The following files provide convenience when running different experiments:

• build_color_summaries.jl - Implements a build_experiments function which builds summaries for all DataGraphs included in the specified ExperimentParams. These are stored as .csv files.

• convert_to_gcare.jl - Implements a convert_dataset function which takes a .graph file (following the Subgraph-Matching format) and converts it into a .txt file (following the G-Care format).

• Experiments.jl - Provides the overall Experiments harness of necessary packages and files. Also defines the timeout constant.

• gcare_summary.py - Code which converts G-Care's custom output format into a DataFrame easily parsed in Julia.

• get_true_cardinalities.jl - Implements a calculate_true_cardinalities function obtains the true cardinality for a given dataset from a cardinality file, or calculates the exact cardinality if the file does not exist. Also has a function to verify true cardinalities.

• graph_results.jl - Provides different ways to graph results using the output DataFrame from an experiment.

• load_datasets.jl - Implements a load_dataset function which creates the appropriate DataGraph for the given dataset.

• load_querysets.jl - Implements a load_querysets function which creates a dictionary mapping each input dataset to its corresponding list of QueryGraphs.

• run_estimators.jl - Implements a run_estimation_experiments function which for each given ExperimentParams creates a .csv file describing the results after running the experiment.

• run_submitted_experiments.sh - This is a bash script used to run the experiments described in the submitted paper and save the corresponidng figures.

• utils.jl - Defines enums for different datasets as well as the ExperimentParams used to describe parameter settings for different experiments.

Generally, to run an experiment, one simply needs to create ExperimentParams describing the parameters for the experiment, call build_experiments to build the summaries used in the experiment, call run_estimation_experiments to obtain the results for the estimation, then choose an appropriate function from graph_results.jl to display the results. Multiple examples of this process are included in Experiments/Scripts.

There are a variety of Julia scripts provided in the Experiments/Scripts folder which perform different experiments then save figures presenting the results to the Experiments/Results/Figures folder.

For example, to find how using different maximum cycle sizes in the summary affects the cardinality estimation, the max_cycle_size.jl script can be called from the main directory:

$ julia Experiments/Scripts/max_cycle_size.jl

The Experiments/Results folder contains a folder for figures, but also any saved ColorSummary or cardinality estimation result from using the experiment utility files will be saved here.

After calling build_experiments, for each ExperimentParams, a .csv file storing information about the ColorSummary-building process is stored in Experiments/Results. After the first line with the header, each row of the file stores the dataset, partitioner, number of colors, build phase, time spent in that build phase, and the overall memory footprint. These values are used in figures evaluating the performance of the framework.

Meanwhile, the actual serialized summaries are stored in Experiments/SerializedSummaries to be used in run_estimation_experiments.

After calling run_estimation_experiments, an estimation result .csv file will be saved in Experiments/Results. After the first line with the header, each row of the file stores the cardinality estimate, true cardinality, estimation time, query type (if included in the given query file), file path to the query file, boolean representing if the estimation failed to compute, and the path width of the query. These values are used in figures evaluating the accuracy of the framework.