Show HN: ΩID – Faster Integrated Information Decomposition (ΦID) with CUDA

4 months ago 7

ΩID is a Python package for calculating the integrated information decomposition (ΦID) of time series data. It is designed for high-performance computing, with optional GPU acceleration via CuPy.

To install OmegaID with GPU support, you need to have a CUDA-enabled GPU and the CUDA toolkit installed. Then, you can install the package using pip:

pip install "omegaid[gpu]"

If you don't have a GPU or don't want to use it, you can install the CPU-only version:

To use the GPU-accelerated version, you need to set the CUPY_ENABLED environment variable to true:

Then, you can use the high-performance calc_phiid_ccs function as follows:

import numpy as np from omegaid.core.phiid import calc_phiid_ccs, calc_phiid_mmi # Generate some random data src = np.random.randn(1000) trg = np.random.randn(1000) # Calculate PhiID using the high-performance CCS method atoms_res, _ = calc_phiid_ccs(src, trg, tau=1) print("CCS Results:", atoms_res) # For theoretical comparison, you can use the MMI method # Note: This method is designed to run on CPU for performance predictability. atoms_res_mmi, _ = calc_phiid_mmi(src, trg, tau=1) print("MMI Results:", atoms_res_mmi)

Benchmarks (Smart Backend Strategy)

The library employs a smart backend strategy for optimal performance:

calc_phiid_ccs: This function leverages the globally configured backend (xp), which can be CuPy for GPU acceleration, offering significant speedups for parallelizable computations.
calc_phiid_mmi: This function is designed to run on the CPU (NumPy) by default, regardless of CuPy availability. This ensures predictable performance for MMI's inherently sequential global reduction operations, avoiding costly GPU-to-CPU data transfers at high dimensions.

Test Parameters: 50,000 samples, float32 data type.

Dims Redundancy NumPy Time (s) NumPy Mem (MB) CuPy Time (s) CuPy Mem (MB) CuPy GPU Mem (MB) Perf Ratio

16	MMI	0.4647	47.50	N/A	N/A	N/A	N/A
16	CCS	0.1542	0.68	0.3264	278.66	126.00	0.47x
64	MMI	0.2362	8.25	N/A	N/A	N/A	N/A
64	CCS	0.2598	-0.43	0.0252	0.03	197.62	10.30x
256	MMI	0.8283	2.79	N/A	N/A	N/A	N/A
256	CCS	0.8251	-1.43	0.0268	0.00	784.00	30.74x
512	MMI	2.0243	3.00	N/A	N/A	N/A	N/A
512	CCS	2.4208	-0.48	0.0279	0.00	1296.00	86.84x
1024	MMI	6.1629	2.50	N/A	N/A	N/A	N/A
1024	CCS	6.1674	0.45	0.0346	0.00	1294.00	178.19x

Conclusion: For maximum performance and efficiency, calc_phiid_ccs is the recommended function, leveraging GPU acceleration. calc_phiid_mmi is provided for its theoretical value and is optimized for CPU execution.

Read Entire Article