Understanding City Traffic Dynamics Utilizing Sensor and Textual Observations
Authors: Pramod Anantharam, Krishnaprasad Thirunarayan, Surendra Marupudi, Amit Sheth, Tanvi Banerjee
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present a large scale evaluation of the proposed approach on a real-world traffic and twitter dataset collected over a year with promising results. |
| Researcher Affiliation | Academia | Pramod Anantharam, Krishnaprasad Thirunarayan, Surendra Marupudi, Amit Sheth, Tanvi Banerjee Kno.e.sis, Wright State University, Dayton OH, USA {pramod,tkprasad,surendra,amit,tanvi}@knoesis.org |
| Pseudocode | Yes | Algorithm 1 describes the selection of a typical traffic dynamics for each hour. Algorithm 2 determines textual events E, that explains anomalies in sensor data. |
| Open Source Code | No | The paper mentions 'utilizing an openly available city traffic event extraction tool (Anantharam 2014)' but does not state that the source code for the RSLDS methodology presented in *this* paper is open source or provide a link to it. |
| Open Datasets | No | The paper mentions collecting data from '511.org and Tweets' but does not provide specific access information (e.g., a direct link, DOI, or repository) for the exact dataset collected and used in the experiments to ensure reproducibility. |
| Dataset Splits | No | The paper describes 'training phase' and 'testing phase' for its models but does not specify explicit train/validation/test dataset splits (e.g., percentages or sample counts) for the data used in the evaluation. |
| Hardware Specification | Yes | Initial processing was done with 2.66 GHz, Intel Core 2 Duo processor with 8 GB main memory. We then exploited inherent embarrassing parallelism to devise a scalable implementation of our approach on Apache Spark (Zaharia et al. 2010) that takes less than a day. The Apache Spark cluster used in our evaluation has 864 cores and 17TB main memory. |
| Software Dependencies | No | The paper mentions 'Apache Spark (Zaharia et al. 2010)' but does not provide a specific version number for Spark or any other software dependencies, which is necessary for reproducibility. |
| Experiment Setup | Yes | We learn one LDS model for each hour of the day and for each day of the week, giving us 24 7 (168) LDS models for each link. Our approach is similar to Switching Linear Dynamics System (SLDS) (Quinn, Williams, and Mc Intosh 2009) that allows discrete switches to select an appropriate LDS model. We use both the day of the week and the hour of the day to index traffic dynamics and learning normalcy model. The input to Algorithm 1 is the speed observations indexed over internal factors. Algorithm 2 determines textual events E, that explains anomalies in sensor data. The radius r is an input parameter which can be changed but it is set to 0.5 km in our experiments. The adjusted duration of an event, Δte = (ˆest h, ˆeet + h), where, h is set to 1 hour (lowest granularity of our analysis). |