Stochastic Variational Inference for Bayesian Time Series Models
Authors: Matthew Johnson, Alan Willsky
ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conclude with a numerical study to validate these algorithms and in particular measure the effectiveness of the approximate updates proposed in Section 5.2. As a performance metric, we evaluate an approximate posterior predictive density on held-out data... First, we compare the performance of SVI and batch mean field algorithms for the HDP-HMM. We sampled a 10-state HMM with 2-dimensional Gaussian emissions and generated a dataset of 100 observation sequences of length 3000 each. We chose a random subset of 95% of the sequences as training sequences and held out 5% as test sequences. We repeated the fitting procedures 5 times with identical initializations drawn from the prior, and we report the median performance with standard deviation error bars. The SVI procedure made only one pass through the training set. Figure 1(a) shows that the SVI algorithm produces fits that are comparable in performance in the time it takes the batch algorithm to complete a single iteration. |
| Researcher Affiliation | Academia | Matthew James Johnson MATTJJ@CSAIL.MIT.EDU Massachusetts Institute of Technology, 77 Massachusetts Ave, Cambridge, MA USA Alan S. Willsky WILLSKY@MIT.EDU Massachusetts Institute of Technology, 77 Massachusetts Ave, Cambridge, MA USA |
| Pseudocode | Yes | Algorithm 1 Stochastic gradient ascent |
| Open Source Code | Yes | Our code is available at github.com/mattjj/pyhsmm. |
| Open Datasets | No | We sampled a 10-state HMM with 2-dimensional Gaussian emissions and generated a dataset of 100 observation sequences of length 3000 each. ... No information on public availability of these generated datasets. |
| Dataset Splits | No | We chose a random subset of 95% of the sequences as training sequences and held out 5% as test sequences. ... The paper defines training and test sets but does not explicitly define a separate validation split. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models or memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers. |
| Experiment Setup | No | The paper mentions 'identical initializations drawn from the prior' but does not specify other experimental setup details such as hyperparameters, learning rates, or optimizer settings. |