Deep Bayesian Nonparametric Tracking
Authors: Aonan Zhang, John Paisley
ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical evaluations demonstrates improved performance and interpretable results when tracking stock prices and Twitter trends. |
| Researcher Affiliation | Academia | Aonan Zhang 1 John Paisley 1 1Department of Electrical Engineering & Data Science Institute, Columbia University, New York, USA. Correspondence to: Aonan Zhang <az2385@columbia.edu>. |
| Pseudocode | Yes | Algorithm 1 Sampling from qθ(Ht|Xt) 1: Get St by passing Xt to an LSTM. (Figure 3 blue part) 2: Sample b Ht,1 from an initial distribution. 3: for j = 2, . . . , Nt do 4: Get parameters (µt,j, (Σ1/2 t,j )) for Ht,j by passing ( b Ht,j 1, St,j 1) to a feed-forward network. (Figure 3 red part) 5: Sample b Ht,j = µt,j + Σ1/2 t,j ε, ε N(0, I). (Figure 3 red part) 6: end for |
| Open Source Code | No | No explicit statement or link providing concrete access to the source code for the methodology described in this paper was found. |
| Open Datasets | Yes | Data source: http://ichart.finance.yahoo.com/ (for stock data) and NFL-related tweets (...) posted during the 2011-2012 season (Sinha et al., 2013). (for Twitter data) |
| Dataset Splits | No | The paper specifies training and testing splits (e.g., 'we crop 1/5 of the data for testing and use the rest for training' or 'We use the first 4/5 of data for training and the rest for testing'), but does not explicitly mention a separate validation set or its split details. |
| Hardware Specification | No | No specific hardware details (such as CPU, GPU models, or memory) used for running the experiments were provided. |
| Software Dependencies | No | The paper mentions software components like Adam optimizer, LSTM, and RELU units, but does not provide specific version numbers for any software libraries or frameworks (e.g., TensorFlow, PyTorch versions). |
| Experiment Setup | Yes | For the CKF models, we set the transition variance λ = 0.1 and the likelihood variance σ2 = 1. For models with drift, we set the damping parameter α = 0.1. For VGP models, we set γt,m iid Gam(1, 1), meaning c = 1 and the integral of the L evy measure over four weeks equals 1. For the encoder part of VAE models (see Figure 3), we set the LSTM latent dimension to be 200. (...) For optimization we use Adam (Kingma & Ba, 2014) with learning rate 10 4. |