

Outputs

Flags Directory

During execution, a directory named flags is created inside the specified output directory. This is used for synchronization between the two parallel Bayesian Optimization processes (fen and rna).

Results

Each process generates a file named best_<model>_results.pk in its output directory. These files contain the best hyperparameters and performance metrics for the respective models.

ACKNOWLEDGMENT

This work has been supported by Grant PID2023-147198NB-I00 funded by MICIU/AEI/10.13039/501100011033 (Agencia Estatal de Investigación) and by FEDER, UE, and by the Ministry of Science and Education of Spain through the national program “Ayudas para contratos para la formación de investigadores en empresas (DIN2019-010887 / AEI / 10.13039/50110001103)”, of State Programme of Science Research and Innovations 2017-2020.