UCI ML Hackathon: Challenge Datasets
Climate change and ecosystem carbon in California
Climate change and ecosystem carbon in California
Donated by: Shane Coffield (email)
This is a fairly small dataset of 2258 points representing 2258 latitude/longitude gridcells in California (a 1/8 degree or 12.5km spatial resolution). The data are tabular (CSV) format, with each row representing one gridcell on the map. There are columns for latitude; longitude; 8 climate variables: 4 seasons of mean daily temperature (deg C) and 4 seasons of mean daily precipitation (mm/day); aboveground live carbon density (ton carbon per hectare); dominant vegetation type (1 for shrub or grass, 2 for forest). There are also 6 files with 6 different future (2090s) climate scenarios: extreme RCP8.5 warming vs moderate RCP4.5 warming and either dry, mean, or wet moisture scenarios. For more details and links to the original data, see the readme file.
Possible Applications: Understanding how ecosystems are connected to climate and might respond to climate change is a major topic of research in Earth System Science, especially because plants can help mitigate climate change. During photosynthesis, plants take carbon out of the atmosphere, and some of that carbon ends up stored in their aboveground biomass (tree trunks are about 50% carbon) or in the soil. This makes them a ‘natural climate solution’, so long as the ecosystems are stable and that carbon is withheld from the atmosphere for a long time. This dataset could be used (1) to help understand the climatic controls on ecosystems - that is, how seasonal patterns of temperature and precipitation help determine how much vegetation can exist at a location and how much carbon can be stored - and (2) to make future projections of how much carbon we would expect to be stored at different locations under different climate scenarios. As Earth System Scientists we want to know whether it’s realistic to expect increased carbon uptake by the land over this century to help mitigate climate change. We also want to know whether changes in temperature or precipitation are going to be more critical for determining future carbon storage, and to quantify uncertainty of any projections. With this dataset we could gain general scientific insight about the relationships between temperature, precipitation, and carbon, and identify patterns of vulnerability. In other words, be wary of overfitting models, and strive for interpretability.
YelpNYC
Yelp reviews of New York City
Donated by: Jiacen Xu (email)
Yelp NYC dataset contains Yelp reviews for restaurants located in New York City. It has 160,255 users, 923 products and 359,052 reviews. So a User-Review-Product graph can be constructed which has 520,230 nodes and 718,144 edges.
Possible Applications: The goal is to classify the reviews is fake or not.
COVIDLies
A Dataset for Detecting COVID-19-Related Misinformation on Social Media
Donated by: Robert Logan (email)
COVIDLies is a dataset of 62 common misconceptions about the COVID-19 along with 6,591 related tweets, identified and annotated by researchers from the UCI School of Medicine. Given a tweet, our data identifies whether any of the known misconceptions are expressed by the tweet, and if so, whether the tweet propagates the misconception (agree/pos), is informative by contradicting it (disagree/neg), or is neither misinformative nor informative (no stance/na). NOTE: More data may be added to the dataset before the start of the hackathon.
Possible Applications: Content moderation on social media platforms, evaluation of stance detection systems, analysis of linguistic patterns in informative/misinformative tweets.
UCI DeID OMOP Clinical Data Warehouse
UCI DeID OMOP Clinical Data Warehouse
Donated by: Alessandro Ghigi, Zhaoxian Hu (email)
A fully anonymized copy of the UCI Health Clinical Data Warehouse that contains most structured EHR data for all UCI Health patients. The dataset is refreshed monthly. No IRB is required for research use. Available clinical information: encounters, conditions (diagnoses), procedures, measurements (lab tests and vital signs), drugs, observations.
Possible Applications: Feasibility studies, clinical projects that can run against DeID data.
UC COvid Research Data Set (UC CORDS)
UC COvid Research Data Set (UC CORDS)
Donated by: Alessandro Ghigi, Zhaoxian Hu (email)
UC CORDS is a HIPAA Limited Data Set comprised of EHR data across the five UC medical campuses associated with individuals tested for COVID. The dataset is refreshed weekly. No IRB is required for research use. Available clinical information: encounters, conditions (diagnoses), procedures, measurements (lab tests and vital signs), drugs, observations.
Possible Applications: Feasibility studies, clinical projects about COVID.
NCATS National Covid Cohort Collaborative (N3C) Data
NCATS National Covid Cohort Collaborative (N3C) Data
Donated by: Alessandro Ghigi, Zhaoxian Hu (email)
A centralized national data resource that the research community can use to study COVID-19. It is by far “the largest collection of EHR data on COVID-19 patients in the world.
Possible Applications: Feasibility studies, clinical projects about COVID-19.
NIH All of Us Research Data
NIH All of Us Research Data
Donated by: Alessandro Ghigi, Zhaoxian Hu (email)
One of the largest biomedical data resources from a diverse cohort of one million or more participants nationwide. Current data available include health and lifestyle surveys, physical measurements, and electronic health records.
Possible Applications: Feasibility studies, clinical projects.
Edge Prediction
Prediction of missing edges in network data
Donated by: Wayne Hayes (email)
This network represents a recent, high-quality, experimentally-determined biological network of the interactions between proteins in a human cell. Each node is one type of protein, with an edge between two proteins if they are known to interact. Detecting such interactions in the lab is very expensive and error-prone (at least $100 per edge detection with current technology), but the network has many potential clinical applications. Thus, reliable computational prediction of new edges can potentially save money and eventually, lives. This dataset contains graphlet-based topological information that allows the prediction of new edges based on current edges. There are over 150,000 features for each and every pair of nodes (there are 9,000 nodes, giving about 40.5 million pairs), but not every feature exists for every pair. Thus, the dataset is sparsely represented. There are 10 folds, with pre-computed features for each. Users should NOT attempt their own folds, since the features must be computed independently for each fold–the same pair of nodes may have different features sets and values in different folds. Each fold consists of a training set and test set, and the training set actually includes the node pairs in the test set, but with a false value for the edge. So technically this problem involves detecting false-negative edges in the training set, with the test set of known actual edges being the comparison.
Possible Applications: For now edge prediction is relatively novel field, but since networks are used to represent a wide variety of data the technique has wide potential application.
Snapshot Serengeti
Snapshot Serengeti
Donated by: Markelle Kelly, Gavin Kerrigan (email)
This data set contains approximately 2.65M sequences of camera trap images, totaling 7.1M images, from seasons one through eleven of the Snapshot Serengeti project. Labels are provided for 61 categories, primarily at the species level (for example, the most common labels are wildebeest, zebra, and Thomson’s gazelle). Approximately 76% of images are labeled as empty. There are also approximately 150,000 bounding box annotations for approximately 78,000 of these images. Note that this is a very large data set containing both (i) images (very large, a few hundred gigabytes per season), (ii) metadata from human labelers (smaller). You may want to limit your experiments to a subset of all of the image data (e.g., a single season) rather than using all of the data.
Possible Applications: Automated detecting and counting of animals in the wild (ecology, climate science), and research on comparing human and machine image classifiers.