History of LOTUS release
=========================
LOTUS 2.3
---------------
September 28, 2005 update
1. Included an option to set the maximum number of regressors used in the node models for
the stepwise model option.
2. Corrected a bug in the generation of the subtrees seqeuence for pruning.
3. For stepwise option, a parent node is declared terminal if one or more of its child
node models contain only the intercept term.
4. Corrected a bug in the computation of the Score and Wald statistics in stepwise
variable selection.
LOTUS 2.2
---------------
June 14, 2005 update
1. Included a stepwise variable selection procedure for multiple linear logistic model.
It uses the score test for entry into the model and the Wald test for staying in the
model (similar to the stepwise procedure used in SAS Proc Logistic). User can specify
the two significance levels for entry and for stay.
2. Redefined cost for stepwise and multiple linear model when missing values are present.
In the older versions, node cost = node model's deviance using nonmissing data. In
current version, cost = node model's deviance using all data at the node with mean
imputation for missing values.
3. Surrogate model is only used, in the presence of missing values, to estimate p when
the model is convergent. Otherwise, the best simple model with mean imputation for
missing values is used.
LOTUS 2.1
---------------
October 18, 2004 update
1. Included a test of significance of "H0: constant model vs H1:best simple model"
at each node of the best simple linear option. Using the critical value
(default = 2) provided by user, a constant model is preferred if the absolute
value of the Wald Statistics of most significant predictor is less than or
equal to the user-specified critical value.
2. For the latex tree, if constant model is fitted to a terminal node, no predictor
is printed beneath the node.
LOTUS 2.0
---------------
July 8, 2004 update
1. Set the maximum number of terminal nodes in the maximal tree to be
tree_size = 500 + 500*(N/50000) for LOTUS(S)
tree_size = 500 + 250*(N/50000) for LOTUS(M)
where N is the sample size.
2. Corrected a bug in choosing the best simple linear model.
3. Corrected a bug caused by floating overflows.
4. For the latex tree with best simple linear option, a +/- sign is given
to indicate positive/negative slope.
5. Included an option to create a new category for missing values in nominal
categorical variables automatically if needed.
LOTUS 1.04
---------------
May 27, 2004 update
1. Corrected an error caused by diverging parameter estimates during Newton-
Raphson iterations.
2. Corrected an error caused by rounding-off errors during Newton-Raphson
iterations.
LOTUS 1.03
---------------
April 27, 2004 update
1. Display LaTeX tree diagram in landscape format if it contains 10 or more
terminal nodes; otherwise portrait format is used.
2. For best simple linear option, the LaTeX tree will show the predictor for
the selected simple model, in addition to the proportion of 0's and 1's
beneath each terminal node.
LOTUS 1.02
---------------
November 16, 2003 update
1. Included the definitions of subroutines median() and mode() in subroutine
grow_tree().
2. Truncate p-hat if it gets too close to 0 or 1. Specifically, if p-hat < e,
set p-hat = e; if p-hat > (1-e), set p-hat = (1-e), where e = 1.0e-15.
LOTUS 1.01
---------------
October 21, 2003 update
1. When Newton-Raphson fails to converge at the root node due to quasi-
separation of data, estimates from the last iteration are reported.
2. Corrected a bug in the computation of MCLASS.
LOTUS 1.00
---------------
September 1, 2003