java4eurostat

Java4eurostat is a Java library for statistical data manipulation. It provides a number of functions to load statistical data into an 'hypercube' structure and index it for fast in-memory computations. A number of specific functions are provided to easily access Eurostat data.

Quick start

Let's start with a simple dataset:

stored as a CSV file example.csv:

country,gender,year,population
Brasil,Male,2013,45.1
Brasil,Female,2013,48.3
Brasil,Total,2013,93.4
Brasil,Male,2014,46.2
Brasil,Female,2014,47.7
Brasil,Total,2014,93.9
Japan,Male,2013,145.1
Japan,Female,2013,148.3
Japan,Total,2013,293.4
Japan,Male,2014,146.2
Japan,Female,2014,147.7
Japan,Total,2014,293.9

This file can be loaded into an hypercube structure with:

StatsHypercube hc = CSV.load("example.csv", "population");

Information on the hypercube structure is shown with hc.printInfo();, which returns:

Information: 12 value(s) with 3 dimension(s).
   Dimension: gender (3 dimension values)
      Female
      Male
      Total
   Dimension: year (2 dimension values)
      2013
      2014
   Dimension: country (2 dimension values)
      Brasil
      Japan

Several input formats are supported. For example, Eurostat data can be loaded directly from the web. For that, only the database code given in Eurostat databases catalog is required. For example, the database on HICP - Country weights (code prc_hicp_cow) can be downloaded and loaded simply with:

StatsHypercube hc2 = EurobaseIO.getData("prc_hicp_cow");

The structure returned with a hc2.printInfo(); is:

Information: 2001 value(s) with 3 dimension(s).
   Dimension: time (21 dimension values)
      1996
      1997
      1998
      ...
   Dimension: geo (35 dimension values)
      AT
      BE
      BG
      ...
   Dimension: statinfo (6 dimension values)
      COWEA
      COWEA18
      COWEA19
      ...

Once loaded, data can be filtered/selected. For example, hc.selectDimValueEqualTo("country","Brasil") selects data for Brasil and hc.selectValueGreaterThan(147) selects data with values greater than 147. Selection criteria can be combined in cascade like hc.selectDimValueEqualTo("country","Brasil").selectDimValueGreaterThan("year",2012) for the selection of Brasil data after 2012. Logical operators 'AND', 'OR' and 'NOT' can also be used to build more complex selection criteria. Totally generic selection criteria can be specified such as:

hc.select(new Criteria(){
	@Override
	public boolean keep(Stat stat) {
		return stat.dims.get("country").contains("r") && Math.sqrt(stat.value)>7;
	}
});

which selects all statistics with country names containing a "r" character, and whose root square value is greater than 7.

A single value can be retrieved with for example hc.selectDimValueEqualTo("country", "Japan", "gender", "Total", "year", "2014").stats.iterator().next().value but the fastest way to retrieve a value and scan a dataset is to use an index with:

StatsIndex index = new StatsIndex(hc,"gender","year","country");

This index is a tree structure based on the dimension values. This structure can be displayed with index.print();:

Total
   2014
      Brasil -> 93.9
      Japan -> 293.9
   2013
      Brasil -> 93.4
      Japan -> 293.4
Male
   2014
      Brasil -> 46.2
      Japan -> 146.2
   2013
      Brasil -> 45.1
      Japan -> 145.1
Female
   2014
      Brasil -> 47.7
      Japan -> 147.7
   2013
      Brasil -> 48.3
      Japan -> 148.3

A statistical value is accessed quickly from the index and its dimension values: double value = index.getSingleValue("Total","2014","Japan");. Scanning a full dataset across its dimensions is very fast with:

for(String gender : index.getKeys())
	for(String year : index.getKeys(gender))
		for(String country : index.getKeys(gender,year)) {
			System.out.println(gender +" "+year+" "+country);
			System.out.println(index.getSingleValue(gender,year,country));
		}

More information

Input formats

CSV

//load
StatsHypercube hc = CSV.load("example.csv", "population");
//save
CSV.save(hc, "C:\datafolder\", "dataset.csv");

JSON-stat

See here for more information on this format.

//load
String jsonStatString = '{"version":"2.0", "class":"dataset", "label":"Population data", "source":"", "id":[...], "size":[...], "dimension":{...}, "value":[...]}';
StatsHypercube hc = JSONStat.load(jsonStatString);
//save
//  not implemented (yet)

Eurostat data

The class EurobaseIO provide several functions to handle Eurostat data. For example, StatsHypercube hc = EurobaseIO.getData("prc_hicp_cow"); loads the database prc_hicp_cow.

Eurostat TSV files can also be downloaded manually from the bulk download facility or using:

//download from Eurostat bulk download facility
EurobaseIO.getDataBulkDownload("eurobase_code","/home/datafolder/");
//load
StatsHypercube hc = EurostatTSV.load("/home/datafolder/eurobase_code.tsv");
//save
//  not implemented (yet)

Filtering on loading

To ensure an efficient usage of memory, a selection criteria can be specified when loading from a data source. For example, StatsHypercube hc = EurobaseIO.getData("prc_hicp_cow", new DimValueEqualTo("geo","BG")) loads only data for country BG.

Basic data structures

The base classes are Stat and StatsHypercube. A Stat object represents a statistical values, element of a StatsHypercube object.

A Stat object is characterised by its value (of course) and a dictionnary of pairs (dimension label, dimension value), which represents its coordinates within the hypercube. Flags can also be attached to a statistical value. The class StatsHypercube is simply characterised by its collection of Stat elements and dimension names.

[TODO: describe HierarchicalCode]

Data access

Data of a hypercube are accessed using either the StatsHypercube.select() method or a StatsIndex object. Access with a StatsIndex is faster, but requires the construction of an index object, which can be resource consumming.

Basic operations based on selection and indexing are presented in the quick start section above.

[TODO: extend description.]

Time series analysis

The class TimeSeriesUtil provides several function for time series analysis such as the computation of moving averages, gap analysis and outlier values detection.

[TODO: extend description.]

R interface

[TODO. based on Rserve]