Hamming Ball Auxiliary Sampling for Factorial Hidden Markov Models
Authors: Michalis Titsias RC AUEB, Christopher Yau
NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We illustrate the application of the approach with simulated and a real data example. To demonstrate Hamming ball (HB) sampling we consider an additive FHMM as the one used in [6] and popularized recently for energy disaggregation applications [7, 10, 11]. In all examples, we compare HB with block Gibbs (BG) sampling. Figure 3 shows the evolution of the error of misclassified bits in X, i.e. the number of bits the state X(t) disagrees with the ground-truth Xtrue. |
| Researcher Affiliation | Academia | Michalis K. Titsias Department of Informatics Athens University of Economics and Business mtitsias@aueb.gr Christopher Yau Wellcome Trust Centre for Human Genetics University of Oxford cyau@well.ox.ac.uk |
| Pseudocode | No | The paper describes the algorithm steps in prose but does not include formal pseudocode blocks or an algorithm listing. |
| Open Source Code | No | The paper does not provide any explicit statements about making code open source or links to code repositories. |
| Open Datasets | Yes | Next we consider a publicly available data set2, called the Reference Energy Disaggregation Data Set (REDD) [11], to test the HB and BG sampling algorithms. The REDD data set contains several types of home electricity data for many different houses recorded during several weeks. |
| Dataset Splits | Yes | For full details regarding the energy disaggregation application see [7, 10, 11]. Next we consider a publicly available data set2, called the Reference Energy Disaggregation Data Set (REDD) [11], to test the HB and BG sampling algorithms. The REDD data set contains several types of home electricity data for many different houses recorded during several weeks. Next, we will consider the main signal power of house_1 for seven days which is a temporal signal of length 604, 800 since power was recorded every second. We further downsampled this signal to every 9 seconds to obtain a sequence of 67, 200 size in which we applied the FHMM described below. unsupervised learning using as training data the first day of recorded data. This involves applying an Metropolis-within-Gibbs type of MCMC algorithm that iterates between the following three steps: i) sampling X, ii) sampling each ewk individually using its own Gaussian proposal distribution and accepting or rejecting based on the M-H step and iii) sampling the noise variance σ2 based on its conjugate Gamma posterior distribution. |
| Hardware Specification | No | The paper mentions 'modern computational hardware' but provides no specific details such as CPU, GPU, or memory specifications used for the experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers (e.g., programming languages, libraries, or frameworks). |
| Experiment Setup | Yes | We simulated K = 5 factor chains (with vk = 0.5 , ρk = 0.05, k = 1, . . . , 5) which subsequently generated observations in the 25-dimensional space according to the additive FHMM from Eq. (12) assuming Gaussian noise with variance σ2 = 0.05. The associated factor vector where selected to be wk = wk Maskk where wk = 0.8 + 0.05 (k 1), k = 1, . . . , 5 and Maskk denotes a 25-dimensional binary vector or a mask. Finally, the bias term w0 was set to zero. We tested three block Gibbs sampling schemes: BG1, BG2 and BG3 that jointly sample blocks of rows of size one, two or three respectively. Regarding HB sampling we considered three schemes: HB1, HB2 and HB3 with radius m = 1, 2 and 3 respectively. The time complexities for these HB algorithms were O(36N), O(256N) and O(676N). Notice that an exact sample from the posterior distribution can be drawn in O(1024N) time. We run all algorithms assuming the same random initialization X(0) so that each bit was chosen from the uniform distribution. |