readLearnTS: Learn, classify and predict cell labels.
In imageHTS: Analysis of high-throughput microscopy-based screens

Description Usage Arguments Details Value Author(s) See Also Examples

readLearnTS trains an SVM classifier using cell features and a training cell set. predictCellLabels predicts cell labels.

1 2	readLearnTS(x, featurePar, trainingSet, access='cache', cost, gamma) predictCellLabels(x, uname, access='cache')

`x`	An imageHTS object.
`uname`	A character vector, containing the well names to segment. See `getUnames` for details.
`featurePar`	A character string, indicating the filename containing the feature parameters.
`trainingSet`	A character string, indicating the filename containing the training cell set. See Details.
`access`	A character string indicating how to access the data. Valid values are `local`, `server` and `cache`, the default. See `fileHTS` for details.
`cost`	An optional numeric vector containing the SVM costs to be explored during the cross-validation parameter grid-search. Default is `c(0.1, 1, 10, 20)`.
`gamma`	An optional numeric vector containing the radial kernel gamma parameters to be explored during the cross-validation parameter grid-search. Default is `c(0.0001, 0.001, 0.01, 0.1)`.

readLearnTS trains an SVM classifier using cell features and a training cell set. Features enumerated in the remove.classification.features field of the feature parameters are not considered for classification. The training set, pointed by trainingSet, is a tab-separated file containing the rows uname, spot, id and label. Each row designates a cell. This file is constructed by using the output of the cellPicker module, see popCellPicker. After completion, readLearnTS writes the a RDA file \'data/classifier.rda\' in the local project directory. This file contains the list returned by readLearnTS.

predictCellLabels uses the trained classifier located in the file \'data/classifier.rda\' and cell features to predict cell labels of wells indicated by uname. For each well, the function writes the file clabels, which contains the predicted cell labels.

If present, popCellPicker shows the predicted cell labels. Several iterations of readLearnTS, predictCellLabels and popCellPicker calls are useful to build an efficient cell classifier.

Returns an invisible list which contains: classifier, the trained classifier obtained by tune.svm and cft, the features that were used to train the classifier.

Gregoire Pau, gregoire.pau@embl.de, 2010

popCellPicker