From:  http://www.dmg.org/public/techreports/pmml1_0.txt
Date:  1999-08-26

<!-- =========================================================================
 Predictive Model Markup Language (PMML) 1.0 - XML Document Type Definition
                           Copyright (C) 1999
                  The Data Mining Group (DMG) at www.dmg.org
==============================================================================
Description:    DTD for the XML based PMML file format Version 1.0
========================================================================== -->

<!ENTITY % predicates
    "(predicate | compound-predicate | true | false)"
>

<!ENTITY % NUMBER "NMTOKEN"> 

<!-- =================================================================
                          Overall structure
================================================================== -->

<!ELEMENT pmml (head?, data-dictionary, (tree-model | regression-model)+)>

<!ATTLIST pmml
    version CDATA #REQUIRED
>

<!-- =================================================================
                           Header Information
================================================================== -->
<!ELEMENT head (application?, annotation*, timestamp?)>

<!ATTLIST head
    copyright CDATA #REQUIRED
    description CDATA #IMPLIED
>

<!-- a timestamp in the format YYYY-MM-DD hh:mm:ss GMT +/- xx:xx -->
<!ELEMENT timestamp (#PCDATA)>

<!-- describes the software application that generated the PMML-->
<!ELEMENT application EMPTY>

<!ATTLIST application
    name CDATA #REQUIRED
    version CDATA #IMPLIED
>

<!ELEMENT annotation (#PCDATA)>


<!-- =================================================================
                            Data Dictionary
================================================================== -->

<!ELEMENT data-dictionary ((categorical | ordinal | continuous)+)>

<!ELEMENT categorical (category+)>
<!ATTLIST categorical
    name CDATA #REQUIRED
>

<!ELEMENT category EMPTY>
<!ATTLIST category
    value CDATA #REQUIRED
    display-value CDATA #IMPLIED
    proportion CDATA #IMPLIED
    missing (true | false) false
>

<!ELEMENT ordinal (order+)>
<!ATTLIST ordinal
    name CDATA #REQUIRED
>

<!ELEMENT order EMPTY>
<!ATTLIST order
    value CDATA #REQUIRED
    display-value CDATA #IMPLIED
    rank CDATA #REQUIRED
    proportion CDATA #IMPLIED
    missing (true | false) false
>

<!-- The predicates indicate the values that represent missing values -->
<!ELEMENT continuous ((%predicates;)*)>
<!ATTLIST continuous
    name CDATA #REQUIRED
    minimum CDATA #IMPLIED
    maximum CDATA #IMPLIED
    mean CDATA #IMPLIED
    median CDATA #IMPLIED
    standard-deviation CDATA #IMPLIED
    inter-quartile-range CDATA #IMPLIED
>



<!-- =================================================================
                           General Tree Model
================================================================== -->

<!ELEMENT tree-model (node)>
<!ATTLIST tree-model
    model-id CDATA #REQUIRED
>

<!-- =================================================================
     The root node of a model should contain a true predicate.
================================================================== -->
<!ELEMENT node ((%predicates;)?, info*, node*, score-distribution*)>
<!ATTLIST node
    score CDATA #IMPLIED
>

<!ELEMENT score-distribution EMPTY>
<!ATTLIST score-distribution
    label CDATA #REQUIRED
    value CDATA #REQUIRED
>

<!ELEMENT info EMPTY>
<!ATTLIST info
    name CDATA #REQUIRED
    value CDATA #REQUIRED
>

<!ELEMENT compound-predicate (%predicates;, (%predicates;)+)>
<!ATTLIST compound-predicate
    bool-op (or | and | xor | cascade) #REQUIRED
>

<!ELEMENT predicate EMPTY>
<!ATTLIST predicate
    attribute CDATA #REQUIRED
    op (eq | ne | lt | le | gt | ge) #REQUIRED
    value CDATA #REQUIRED
>

<!ELEMENT true EMPTY>

<!ELEMENT false EMPTY>

<!-- =================================================================
                        Regression Model
================================================================== -->
<!ELEMENT regression-model (factor-list?, covariate-list?,
                            predictor-to-parameter-correlation-matrix?,
                            parameter-table)> 
<!ATTLIST regression-model 
    model-id CDATA #REQUIRED
    response-variable-name          CDATA           #REQUIRED 
    number-parameters               %NUMBER;        #REQUIRED 
    model-type                      (regression | general-linear | log-linear | multinomial-logistic) #REQUIRED
    verbose-model-specification     CDATA           #IMPLIED
> 

<!ELEMENT factor-list (var-name+)> 

<!ELEMENT covariate-list (var-name+)> 

<!ELEMENT var-name (CDATA)> 

<!ELEMENT predictor-to-parameter-correlation-matrix (predictor-to-parameter-cell+)> 

<!ELEMENT predictor-to-parameter-cell (#PCDATA)> 

<!ATTLIST predictor-to-parameter-cell 
    predictor-name                  CDATA           #REQUIRED 
    parameter-name                  CDATA           #REQUIRED
> 

<!ELEMENT parameter-table (parameter-cell+)> 

<!ELEMENT parameter-cell EMPTY> 

<!ATTLIST parameter-cell 
    target-category                CDATA           #REQUIRED 
    parameter-name                 CDATA           #REQUIRED 
    beta                           %NUMBER;        #REQUIRED 
    std-error                      %NUMBER;        #IMPLIED 
    df                             %NUMBER;        #IMPLIED
>