Timelines
I. Development Timeline
9/10 - 9/16
- Ported Prof. Rana's code to Borland C++.
- Created TESGraphWindow window prototype based on sigmoid scale.
9/17 - 9/23
- Added "Scale" menu to TESGraphWindow with logarithmic, sigmoid,
and linear scale options.
- Added generation numbers to horizontal axis of TESGraphWindow.
9/24 - 9/30
- Created prototype TESWindow, with working "Run" button, "ES"
button, and "Generation" display.
10/1 - 10/7
- Added working "Network" and "Problem" buttons to TESWindow,
as well "Best" display, "Plot" button, and "Pause" and "Reset" buttons.
10/8 - 10/14
- Created prototype TNVisWindow, with weights mapped directly to
line brightness.
10/27 - 11/4
- Added "mutation slider" to TESWindow
- Added vertical scale to TESGraphWindow.
- Added horizontal scroll bar to scroll through samples in
TNVisWindow.
- Added ability to inspect inputs and activations by clicking on
a node in TNVisWindow.
- Used combination of line length and brightness, rather than just
brightness, to indicate weight strength in TNVisWindow.
11/5 - 11/11
- Created prototype TGenerationWindow using compact matrix
representation of networks.
- Attempted to draw lines for all weight-parents and sigma-parents
in TGenerationWindow, before realizing it was too cluttered.
- Added ability to display weight-parents for a network by clicking
on it in TGenerationWindow.
11/12 - 11/18
- Added TFamilyTreeWindow.
- Added "Zoom" and "Network" menus to THeredityWindow.
11/19 - 12/2
- Added "File" menu to TNVisWindow allowing load & save
of network files.
- Added "Network" menu to TNVisWindow with "Eval" option.
- Added sigma visualization to TNVisWindow. Added "View" menu
to allow user to turn sigma visualization on and off.
12/10 - 12/16
- Added generation numbers to TGenerationWindow.
1/14 - 1/20
- Added "BackProp" option to "Network" menu in TNVisWindow.
1/21 - 1/27
- Added attractive "IE4-Style" button class.
1/28 - 2/3
- Cleaned up layout of new buttons
- Added ability to slide a weight up and down by dragging it in
TNVisWindow.
II. Experiment Timeline
10/8 - 10/14
- False results with parity problem. Only a subset of the problem
was actually being solved, plus the black line bug caused me to think
I was gaining insights which I actually wasn't.
11/19 - 12/2
12/3 - 12/9
- Solved balance-scale problem.
- Solved balloon classification problem.
- Solved housing value problem.
- Obtained interesting fitness graphs for tic-tac-toe and
housing problems.
1/28 - 2/3
- Observed interesting behavior when sliding weights while
backpropagation algorithm is running.
2/11 - 2/17
- Obtained results for CPU Performance, Image Recognition,
and Breast Cancer Diagnosis problems.