Projection
The Projection panel allows to test, tweak and observe how different
algorithms perform projection or data dimensionality reduction on
samples living in a N-dimensional space: "the canvas".
Projection is usually performed in one of two ways:
- N-dimensions to 1-dimension (e.g. LDA)
- N-dimensions to M-dimensions (e.g. PCA)
where M is usually lower than N (a counter-example: Kernel PCA).
The projection is the only type of algorithm that affects the
current data within the canvas (all other methods only do
computation on the basis of the canvas data). It however stores a
copy of the original data that is used by all projection methods
(useful e.g. for comparison). To revert to the original unprojected
data, it suffices to use the Revert button.
However, if multiple projections need to be applied (e.g. to combine
PCA + LDA in successive iterations), the user needs to use the
Project button for the first method, and the Re-Project button for
the second one. Re-Project will replace the current stored data with
the first projection, and perform the second projection on it. The
projected (or re-projected) data can then be used for
classification, clustering or other problems.
The canvas will display the results of the projection in multiple
layers, which can be changed using the display options. These are:
- Samples: the projected samples, these replace the original
data (which can be re-obtained using the Revert button)
- Learned Model: Additional information about the projection
(e.g. projection lines)
The other display layers (e.g. Density Map) are not used in
projection.
In Practice
The easiest way to perform projection is to:
- Draw some samples (left-click) somewhere in the canvas
- Click on "Project"
This should compute the projection according to the algorithm
selected, and replace the current samples with the projected ones.
Options and Commands
The interface for projection (the right-hand side of the Algorithm
Options dialog) provides the following commands:
- Project: perform the projection using the currently selected
algorithm and options, pressing this button multiple times will
always perform the projection from the same original data
- Revert: return the original data into the canvas (erase the
projected data)
- ReProject: perform a new projection on top of the current
projected data (instead of the stored original data)
and the following options:
- Auto-Fit: adjust the canvas to fit the projected data after
the projection (useful for widespread data that will be
normalized by the projection)
- Manual Selection: manually select the training samples (if the
algorithm requires training data)
All other options are algorithm-dependent and should be described in
the help menu of the algorithm itself.