def main():
args = parseArgs()
X, _, _, _ = filehandlers.loadTrainingSets(args.input)
pca = sklearn.decomposition.PCA(args.variance)
pca.fit(X)
numComponents = pca.components_.shape[0]
numFeatures = pca.components_.shape[1]
print "{:.1f}% ({} / {}) features to retain {}% of the variance".format(100. * numComponents / numFeatures, numComponents, numFeatures, 100 * args.variance)
filehandlers.savePca(args.output, pca)
def runTests(filename, pcaFile=None):
print "TRAINING/TESTING {}".format(filename)
trainingInputs, trainingOutputs, validationInputs, validationOutputs = filehandlers.loadTrainingSets(filename)
if pcaFile:
pca = filehandlers.loadPca(pcaFile)
trainingInputs = pca.transform(trainingInputs)
validationInputs = pca.transform(validationInputs)
numExamples = trainingInputs.shape[0]
numFeatures = trainingInputs.shape[1]
print "Loaded training data with shape {} and {}".format(trainingInputs.shape, trainingOutputs.shape)
print "Loaded validation data with shape {} and {}".format(validationInputs.shape, validationOutputs.shape)
avgPredictions = trainingOutputs.mean(axis=0)
print "Baselines"
print "\tPredict zero on all outputs (val): {}".format(rms(0, validationOutputs.ravel()))
print "\tPredict one on all outputs (val): {}".format(rms(1, validationOutputs.ravel()))
print "\tPredict average on all outputs (val): {}".format(rms(avgPredictions.repeat(validationOutputs.shape[0]), validationOutputs.ravel()))
print "\tPredict one on all outputs then norm (val): {}".format(rms(normalizePredictions(numpy.ones(validationOutputs.shape)).ravel(), validationOutputs.ravel()))
tiledAverage = numpy.tile(avgPredictions, (validationOutputs.shape[0], 1))
print "\tPredict average on all outputs then norm (val): {}".format(rms(normalizePredictions(tiledAverage).ravel(), validationOutputs.ravel()))
experimentLinearRegression(trainingInputs, trainingOutputs, validationInputs, validationOutputs)
#experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, 20, 5)
#experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, 20, 5, maxFeatures="auto")
#experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, 20, None)
#experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, 20, None, maxFeatures=int(numFeatures * 0.75))
#experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, 40, None, maxFeatures=int(numFeatures * 0.5))
#experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, 80, None, maxFeatures=int(numFeatures * 0.5))
#experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, 160, None, maxFeatures=int(numFeatures * 0.5))
#for numTrees in [ 40, 80, 160 ]:
# experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, numTrees, None, maxFeatures=int(numFeatures * 0.5), minSplit=5)
experimentRandomForest(trainingInputs, trainingOutputs, validationInputs, validationOutputs, 40, None, maxFeatures=int(numFeatures * 0.5), minSplit=20, saveAsPath="randomForest-40t-0.5f-20mss.pickle")