|
In this example we
highlight one of the most powerful capability of SAFIR-Xp. More
than the choice of indicators or the zigzag definition, the
selection of the training period is really crucial to make
SAFIR-Xp build a reliable knowledge base. However, in the
previous examples the training period was always the first part
of the data, even if it was not the most adequate.
The principle of
the Search for train/test periods
is a scan of the data through a moving window. The size of this
window is typically the same as the one used during
Explore
(about 1000 bars x number of indicators).
At each step, SAFIR-Xp
evaluates:
§
Its capability to learn on the train period,
§
Its capability to generalize on the test period,
§
And its capability to generalize on the remaining
part of the data file.
The moving window is
then shifted by an offset. You can see the window moving in the
Criteria and prices view. If the test is positive the
train part is redrawn in red color, else in yellow. The test
part is redrawn in white color.
At scanning
completion, the successful periods are sorted according their
result on the whole Explore file or on the Check portfolio.
The best results
give place to the creation of new workspaces. Each of them
receives in its Explore portfolio the train/test interval
defined in the previous step. It is possible to transfer the
remainder of the original data file in the check portfolio of
the new workspaces. It is also possible to transfer the original
Check portfolio.
Exploration
parameters, indicators, and zigzag settings of the original
files are copied.
For each new workspace, SAFIR-Xp
runs an Explore and a Check.
When the command
completes, you have at your disposal new workspaces with
solutions. If there are good results, you can save, create new
configurations, modify the parameters, make other checks or
export some systems. If results are bad, do not hesitate to kill
the workspace.
All the parameters
and the options of the analysis can be set in a dialog box. |