GeneXproTools 4.0 ships with four different Sample
Runs for Time Series Prediction: Dow Jones,
Exchange Rate, Earthquakes,
and Sunspots. To try any one of them, you just have to click
its link on the Welcome Screen of GeneXproTools.
All these four sample runs use real-world data and you will see
that, like with all complex phenomena, it takes some cunning to
design very good models: some times good results can be obtained
using raw data, but other times by smoothing the data you can obtain
better results. In three of the sample runs (Dow
Jones, Exchange Rate, and Earthquakes)
an integral smoothing algorithm was used, and if you compare both
the plots and behavior of these time series with the Sunspots
series (in this case, the original raw data was used), you will see
that, as a rule, it is much easier to create good models using
smoothed data.
Notwithstanding, GeneXproTools has exceptional tools to assist you
in creating good predictive models. The most important is the vast
number of built-in
functions that allow you to create really complex models. As an
example, see the default functions we recommend for Time Series
Prediction in the Run Templates. Try
experimenting with different function sets by mixing some functions
chosen by hand with others picked randomly with the Random button in
the Functions Panel.
Also important is being able to make predictions quickly with all
the best-of-generation models created in a run (they are all kept in
the History Panel) and then choosing the best
one. For that you just have to select the model you want to test in
the History Panel and then make predictions
with it in the Predictions Panel.
These are some of the tips and strategies you might want to try with
the four sample runs GeneXproTools offers for Time Series
Prediction:
The Dow Jones sample run is an interesting
time series application that uses monthly closings of the Dow-Jones industrial
index. For this sample run, each value from the original raw time series
(data file dj.dat obtained from the Time Series Data
Library) was replaced by its moving average using a smoothing period
p = 10. You will see that GeneXproTools is able to make very
good predictions about the monthly closings of the Dow-Jones if you
choose Testing Mode and then compare past known behavior with your
models' predictions in the Predictions Panel.
The Exchange Rate sample run
uses the monthly average exchange rate of Australian dollar for US
dollar. For this sample run, each value from the original raw time series
(data file usexchm.dat obtained from the Time Series Data
Library) was replaced by its moving average using a smoothing period
p = 12. You will see that GeneXproTools is able to make very
good predictions about exchange rates if you choose Testing Mode and
then compare past known behavior with your models' predictions in
the Predictions Panel.
In the Earthquakes sample run a
time series with the number of earthquakes per year with magnitude 7.0 or
greater in the period 1900-1998 is used. And again, each value from the original raw
time series (data file earthq.dat obtained from the Time Series Data
Library) was replaced by its moving average using a smoothing period
p = 10. You will see that GeneXproTools is able to make very
good predictions about earthquakes if you choose Testing Mode and
then compare past known behavior with your models' predictions in
the Predictions Panel.
The Sunspots sample run uses the
well-known Wolfer sunspots data. For this sample run, 100 observations of the Wolfer sunspots series were used with an embedding dimension of 10 and a delay time of one.
You will see that GeneXproTools is able to make very good
predictions about sunspots if you choose Testing Mode and then
compare past known behavior with your models' predictions in the Predictions
Panel.
|