Desbordante user guide

This manual explores the main Desbordante concepts and navigates through the general usage scenario.

Try the deployed instance

View the Project on GitHub Mstrutov/Desbordante

Mining Exact and Approximate Functional Dependencies

A functional dependency (FD) is a constraint between two groups of attributes in a relation. A dependency FD: X → Y means that the values of Y are determined by the values of X. Two tuples sharing the same values of X will necessarily have the same values of Y.

Dependency discovery algorithms allow the user to obtain the FDs held in the data without any prior knowledge of the dataset. The algorithms have the same goal, namely, the discovery of all minimal non-trivial functional dependencies in a given dataset. Basically, it means that Desbordante will process your csv table and output a list of FDs excluding the self-evident ones like ABC→B and minimizing its size as much as possible, ensuring the full list is still deducible. It is worth noting that all the FDs will only have one attribute on the right since ABC→DE is equivalent to a pair of ABC→D and ABC→E.

Relevant data dependencies in real-world datasets are often not exact, so existing discovery algorithms fail to detect them. To cope with this problem, the definition of exact dependencies can be relaxed to allow for a certain degree of violation. Such relaxed dependencies are often called approximate dependencies (AFDs). An AFD A→B holds even if some rows violate the definition of FD, given that the fraction of such rows is small enough.

Configuring the task

There are two major groups of parameters that can be modified.

The first one is dataset parameters — you have to manually specify the csv delimiter, as well as whether it has a header row or the file immediately starts with the data values.

The second one is algorithm parameters. Each algorithm may have up to three different parameters: arity constraint, error threshold, and threads.

Some parameters may be disabled depending on the chosen algorithm. Error threshold is available only to AFD algorithms, threads is enabled only for parallelized algorithms, and arity constraint is supported only by certain algorithms.

Running the task

Click “Analyze” to launch the task. The execution will start right after the dataset is uploaded. The progress bar will help you track the algorithm progress, however, some algorithms do not report on their progress — the bar will appear motionless in this case.

Analyzing the results

The plain list of FDs held on the dataset is usually too large to be useful on its own. You can use three tabs: Attributes, Dependencies, and Dataset to filter, select and view the most prospective dependencies. The intended order of using these tabs goes as follows.

First, pick the attributes you are interested in using the Attributes tab. The subsequent steps will use only those FDs which have at least one chosen attribute from the left diagram in their left-hand side and at least one from the right diagram in their right-hand side. The diagram can help you decide which attributes are the most interesting ones. The attribute area size on the diagram is higher for the attributes which appear in a greater number of dependencies, with shorter ones being given greater weight.

Click on an attribute to select it. If the number of attributes is too large, the least interesting ones will be hidden in the other section. Click on it to expand it and view those attributes. You can return to the previous diagram state by pressing the button above the diagram. You can also filter the attributes by name on-the-fly.

Then, navigate to the Dependencies tab to access the refined list of FDs. You can sort them lexicographically by LHS or RHS attribute names, and use a regex that goes through the list leaving only the matching FDs. If your dataset contains single-column keys, the algorithm would yield a set of FDs A→X for every column X and key A. To keep or filter such dependencies out, you can select whether to show keys or not. The list is paginated.

Finally, choose a dependency you are interested in and switch to the Dataset tab. You can view the analyzed dataset, whose columns that belong to the selected FD are highlighted. Click the “Show non-selected” button to hide the other columns.

Further Reading

For a detailed explanation of exact and approximate functional dependency concepts see:

[1] Thorsten Papenbrock, Jens Ehrlich, Jannik Marten, Tommy Neubert, Jan-Peer Rudolph, Martin Schönberg, Jakob Zwiener, and Felix Naumann. 2015. Functional dependency discovery: an experimental evaluation of seven algorithms. Proc. VLDB Endow. 8, 10 (June 2015), 1082—1093.

[2] Sebastian Kruse and Felix Naumann. 2018. Efficient discovery of approximate dependencies. Proc. VLDB Endow. 11, 7 (March 2018), 759—772.

Back to the main page