Short: artificial intelligence quickstart utility Author: Timm S. Müller Type: util/wb Version: 1.0 Requires: os3.x (v39), 68040, guigfx.library, render.library, neuralnet.library released at the 'Bockbier & Hackbraten' neoscientific summer conference overview ---------------------- Madrid is a commodity designed to launch applications. It is not another ToolManager clone, neither it tries to serve the misguiding concept of startbars. Madrid's approach is different. Madrid opens a panel on your Workbench, allowing you to draw simple figures with the mouse. After a short training phase, Madrid securely recognizes these figures. Madrid is not conditioned to recognize pre-defined figures. It is mandatory for you to teach it your own, personal, individual set of figures. Once Madrid recognizes figures securely, you can drag'n'drop icons on its window, thus assigning the respective application to the current figure. Then activate launching. Now Madrid will start the application the next time you redraw its figure. Madrid only recognizes 7 different figures. I've found that more figures would delay the training phase too much, thus confusing people when they first try to get used to this concept. Madrid can be coustomized nicely, e.g. you can 'nail' it to your Workbench's backdrop, and it uses nice graphical buttons and background patterns. requirements -------------------------------------- - os3.0 (v39) - 68040 (might function with 68030 + FPU) - datatypes.library, ILBM datatype - guigfx.library - render.library suggested add-on: - 'wbrun' (util/cli/WBRun_fix.lha) installation ---------------------- - copy Madrid whereever you want. copy Madrid.buttons (and Madrid.pattern if you like) to the same directory. basic usage ---------------------- Madrid is controlled via mouse and keyboard. Hold the left mouse button while drawing a figure. The buttons (from upper-left to lower-left): - square - quit (this button is visible only in backdrop mode) - two frames - hide. - single frame - toggle backdrop mode. - arrow - toggle launching. The lamps at the right act as indicators for the seven different figures. The brighter, the better a figure is recognized. Click a lamp to teach Madrid the current figure. Once a figure is clearly recognized, you can drag icons on Madrid's window. This will assign an application to the current figure. keyboard control: escape - quit h - hide b - toggle backdrop mode a - toggle launching 1...7 - teach the current figure training ---------------------- I admit that Madrid might appear a bit confusing at first sight. Training is an annoying procedure. You won't see any success when you first teach it a set of simple figures. But it WILL learn, I promise. Proceed as follows. 1. First try to imagine the figures you would like to use for launching applications in your daily work. I use S - shell. e - editor. W - WWW browser. M - picture viewer. a weird cross - commodity exchange. arrow - TCP stack. C - IRC client. 2. Don't choose Madrid's window too small, and move it to a place on your desktop where it can reside. 3. Draw the figures some times in order to get used to them (this means: first train the figures to yourself :-) 4. Teach the figures clock-wise. Draw a figure and assign it to a button at the right. (You can also press the keys 1...7 for the same purpose.) You will notice that the brain is completely stupid in the first training cycle. In the second or third training cycle it might weakly recognize some figures. In the fourth to fifth training cycle Madrid should recognize most figures. Intensify training until each figure is securely recognized. 5. Finally, drag the respective icons on the window when a figure has been recognized. Then toggle the arrow button at the lower left. Use the backdrop button (or press 'b') to nail Madrid's window to the Workbench. Configuration should be complete by now. If you do not succeed in teaching Madrid your set of figures, refer to the FAQ section 'troubleshooting'. It is possible that some programs do not start, e.g. the Shell won't be launched in Madrid's default configuration. This is because Madrid tries to call it with the shell command 'run'. Refer to the next section. configuration ---------------------- Madrid by default starts programs with the command line run >NIL: program_name You can modify the function calls in Madrid's configuration file (ENV/ENVARC:madrid.settings). The default commandline (run >NIL: %s) can be modified with the DEFAULTCOMMANDLINE setting. I recommend to use "wbrun" instead of "run". (see also: util/cli/WBRun_fix.lha on Aminet) settings: ACTIONn - command line for each figure. DEFAULTCOMMANDLINE - by default "run >NIL: %s". you can enter a different template, e.g. use "wbrun >NIL: %s" if you want to create commandlines that launch programs as if they were started from the Workbench. SHOWINPUT - visualize quantized input coordinates. default: NO THRESHOLD_ACCEPT - Madrid is quite sure about a figure when the difference between the closest and second closest match is above this threshold (in percent, default: 45) BACKCOLOR - hexadecimal value for the window's background color. default: 000044 SHOWBUTTONS - Use this setting to hide buttons and patterns. default: YES BUTTONFILE - by default "Madrid.buttons". you can specify a different location here or a path/filename of another buttons file. PATTERNFILE - by default "Madrid.pattern". you can specify a different file/location for the button's background pattern here. the 'Madrid.buttons' and 'Madrid.pattern' files will be looked up in the same directory as Madrid by default. Modify the graphics if you like. Madrid is a commodity. Use the regular commodity tooltypes for customization. The default hotkey is LALT F10. Additional tooltypes: SETTINGSFILE - alternative settings file BRAINFILE - alternative brain file history -------------------------------------- v1.0 - added keyboard shortcuts - added SHOWBUTTONS setting - added hide button v0.9 - added button background pattern - Madrid is now a commodity - sources are re-included v0.8 - fixed enforcer hits that could occur when Madrid was quitted - now the difference between the closest and the second closest match is measured. more secure recognition. - added DEFAULTCOMMANDLINE - added BUTTONFILE v0.7 - changed the network topology to 50:22:7 - added graphical buttons. - sources no longer included. v0.6 - implemented automatic saving of the settings. - now there are 7 different figures recognized instead of 4. v0.5 - now using absolute coordinates. better recognition. v0.4 - changed the neural network to use delta cartesian coordinates instead of polar coordinates. v0.3 - initial release at the 'Bockbier & Hackbraten' neoscientific summer conference