Infomat 1.0
A Short Manual
Magnus Rosell 2007
Content
- Introduction
- Interface overview
- Main View
- Menu and Toolbar
- File
- Image
- Windows
- Tools
- Algorithms
- Help
- Grouping Panel
- Example
Introduction
Infomat is a vector space visualization tool. It allows you to display a matrix,
and group, order and alter it. You may do this along the rows or columns.
Any information stored in a matrix may be investigated using Infomat,
but it is primarily aimed at Information Retrieval. It is optimized for
sparse matixes.
Further information can be found in the
"readme.txt",
the
javadoc
of the program,
and on the
Infomat website.
Interface overview
The main window is divided into four sections. At the top is the menu,
below that the toolbar and under that the grouping panel. The
main view is the fourth section.
There are several other windows that appear in certain situation.
From the beginning the over view is shown. A rectangle in
it shows what part of the matrix that is currently shown in the main view.
In the following sections the main window sections and the other windows will
be described briefely
Main View
The main view displays the matrix. The opacity of the pixels are proportional
to the weight of the matrix elements they represent.
Menu and Toolbar
This sections contains a short account for the available menu options.
As the toolbar contains convenient short cuts to some of the options it
is described here as well.
When a matrix is loaded and the mouse pointer is moving over the main view
the row and column objects is is currently pointoing to is displayed in the
toolbar.
File
In the file meny you can save and load matrix files.
You can also load "corpus files" and "document grouping files". These serves as
a connection between my personal clustering program. You are probably better off
not looking in to them... :)
It is also possible to save the picture in the main view as .png-file.
File Formats
The xml-formats are quite straight forward. You should be able to figure them
out by looking at the examples, see the "readme.txt".
Image
The toolbar is divied into two sections, with two and six buttons.
The four first menu options on the Image menu corresponds to the four first icons in the
second button section:
- Pixel selection
When the mouse is clicked on a pixel in the main view information on it is displayed in a separate window.
- Drag
For moving the selected zoom area.
- Zoom Selection
By clicking, dragging and releasing the mouse within the view that area is zoomed in.
- Zoom
By clicking in the view the zoom amount is increased.
The following two menu options concers the columns.
One can delete columns by click-drag-release (the second last icon in the toolbar)
and toggle the group separators on/off (the first icon).
The same features for the rows are found in the next two options.
The last option in the Image menu is the possibility to reset the zoom entirely.
All functions work in both the main view and the overview.
Windows
The options in the Windows menu are all toggle options. You can turn
the toolbar, the grouping panel, the pixel info window and the overview window
on or off.
The pixel info window shows information on the pixel the mouse pointer is currently
pointing on. It is easily confused with the window that appears in the "pixel selection
process" descirbed in the section on the Image menu.
The grouping panel is described in its own section.
Tools
There are two tools. To invert the matrix speaks for it self.
"Purge matrix" removes all objects that are not displayed in the overview.
If you for instance have deleted certain uniteresting objects from a grouping this function
removes the from the matrix and from all other groupings.
Algorithms
There are three algorithms. If you choose either one a algorithm window will appear.
In it you can decide if you want to cluster rows or columns. The algorithms all have
some proerties that can be altered, like for instance the number of clusters. The algorithm
window explains these properties rather weill.
- K-Means
K-Means clustering
- Relative Clustering
An algorithm that cluster the columns (or rows) relative the rows (columns).
The columns objects that have the higest weight in the first row cluster is assembled into a first column cluster,
and so on.
- Random Clustering
Just what it sounds like.
Help
Try them! :)
Grouping Panel
Through the grouping panel all handling of the groupings is deviced. It is divided into two sections,
one for rows and one for columns. They work similarily.
The topmost drop down menu displays the currently selected grouping. When you choose the grouping here
the order of the objects along the dimension (rows or columns) changes.
The bottom drop down menu selects the coloring grouping. For the rows this leads to a coloring of the pixels,
and for the columns a coloring of the background columns. The pixels are averaged over the matrix elements they
represents, while the column coloring is averaged over the entire columns.
When the "S" button beside each drop down menu is pressed a grouping settings window is displayed.
Grouping Settings Window
The grouping settings window looks a little different depending on which of the four groupings it concerns.
They all have the following sections:
- Name panel
Here the name of the grouping is displayed. You can alter it.
- Groups panel
Here all the groups are displayed. You can alter the names.
- Reordering panel
By changing the order of the numbers in the text filed and pressing the "Apply" button
you can change the order of the groups in the grouping. If you leave a group out it is deleted - a very convenient
way to remove one or more groups.
- File panel
Here you can load and save groupings. For either to work there has to be a matrix loaded.
For coloring groupings you can change the color of each group in the groups panel. The change does not take effect
until you press the "Apply" button in the coloring panel which (for coloring groupings) is located between the
reordering panel and the file panel. There you can also reset the coloring to the default colors.
The opacity of the pixels can be altered in the "row show" grouping settings window and the opacity of the column coloring
in the "column coloring" grouping settings window. By default the column coloring opacity has a lower range than the
pixel opacity.
Example
In the directory .../Infomat/example/
you find a few files to start with.
-
"matrix.xml" is a matrix file consisting of 24 Swedish newspaper articles
represented by 60 words after lemmatization. It is weighed with common
tf*idf-weighting. Load it from the file menu.
-
"article categorization.xml" is a categorization of the articles made by
the newspaper. You could make a clustering and compare it to this categorization:
Make the clustering the shown grouping and the categorization the color grouping
along the dimension of your choice.
-
"corpus.xml" is the same matrix in the "corpus-format".
-
"doc grouping - categorization.xml" is the same article categorization in
the "document grouping format". These two later are a bridge between another
program of the author and Infomat and you are probably better of not using
them.