Max Markov Chain
Authors: Yu Zhang, Mitchell Bucklew
IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare MMC with several baselines with synthetic and real-world datasets to demonstrate MMC as a valuable alternative for stochastic modeling. 4 Evaluation For the purpose of comparison, we chose the High-order Markov Chain (HMC), a popular approximate HMC model (MTD [Raftery, 1985]), and First-order Markov Chain (FMC) as the baselines. We evaluated first with synthetic datasets randomly generated and then with real-world datasets. |
| Researcher Affiliation | Academia | Yu Zhang and Mitchell Bucklew School of Computing and Augmented Intelligence, Arizona State University {yzhan442, mbucklew}@asu.edu |
| Pseudocode | No | The paper provides a summary of the parameter optimization procedure as numbered steps but not in a formal pseudocode block or algorithm environment. |
| Open Source Code | No | The paper states 'All implementations are in Python' but does not provide a specific link to the source code or explicitly state that the code is being released. |
| Open Datasets | Yes | We evaluated first with synthetic datasets randomly generated and then with realworld datasets. The real-world datasets were chosen since they are commonly used to evaluate sequence modeling with long-term dependencies. For example, the inflation dataset was used in MTD [Raftery, 1985]. In this experiment we use the inflation rates based on the US Consumer Price Index (CPI) from 1821 to 1999. Apple Stock We took it further to consider Apple stock price from NASDAQ using 1 hour interval. |
| Dataset Splits | No | For each setting, we randomly split the data 75% for training and 25% for testing; we ran multiple trials for the average performance. The paper specifies train/test splits but does not mention a distinct validation split. |
| Hardware Specification | Yes | Experiments were run on Paperspace C7 instances with 12 v CPUs and 30GB RAM. |
| Software Dependencies | No | All implementations are in Python. The paper mentions Python but does not specify versions of Python or any specific libraries/packages used. |
| Experiment Setup | Yes | The same order of MMC, HMC, and MTD was used in the same evaluation setting. For each setting, we randomly split the data 75% for training and 25% for testing; we ran multiple trials for the average performance. We chose order 3 for MMC, MTD, and HMC, resulting in 176 data samples. We chose order 5 for MMC, MTD, and HMC and ran 100 trials to compute the averages. |