src/stat-models/regression-clos.lisp

   1 ;;; -*- mode: lisp -*-
   2
   3 ;;; Time-stamp: <2009-01-31 14:20:14 tony>
   4 ;;; Creation:   <2008-10-03 02:07:10 tony>
   5 ;;; Author:     AJ Rossini <blindglobe@gmail.com>
   6 ;;; Copyright:  (c)2007, AJ Rossini.  BSD, LLGPL, or GPLv2, depending
   7 ;;;             on how it arrives.
   8 ;;; Purpose:    redoing regression in a CLOS based framework.  See
   9 ;;;             regression.lsp for basis of work.
  10
  11 ;;; What is this talk of 'release'? Klingons do not make software
  12 ;;; 'releases'.  Our software 'escapes', leaving a bloody trail of
  13 ;;; designers and quality assurance people in its wake.
  14
  15 ;;; This organization and structure is new to the 21st Century
  16 ;;; version.
  17
  18 (in-package :lisp-stat-regression-linear-clos)
  19
  20 ;;; Regresion Model CLOS, data structures
  21
  22 (defclass regression-model-clos (statistical-model)
  23   ((x :initform nil :initarg :x :accessor x)
  24    (y :initform nil :initarg :y :accessor y)
  25    (included :initform nil :initarg :y :accessor y)
  26    (total-sum-of-squares :initform nil :initarg :y :accessor y)
  27    (residual-sum-of-squares :initform nil :initarg :y :accessor y)
  28    (predictor-names :initform nil :initarg :y :accessor y)
  29    (response-name :initform nil :initarg :y :accessor y)
  30    (case-labels :initform nil :initarg :y :accessor y)
  31    (needs-computing :initform T :initarg :compute? :accessor compute?))
  32   (:documentation "Normal Linear Regression Model through CLOS.
  33   Historical design based on what was done for LispStat, not modern."))
  34
  35 (defclass model-specification ()
  36   ((spec-string :initform nil
  37                 :initarg :specification
  38                 :accessor :specification)
  39    (spec-form :initform nil
  40               :initarg :spec-form
  41               :accessor :spec-form)
  42    (model-class :initform nil))
  43   (:documentation "container for mathematical structure"))
  44
  45 (defclass bayesian-model-specification (model-specification)
  46   ((prior-model-class)
  47    (spec-string :initform nil
  48                 :initarg :specification
  49                 :accessor :specification)
  50    (spec-form :initform nil
  51               :initarg :spec-form
  52               :accessor :spec-form))
  53   (:documentation "adds structure holding priors to the model"))
  54
  55 ;;; The following should be self-created based on introspection of
  56 ;;; available:
  57 ;;; ## inferential technologies (bayesian, frequentist, etc),
  58 ;;; ## optimization criteria (likelihood, least-squares, min-entropy,
  59 ;;;    minimax, etc)
  60 ;;; ## simplification macros, i.e. mapping directly to linear
  61 ;;;    regression and other applications. fast specialized
  62 ;;;    algorithms for edge cases and narrow conditions.
  63 ;;; ##
  64
  65 (defparameter *model-class-list*
  66   '((linear-regression frequentist)
  67     (generalized-linear-regression  parametric)
  68     (linear-regression bayesian)
  69     ()))
  70
  71
  72
  73 ;;; Regression model generics and methods
  74
  75 (defgeneric regression-model (model-spec data-pointer &key debug)
  76   (:documentation "CLOSy framework for regression, using numerics from "))
  77
  78 (defmethod regression-model
  79     ((regr-inst regression-model-clos)
  80      (data-ptr  data-pointer)
  81      &key debug)
  82   "Args: (regr-inst regressino-model-clos)
  83
  84 Returns a fitted regression model object. To examine the model further
  85 assign the result to a variable and send it messages.  Example (data
  86 are in file absorbtion.lsp in the sample data directory/folder):
  87
  88  (def fit-m (fit (new 'regression-model-clos (list iron aluminum) absorbtion)))
  89  (print fit-m)
  90  (plot  fit-m :feature 'residuals)
  91 "
  92   (let ((x (get-variable-matrix (x regr-inst) data-ptr))
  93         (y (get-variable-vector (y regr-inst) data-ptr)))
  94
  95
  96
  97 ;;;;; More mischief from a different time
  98
  99
 100 ;; regression-model is the old API, but regression as a generic will
 101 ;; be the new API.  We need to distinguish between APIs which enable
 102 ;; the user to do clear activities, and APIs which enable developers
 103 ;; to do clear extensions and development, and underlying
 104 ;; infrastructure to keep everything straight and enabled.
 105
 106 ;; There are conflicting theories for how to structure the
 107 ;; specification of mathematical models, along with the statistical
 108 ;; inference, along with the data which is instantiating the model.
 109 ;;
 110 ;; i.e.:  mathematical model for the relationships between components,
 111 ;; between a component and a summarizing parameter, and between
 112 ;; parameters.
 113 ;;
 114 ;; statistical inference describes the general approach for
 115 ;; aggregating into a decision and has impliciations for the scale up
 116 ;; from the model on a single instance to the generalization.
 117 ;;
 118 ;; The data represents the particular substantive context that is
 119 ;; driving the model/inference combination, and about which we hope to
 120 ;; generate knowledge.
 121 ;;
 122 ;; numerical analysis selects appropriate algorithms/implementations
 123 ;; for combining the above 3.
 124 ;;
 125 ;; the end result is input on the decision being made (which could be
 126 ;; specific (decision analysis/testing), risk-analysis (interval
 127 ;; estimation) , most likely/appropriate selection (point estimation)
 128 ;;
 129
 130 #|
 131  (defclass model ()
 132   ((type structure)))
 133
 134  (defgeneric regression ;; assumes x/y from lisp-matrix -- start of a set of generics.
 135     (model dataset)
 136   "Args: (x y &key (intercept T) (print T) (weights nil)
 137           included predictor-names response-name case-labels)
 138 X           - list of independent variables or X matrix
 139 Y           - dependent variable.
 140 INTERCEPT   - T to include (default), NIL for no intercept
 141 PRINT       - if not NIL print summary information
 142 WEIGHTS     - if supplied should be the same length as Y; error
 143               variances are
 144               assumed to be inversely proportional to WEIGHTS
 145 PREDICTOR-NAMES, RESPONSE-NAME, CASE-LABELS
 146             - sequences of strings or symbols.
 147 INCLUDED    - if supplied should be the same length as Y, with
 148               elements nil to skip a in computing estimates (but not
 149               in residual analysis).
 150 Returns a regression model object. To examine the model further assign the
 151 result to a variable and send it messages.
 152 Example (data are in file absorbtion.lsp in the sample data directory):
 153   (def m (regression-model (list iron aluminum) absorbtion))
 154   (send m :help) (send m :plot-residuals)"
 155   (let ((m (send regression-model-proto :new)))
 156     (format t "~%")
 157     (send m :doc doc)
 158     (send m :x x)
 159     (send m :y y)
 160     (send m :intercept intercept)
 161     (send m :weights weights)
 162     (send m :included included)
 163     (send m :predictor-names predictor-names)
 164     (send m :response-name response-name)
 165     (send m :case-labels case-labels)
 166     (if debug
 167         (progn
 168           (format t "~%")
 169           (format t "~S~%" (send m :doc))
 170           (format t "X: ~S~%" (send m :x))
 171           (format t "Y: ~S~%" (send m :y))))
 172     (if print (send m :display))
 173     m))
 174 |#