Members in attendance Alan, Tu, Mike, Tim

Mike's Program- Michael illustrated his code and gave a transcript of a test case. Then there was discussion of some relevant topics. Firstly, it was decided that the user would need to create a header file. The header file will appear something like.

.name_binning Strategy

.questions_15 //number of question on survey

If the user did not make a header there would be a default binding strategy.

Secondly, in regards to the first topic, Alan proposes a minimum of two types of binding strategies. One would be for bin lengths to be uniform; the second would have individual bin lengths defined by the user. Michael appended to these strategies the idea of bins based upon measures of standard deviation away from the norm.

Regarding outlying survey intervals were the Expert doctor states that there is potential for data, but no patient fell into that interval. We decided that if the header were used we would have the option to include or exclude the interval from the decomposer. But if the user gave no header, the binner would take range information from the sample set and use this to create max and min values.

Tim points out that there can even be a hybrid of this. For instance, the Expert only specifies some of the columns. This should be allowed as well.

Alan advises four frames.

• One would hold various texts, including the survey data. One would be able to shift through the various 'important' texts.
• The second would hold graphical data, such as a histogram of 'binned' data per variable. Michael also thought that a correlation matrix between variables could be included here. This matrix could help in choosing vacuous variables, in choosing bound sets and in interpreting the output.
• The third would be a graphical representation of the decomposed switching function.
• The fourth pane would have many possibilities. They could be case statements, English representations of the switching function modules.

We talked about the possibility of using VBASIC rather than VC++.

We have decided to put an agenda together at the end of each meeting. This could be appended throughout the following week.

Tim (by Tuesday): look into VBASIC and using C++ code as libraries rather than as GUI development tools. Build VC++ low level interface. The interface will focus on aspects that we discussed last Tuesday (see minutes).

Mike: further refines his binding code

Next meeting.

This will be held Thursday, February 3 2000 at 8:00 PM in PCAT.