def __init__(self, data, dimsKeep):
print datetime.now(), 'starting PCA...'
#self.pca = PCA(data)
self.pca = cached(PCA, data)
print datetime.now(), 'done with PCA.'
self.dimsKeep = dimsKeep
def __init__(self, data, dimsKeep):
print datetime.now(), 'starting PCA...'
#self.pca = PCA(data)
self.pca = cached(PCA, data)
print datetime.now(), 'done with PCA.'
self.dimsKeep = dimsKeep
def getData(self, patchSize, number, seed=None):
patches = cached(loadCS294Images,
patchSize=patchSize,
number=number,
seed=seed)
return patches
def getData(self, patchSize, number, seed = None):
samples, labelMatrix, labelStrings = cached(makeUpsonRovio3.randomSampleMatrixWithLabels,
makeUpsonRovio3.trainFilter,
color = (self.colors == 3),
Nw = patchSize, Nsamples = number, seed = seed,
imgDirectory = '../data/upson_rovio_3/imgfiles')
return samples.T # one example per column
def plotCov(data, saveDir = None, prefix = 'imgdata', show = False):
cv = cached(cov, data)
#cv = cov(data)
saveto = os.path.join(saveDir, '%s_cov.png' % prefix) if saveDir else None
image = pil_imagesc(cv, saveto = saveto, show = False)
if show:
image.show()
return cv
def plotCov(data, saveDir=None, prefix='imgdata', show=False):
cv = cached(cov, data)
#cv = cov(data)
saveto = os.path.join(saveDir, '%s_cov.png' % prefix) if saveDir else None
image = pil_imagesc(cv, saveto=saveto, show=False)
if show:
image.show()
return cv
def getData(self, patchSize, number, seed=None):
samples, labelMatrix, labelStrings = cached(
makeUpsonRovio3.randomSampleMatrixWithLabels,
makeUpsonRovio3.trainFilter,
color=(self.colors == 3),
Nw=patchSize,
Nsamples=number,
seed=seed,
imgDirectory='../data/upson_rovio_3/imgfiles')
return samples.T # one example per column
def getBigAndSmallerSamples(fileFilter, whitener, seed, isColor, Nw, Nwshift, Nsamples):
# Sample 1.5x larger windows (e.g. 15 px on a side). This is for
# patches that overlap half with their
# neighbors. largeSamples.shape is (Nsamples, Nw15^2)
NwLarge = Nw + Nwshift
largeSampleMatrix, labelMatrix, labelStrings = cached(randomSampleMatrixWithLabels, fileFilter, seed, isColor, Nw = NwLarge, Nsamples = Nsamples)
stackedSmall = large2StackedSmall(largeSampleMatrix, Nw, Nwshift)
stackedSmallWhite,junk = whitener.raw2normalized(stackedSmall.T)
# dataWhite.shape = (Nw^2, 4*Nsamples). Now with one example per COLUMN!
return largeSampleMatrix.T, stackedSmall.T, stackedSmallWhite, labelMatrix.T, labelStrings
def getDataAndLabels(self, patchSize, number, seed = None):
approxMbCachefile = prod(patchSize) * number * 4.0 / 1000000
if approxMbCachefile < MAX_CACHE_SIZE_MB:
print 'NYU2_Labeled layer:: Predicted approx size of %f MB, using cache' % (approxMbCachefile)
patches, labels = cached(loadNYU2Data, patchSize = patchSize, number = number,
rgbColors = self.colorChannels, depthChannels = self.depthChannels,
seed = seed)
else:
# Skip cache, just run
print 'NYU2_Labeled layer:: Skipping cache, approx size of %f MB > max cache size of %s MB' % (approxMbCachefile, repr(MAX_CACHE_SIZE_MB))
patches, labels = loadNYU2Data(patchSize = patchSize, number = number,
rgbColors = self.colorChannels, depthChannels = self.depthChannels,
seed = seed)
return patches, labels
def testPca():
random.seed(0)
NN = 5000
dim = 200
data = random.randn(NN, dim)
pca = cached(PCA, data)
#pca = PCA(data)
pc = pca.pc()
asPc = pca.toPC(data)
dataWhite = pca.toZca(data, epsilon=1e-6)
def testPca():
random.seed(0)
NN = 5000
dim = 200
data = random.randn(NN,dim)
pca = cached(PCA, data)
#pca = PCA(data)
pc = pca.pc()
asPc = pca.toPC(data)
dataWhite = pca.toZca(data, epsilon = 1e-6)
def getDataAndLabels(self, patchSize, number, seed=None):
approxMbCachefile = prod(patchSize) * number * 4.0 / 1000000
if approxMbCachefile < MAX_CACHE_SIZE_MB:
print 'NYU2_Labeled layer:: Predicted approx size of %f MB, using cache' % (
approxMbCachefile)
patches, labels = cached(loadNYU2Data,
patchSize=patchSize,
number=number,
rgbColors=self.colorChannels,
depthChannels=self.depthChannels,
seed=seed)
else:
# Skip cache, just run
print 'NYU2_Labeled layer:: Skipping cache, approx size of %f MB > max cache size of %s MB' % (
approxMbCachefile, repr(MAX_CACHE_SIZE_MB))
patches, labels = loadNYU2Data(patchSize=patchSize,
number=number,
rgbColors=self.colorChannels,
depthChannels=self.depthChannels,
seed=seed)
return patches, labels
def main():
resman.start('junk', diary = False)
stica = loadFromPklGz('results/130407_132841_76b6586_rapidhacks_upson3_1c_l2_first/stackedTica_mod.pkl.gz')
layer1Whitener = loadFromPklGz('../data/upson_rovio_3/white/train_10_50000_1c.whitener.pkl.gz')
layerSizePlan = [10, 15, 23, 35, 53, 80, 120, 180]
visLayer = 1
largeSampleMatrix, labelMatrix, labelStrings = cached(randomSampleMatrixWithLabels, trainFilter,
seed = 0, color = False,
Nw = layerSizePlan[visLayer], Nsamples = 50000)
seed = 0
Nw = layerSizePlan[visLayer-1] # e.g. 10
Nwbig = layerSizePlan[visLayer] # e.g. 15
Nwshift = Nwbig - Nw # e.g. 15 - 10 = 5
Nsamples = 1000
temp = getBigAndSmallerSamples(trainFilter, layer1Whitener, seed, False, Nw, Nwshift, Nsamples)
largeSampleMatrix, stackedSmall, stackedSmallWhite, labelMatrix, labelStrings = temp
pooled = stica.getRepresentation(largeSampleMatrix)
plotTopActivations(pooled, largeSampleMatrix, (Nwbig,Nwbig), resman.rundir, nActivations = 50, nSamples = 20)
pl = (pooled.T - pooled.mean(1)).T
for ii in range(len(labelStrings)):
print 'finding top for', labelStrings[ii]
if labelMatrix[ii,:].sum() == 0:
print ' skipping, no examples'
continue
avgActivationForClass = (pl * labelMatrix[ii,:]).mean(1)
sortIdx = argsort(avgActivationForClass)
topNeurons = sortIdx[-1:-(50+1):-1]
plotTopActivations(pooled[topNeurons,:], largeSampleMatrix, (Nwbig,Nwbig), resman.rundir,
nActivations = 50, nSamples = 20, prefix = 'topfor_%s' % labelStrings[ii])
resman.stop()
def learnNextLayer(self, params):
nLayers = len(self.ticas)
print 'StackedTica currently has %d layers, learning next' % nLayers
# TODO: only works for one extra layer!
Nw = 10
Nwshift = 5
Nsamples = 50000
#Nsamples = 1000; print 'HACK!'
# Get data and norm it
nextLayerData = cached(makeNewData, self.ticas[-1], self.l1whitener, seed = 0,
isColor = self.isColor, Nw = Nw, Nwshift = Nwshift,
Nsamples = Nsamples)
colNorms = sqrt(sum(nextLayerData**2, 0) + (1e-8))
nextLayerData = nextLayerData / colNorms
# Parameters
if params['dataCrop']:
print '\nWARNING: Cropping data from %d examples to only %d for debug\n' % (nextLayerData.shape[1], params['dataCrop'])
nextLayerData = nextLayerData[:,:params['dataCrop']]
if params['hiddenISize'] != params['hiddenJSize']:
raise Exception('hiddenISize and hiddenJSize must be the same')
hiddenLayerShape = (params['hiddenISize'], params['hiddenJSize'])
neighborhoodParams = ('gaussian', params['neighborhoodSize'], 0, 0)
if self.saveDir:
layerLogDir = os.path.join(self.saveDir, 'layer_%02d' % (nLayers+1))
os.makedirs(layerLogDir)
else:
layerLogDir = ''
# Print/plot data stats
if layerLogDir:
pseudoImgShape = (int(sqrt(nextLayerData.shape[0])), int(sqrt(nextLayerData.shape[0])))
plotImageData(nextLayerData, pseudoImgShape, layerLogDir, pc('data_raw'))
printDataStats(nextLayerData)
# Learn model
tica = TICA(nInputs = nextLayerData.shape[0],
hiddenLayerShape = hiddenLayerShape,
neighborhoodParams = neighborhoodParams,
lambd = params['lambd'],
epsilon = 1e-5,
saveDir = layerLogDir)
manuallySkipExpensivePart = False
if not manuallySkipExpensivePart:
beginTotalCost, beginPoolingCost, beginReconstructionCost, grad = tica.cost(tica.WW, nextLayerData)
tic = time.time()
tica.learn(nextLayerData, maxFun = params['maxFuncCalls'])
execTime = time.time() - tic
if layerLogDir:
saveToFile(os.path.join(layerLogDir, 'tica.pkl.gz'), tica) # save learned model
endTotalCost, endPoolingCost, endReconstructionCost, grad = tica.cost(tica.WW, nextLayerData)
print 'beginTotalCost, beginPoolingCost, beginReconstructionCost, endTotalCost, endPoolingCost, endReconstructionCost, execTime ='
print [beginTotalCost, beginPoolingCost, beginReconstructionCost, endTotalCost, endPoolingCost, endReconstructionCost, execTime]
else:
pdb.set_trace()
# Plot some results
#plotImageRicaWW(tica.WW, imgShape, saveDir, tileShape = hiddenLayerShape, prefix = pc('WW_iterFinal'))
if layerLogDir:
self.plotResults(layerLogDir, tica, nextLayerData, pseudoImgShape, hiddenLayerShape)
self.ticas.append(tica)
def getData(self, patchSize, number, seed = None):
patches = cached(loadCS294Images, patchSize = patchSize, number = number, seed = seed)
return patches