linearregressionmodel.java

化学图形处理软件

        // lets build the model
        try {
            this.modelfit = (LinearRegressionModelFit)revaluator.call("buildLM", 
                    new Object[]{ getModelName(), this.params });
        } catch (Exception re) {
            throw new QSARModelException(re.toString());
        }
    }


    /**
     * Sets parameters required for building a linear model or using one for prediction.
     *
     * This function allows the caller to set the various parameters available
     * for the lm() and predict.lm() R routines. See the R help pages for the details of the available
     * parameters.
     * 
     * @param key A String containing the name of the parameter as described in the 
     * R help pages
     * @param obj An Object containing the value of the parameter
     * @throws QSARModelException if the type of the supplied value does not match the 
     * expected type
     *
     */
    public void setParameters(String key, Object obj) throws QSARModelException {
        // since we know the possible values of key we should check the coresponding
        // objects and throw errors if required. Note that this checking can't really check
        // for values (such as number of variables in the X matrix to build the model and the
        // X matrix to make new predictions) - these should be checked in functions that will
        // use these parameters. The main checking done here is for the class of obj and
        // some cases where the value of obj is not dependent on what is set before it

        if (key.equals("y")) {
            if (!(obj instanceof Double[])) {
                throw new QSARModelException("The class of the 'y' object must be Double[]");
            }
        }
        if (key.equals("x")) {
            if (!(obj instanceof Double[][])) {
                throw new QSARModelException("The class of the 'x' object must be Double[][]");
            }
        }
        if (key.equals("weights")) {
            if (!(obj instanceof Double[])) {
                throw new QSARModelException("The class of the 'weights' object must be Double[]");
            }
        }
        if (key.equals("interval")) {
            if (!(obj instanceof String)) {
                throw new QSARModelException("The class of the 'interval' object must be String");
            }
            if (!(obj.equals("confidence") || obj.equals("prediction"))) {
                throw new QSARModelException("The type of interval must be: prediction or confidence");
            }
        }
        if (key.equals("newdata")) {
            if ( !(obj instanceof Double[][])) {
                throw new QSARModelException("The class of the 'newdata' object must be Double[][]");
            }
        }
        this.params.put(key,obj);
    }
       
        
    /**
     * Uses a fitted model to predict the response for new observations.
     *
     * This function uses a previously fitted model to obtain predicted values
     * for a new set of observations. If the model has not been fitted prior to this
     * call an exception will be thrown. Use <code>setParameters</code>
     * to set the values of the independent variable for the new observations and the
     * interval type.
     * @throws QSARModelException if the model has not been built prior to a call
     * to this method. Also if the number of independent variables specified for prediction
     * is not the same as specified during model building
     */
    public void predict() throws QSARModelException {
        if (this.modelfit == null) 
            throw new QSARModelException("Before calling predict() you must fit the model using build()");

        Double[][] newx = (Double[][])this.params.get(new String("newdata"));
        if (newx[0].length != this.nvar) {
            throw new QSARModelException("Number of independent variables used for prediction must match those used for fitting");
        }
            
        try {
            this.modelpredict = (LinearRegressionModelPredict)revaluator.call("predictLM",
                    new Object[]{ getModelName(), this.params });
        } catch (Exception re) {
            throw new QSARModelException(re.toString());
        }
    }

    /**
     * Returns an object summarizing the linear regression model.
     *
     * The return object simply wraps the fields from the summary.lm
     * return value. Various details can be extracted from the return object,
     * See {@link LinearRegressionModelSummary} for more details.
     *
     * @return A summary for the linear regression model
     * @throws QSARModelException if the model has not been built prior to a call
     * to this method
     */
    public LinearRegressionModelSummary summary() throws QSARModelException {
        if (this.modelfit == null) 
            throw new QSARModelException("Before calling summary() you must fit the model using build()");

        LinearRegressionModelSummary s = null;
        try {
            s = (LinearRegressionModelSummary)revaluator.call("summaryModel",
                    new Object[]{ getModelName() });
        } catch (Exception re) {
            throw new QSARModelException(re.toString());
        }
        return(s);
    }


    /**
     * Loads an LinearRegressionModel object from disk in to the current session.
     *
     * @param fileName The disk file containing the model
     * @throws QSARModelException if the model being loaded is not a linear regression model
     * object
     */
    public void  loadModel(String fileName) throws QSARModelException {
        // should probably check that the fileName does exist
        Object model = (Object)revaluator.call("loadModel", new Object[]{ (Object)fileName });
        String modelName = (String)revaluator.call("loadModel.getName", new Object[] { (Object)fileName });

        if (model.getClass().getName().equals("org.openscience.cdk.qsar.model.R.LinearRegressionModelFit")) {
            this.modelfit =(LinearRegressionModelFit)model;
            this.setModelName(modelName);
            Double tmp = (Double)revaluator.eval("length("+modelName+"$coefficients)-1");
            nvar = (int)tmp.doubleValue();
        } else throw new QSARModelException("The loaded model was not a LinearRegressionModel");
    }
    /**
     * Loads an LinearRegressionModel object from a serialized string into the current session.
     *
     * @param serializedModel A String containing the serialized version of the model
     * @param modelName A String indicating the name of the model in the R session
     * @throws QSARModelException if the model being loaded is not a linear regression model
     * object
     */
    public void  loadModel(String serializedModel, String modelName) throws QSARModelException {
        // should prxbably check that the fileName does exist
        Object model = (Object)revaluator.call("unserializeModel", new Object[]{ (Object)serializedModel, (Object)modelName });
        String modelname = modelName;

        if (model.getClass().getName().equals("org.openscience.cdk.qsar.model.R.LinearRegressionModelFit")) {
            this.modelfit =(LinearRegressionModelFit)model;
            this.setModelName(modelname);
            Double tmp = (Double)revaluator.eval("length("+modelName+"$coefficients)-1");
            nvar = (int)tmp.doubleValue();
        } else throw new QSARModelException("The loaded model was not a LinearRegressionModel");
    }

    /* interface to fit object */

    /**
     * Gets the rank of the fitted linear model.
     *
     * This method only returns meaningful results if the <code>build</code>
     * method of this class has been previously called.
     * 
     * @return An integer indicating the rank
     */
    public int getFitRank() { return(this.modelfit.getRank()); }

    /**
     * Returns the residuals.
     *
     * The residuals are the response minus the fitted values.
     * This method only returns meaningful results if the <code>build</code>
     * method of this class has been previously called.
     * @return A double[] contaning the residuals for each observation
     */
    public double[] getFitResiduals() { return(this.modelfit.getResiduals()); }
    
    /**
     * Returns the estimated coefficients.
     *
     * This method only returns meaningful results if the <code>build</code>
     * method of this class has been previously called.
     * @return A double[] containing the coefficients
     */
    public double[] getFitCoefficients() { return(this.modelfit.getCoefficients()); }

    /**
     * Returns the residual degrees of freedom.
     *
     * This method only returns meaningful results if the <code>build</code>
     * method of this class has been previously called.
     * @return An integr indicating the residual degrees of freedom
     */
    public int getFitDFResidual() { return(this.modelfit.getdfResidual()); }

    /**
     * Returns the fitted mean values.
     *
     * This method only returns meaningful results if the <code>build</code>
     * method of this class has been previously called.
     * @return A double[] containing the fitted  values
     */
    public double[] getFitFitted() { return(this.modelfit.getFitted()); }
    




    /* interface to predict object */

    /**
     * Returns the degrees of freedom for residual.
     *
     * @return An integer indicating degrees of freedom
     */
    public int getPredictDF() { return(this.modelpredict.getDF()); }

    /**
     * Returns the residual standard deviations.
     *
     * @return A double indicating residual standard deviations
     */
    public double getPredictResidualScale() { return(this.modelpredict.getResidualScale()); }

    /**
     * Returns the predicted values for the prediction set. 
     *
     * This function only returns meaningful results if the <code>predict</code>
     * method of this class has been called.
     *
     * @return A double[] containing the predicted values
     */
    public double[] getPredictPredicted() { return(this.modelpredict.getPredicted()); }

    /**
     * Returns the lower prediction bounds. 
     *
     * By default the bounds (both lower and upper) are confidence bounds. However
     * the call to <code>predict</code> can specify prediction bounds.
     * This function only returns meaningful results if the <code>predict</code>
     * method of this class has been called.
     *
     * @return A double[] containing the lower bounds for the predictions
     */
    public double[] getPredictLowerBound() { return(this.modelpredict.getLower()); }

    /**
     * Returns the upper prediction bounds. 
     *
     * By default the bounds (both lower and upper) are confidence bounds. However
     * the call to <code>predict</code> can specify prediction bounds.
     * This function only returns meaningful results if the <code>predict</code>
     * method of this class has been called.
     *
     * @return A double[] containing the upper bounds for the predictions
     */
    public double[] getPredictUpperBound() { return(this.modelpredict.getUpper()); }

    /** 
     * Returns the standard error of predictions.
     *
     * This function only returns meaningful results if the <code>predict</code>
     * method of this class has been called.
     *
     * @return A double[] containing the standard error of predictions.
     */
    public double[] getPredictSEPredictions() { return(this.modelpredict.getSEFit()); }
}