Time series

Time series modeling

In PowSyBl, time series are modeled by:

• A name to uniquely identify a time series inside a store.

• A data type which is either double or String.

• A time index to define a list of instants for which data exists. Three different implementations of the time index are available in the framework, depending on the need:

  • Regular index: the time step size is constant

  • Irregular index: the time step size varies

  • Infinite index: the time series contains only two points, one at instant 0 and another at instant Long.MAX_VALUE

• Metadata: a list of key/value string data

• Data chunks: an ordered list of data that will be associated with instants of the time index. The data chunks may be compressed or uncompressed.

An uncompressed JSON data chunk looks like:

{
  "offset" : 0,
  "values" : [ 1.0, 1.0, 1.0, 3.0 ]
}

An uncompressed data chunk is modeled with a double (or String) array and an offset. It defines values associated to instants of the time index from offset to offset + values.length.

It is possible to compress the data chunks, using, for example, the RLE. The JSON serialization of compressed data chunks looks like: Output:

{
  "offset" : 0,
  "uncompressedLength" : 4,
  "stepValues" : [ 1.0, 3.0 ],
  "stepLengths" : [ 3, 1 ]
}

Time series can be imported from CSV data:

// Creating a map of TimeSeries per version by parsing a CSV file
Map<Integer, List<TimeSeries>> timeSeriesPerVersion = TimeSeries.parseCsv(pathToCSV);

// Creating a map of TimeSeries per version by parsing the CSV string with a specified configuration
TimeSeriesCsvConfig timeSeriesCsvConfig = new TimeSeriesCsvConfig(';', true, TimeFormat.MILLIS);
Map<Integer, List<TimeSeries>> timeSeriesPerVersion = TimeSeries.parseCsv(csvAsString, timeSeriesCsvConfig);

Calculated time series

Starting from a double time series, it is possible to create calculated time series using a Groovy script.

For instance, let us consider the following example. Let’s say we have created a first double time series named dts in a script, it is then possible to create new time series a and b by writing:

ts['a'] = ts['dts'] + 1
ts['b'] = ts['a'] * 2

The time series a and b, serialized in JSON format, then look like:

[ {
  "name" : "a",
  "expr" : {
    "binaryOp" : {
      "op" : "PLUS",
      "timeSeriesName" : "dts",
      "integer" : 1
    }
  }
}, {
  "name" : "b",
  "expr" : {
    "binaryOp" : {
      "op" : "MULTIPLY",
      "binaryOp" : {
        "op" : "PLUS",
        "timeSeriesName" : "dts",
        "integer" : 1
      },
      "integer" : 2
    }
  }
} ]

The calculated time series are evaluated on the fly during array conversion or iteration (through iterators or streams): only the arithmetic expression is stored.

Here is the list of supported vector operations:

Operator	Purpose	Example	Return type
+	addition	ts['a'] + ts['b']	Numerical
-	substraction	ts['a'] - ts['b']	Numerical
*	multiplication	ts['a'] * ts['b']	Numerical
/	division	ts['a'] / ts['b']	Numerical
==	1 if equals, 0 otherwise	ts['a'] == ts['b']	Boolean
!=	1 if not equals, 0 otherwise	ts['a'] != ts['b']	Boolean
<	1 if less than, 0 otherwise	ts['a'] < ts['b']	Boolean
<=	1 if less than or equals to, 0 otherwise	ts['a'] <= ts['b']	Boolean
>	1 if greater, 0 otherwise	ts['a'] > ts['b']	Boolean
>=	1 if greater than or equals to, 0 otherwise	ts['a'] >= ts['b']	Boolean
-	negation	-ts['a']	Numerical
abs	absolute value	ts['a'].abs()	Numerical
time	convert to time index vector (epoch)	ts['a'].time()	Numerical
min	min value	ts['a'].min(10)	Boolean
max	max value	ts['a'].max(10)	Boolean
min	point-to-point min values between timeseries	min(ts['a'], ts['b'])	Numerical
max	point-to-point max values between timeseries	max(ts['a'], ts['b'])	Numerical

In the Groovy DSL syntax, both timeSeries['a'] and ts['a'] are supported and are equivalent.

To compare a time index vector to a literal date, the time('2018-01-01T00:00:01Z') function is available. For instance, the following code creates a time series of 0 and 1 values:

a = ts['dts'].time() < time('2018-01-01T00:00:01Z')