Edit on GitHub

Monotonic Detection and Plotting

One important preprocessing step for time series data processing is the detection of monotonic data and transform it to non-monotonic type by using the finite difference operator.

Artificial Data Example

Let's create an artificial monotonic data and apply our monotonic transformer to normalize it. We can use the Plotter filter to visualize the generated data.

using TSML

Random.seed!(123)
pltr = Plotter(Dict(:pdfoutput => false))
mdates = DateTime(2017,12,1,1):Dates.Hour(1):DateTime(2017,12,31,10) |> collect
mvals = rand(length(mdates)) |> cumsum
df =  DataFrame(Date=mdates ,Value = mvals)
fit_transform!(pltr,df);

Now that we have a monotonic data, let's use the Monotonicer to normalize and plot the result:

using TSML

mono = Monotonicer(Dict())

pipeline = mono |> pltr

res=fit_transform!(pipeline,df);

Real Data Example

We will now apply the entire pipeline starting from reading csv data, aggregate, impute, normalize if it's monotonic, and plot. We will consider three different data types: a regular time series data, a monotonic data, and a daily monotonic data. The difference between monotonic and daily monotonic is that the values in daily monotonic resets to zero or some baseline and cumulatively increases in a day until the next day where it resets to zero or some baseline value. Monotonicer automatically detects these three different types and apply the corresponding normalization accordingly.

using TSML

regularfile = joinpath(dirname(pathof(TSML)),"../data/typedetection/regular.csv")
monofile = joinpath(dirname(pathof(TSML)),"../data/typedetection/monotonic.csv")
dailymonofile = joinpath(dirname(pathof(TSML)),"../data/typedetection/dailymonotonic.csv")

regularfilecsv = CSVDateValReader(Dict(:filename=>regularfile,:dateformat=>"dd/mm/yyyy HH:MM"))
monofilecsv = CSVDateValReader(Dict(:filename=>monofile,:dateformat=>"dd/mm/yyyy HH:MM"))
dailymonofilecsv = CSVDateValReader(Dict(:filename=>dailymonofile,:dateformat=>"dd/mm/yyyy HH:MM"))

valgator = DateValgator(Dict(:dateinterval=>Dates.Hour(1)))
valnner = DateValLinearImputer(Dict(:dateinterval=>Dates.Hour(1)))
stfier = Statifier(Dict(:processmissing=>true))
mono = Monotonicer(Dict())
pltr = Plotter(Dict(:pdfoutput => false))

Regular TS Processing

Let's test by feeding the regular time series type to the pipeline. We expect that for this type, Monotonicer will not perform further processing:

  • Pipeline with Monotonicer: regular time series
pipeline = regularfilecsv |> valgator |> valnner |> mono |> pltr

fit_transform!(pipeline);
  • Pipeline without Monotonicer: regular time series
pipeline = regularfilecsv |> valgator |> valnner |> pltr

fit_transform!(pipeline);

Notice that the plots are the same with or without the Monotonicer instance.

Monotonic TS Processing

Let's now feed the same pipeline with a monotonic csv data.

  • Pipeline without Monotonicer: monotonic time series
pipeline = monofilecsv |> valgator |> valnner |> pltr

fit_transform!(pipeline);
  • Pipeline with Monotonicer: monotonic time series
pipeline = monofilecsv |> valgator |> valnner |> mono |> pltr

fit_transform!(pipeline);

Notice that without the Monotonicer instance, the data is monotonic. Applying the Monotonicer instance in the pipeline converts the data into a regular time series but with outliers.

We can use the Outliernicer filter to remove outliers. Let's apply this filter after the Monotonicer and plot the result.

  • Pipeline with Monotonicer and Outliernicer: monotonic time series
using TSML: Outliernicer
outliernicer = Outliernicer(Dict(:dateinterval=>Dates.Hour(1)));

pipeline = monofilecsv |> valgator |> valnner |> mono |>  outliernicer |> pltr
fit_transform!(pipeline);

Daily Monotonic TS Processing

Lastly, let's feed the daily monotonic data using similar pipeline and examine its plot.

  • Pipeline without Monotonicer: daily monotonic time series
pipeline = dailymonofilecsv |> valgator |> valnner |> pltr
fit_transform!(pipeline);

This plot is characterized by monotonically increasing trend but resets to certain baseline value at the end of the day and repeat similar trend daily. The challenge for the monotonic normalizer is to differentiate between daily monotonic from the typical monotonic function to apply the correct normalization.

  • Pipeline with Monotonicer: daily monotonic time series
pipeline = dailymonofilecsv |> valgator |> valnner |> mono |> pltr
fit_transform!(pipeline);

While the Monotonicer filter is able to transform the data into a regular time series, there are significant outliers due to noise and the nature of this kind of data or sensor.

Let's remove the outliers by applying the Outliernicer filter and examine the result.

  • Pipeline with Monotonicer and Outliernicer: daily monotonic time series
pipeline = dailymonofilecsv |> valgator |> valnner |> mono |> outliernicer |> pltr
fit_transform!(pipeline);

The Outliernicer filter effectively removed the outliers as shown in the plot.