######################################## # The contents of this file are subject to the MLX PUBLIC LICENSE version # 1.0 (the "License"); you may not use this file except in # compliance with the License. # # Software distributed under the License is distributed on an "AS IS" # basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See # the License for the specific language governing rights and limitations # under the License. # # The Original Source Code is "compClust", released 2003 September 03. # # The Original Source Code was developed by the California Institute of # Technology (Caltech). Portions created by Caltech are Copyright (C) # 2002-2003 California Institute of Technology. All Rights Reserved. ######################################## """ A simple pychart based summary plotter. """ import colorsys import string import tempfile import os import copy import Tkinter import re import types import Pmw import MLab import Numeric import gracePlot from Scientific.Functions import LeastSquares from compClust.score import roc from compClust.visualize import plot from compClust.mlx.datasets import Dataset from compClust.mlx.labelings import Labeling from compClust.mlx.labelings import subsetByLabeling from compClust.mlx.views import RowPCAView, RowSubsetView from compClust.util import DistanceMetrics from compClust.util import InterpreterTools from compClust.util import NaN from compClust.visualize import IPlot def estimateLD(matrixSpectrum): """ estimateLD(s) where s is the variances from an svd decompistion """ maxVariance = matrixSpectrum[0] dims = len(matrixSpectrum) estK, fit = LeastSquares.leastSquaresFit(lambda p, x: (1/(x**p[0]))*maxVariance, (.5,), zip(range(1,dims+1), matrixSpectrum)) normalizedShannonEntropy = -1*(MLab.sum(matrixSpectrum*MLab.log(matrixSpectrum)) / dims) return(estK[0], maxVariance, fit, matrixSpectrum, normalizedShannonEntropy) def visualizeLD(matrixSpectrum, name=None, g=None): dims = len(matrixSpectrum) estK, maxVariance, fit, matrixSpectrum, entropy = estimateLD(matrixSpectrum) estSpectrum = map(lambda x: (1.0/(x**estK))*maxVariance, range(1,dims+1)) if g is None: g = gracePlot.gracePlot() g.plot(matrixSpectrum , linetype = 'none') g.hold(1) g.plot(estSpectrum) if name is None: g.title('Variance vs Number of Dimensions') else: g.title('%s: Variance vs Num Dims'%(name)) g.subtitle('Maxvar = %3.2f, Linear Dependence = %3.2f, Entropy = %3.2f'%(maxVariance, estK, entropy)) g.xlabel('Number of Dimensions') g.ylabel('Variance') g.legend(['Real Spectrum', 'x\S-%3.2f \N fit = %3.4f'%(estK, fit)]) return(g) def pca(dataset, labeling=None, plot=None): """ just a quick little pca plotter """ if plot is None: g = gracePlot.gracePlot() else: g = plot pcaDataset = RowPCAView(dataset) variances = pcaDataset.getVariances() colVariances = Labeling(pcaDataset) map(colVariances.addLabelToCol, variances, range(pcaDataset.getNumCols())) variances.sort() variances.reverse() if labeling is None: # We don't have cluster labels make everything identicle xdata = pcaDataset.getColData(colVariances.getColsByLabel(variances[0])[0]) ydata = pcaDataset.getColData(colVariances.getColsByLabel(variances[1])[0]) g.plot(xdata, ydata, linetype='none') else: # we have cluster labels g.hold() for cluster in labeling.getLabels(): print cluster # this is a hack to make it work untill I can figure out # why the subseting on the pca view to work xdata = copy.copy(RowSubsetView(pcaDataset, labeling.getRowsByLabel(cluster)).getColData(colVariances.getColsByLabel(variances[0])[0])) ydata = copy.copy(RowSubsetView(pcaDataset, labeling.getRowsByLabel(cluster)).getColData(colVariances.getColsByLabel(variances[1])[0])) g.plot(xdata, ydata, linetype='none') print colVariances.getLabelsByCol(0) g.xlabel('1st PC- %3.2f variance captured'%(variances[0])) g.ylabel('2nd PC- %3.2f variance captured'%(variances[1])) g.title('2D-PCA projection') g.multi(2,1) g.focus(1,0) print variances visualizeLD(variances, g=g) return(g) def iPCA(ds, labeling=None, annotes=None): """ generate an interactive PCA plot """ removeAnnotes= 0 if annotes is None: annotes = Labeling(ds, 'tmp') annotes.labelRows(range(0, ds.numRows)) removeAnnotes = 1 def drawLine(event, labels): index = event.widget.element_closest(event.x, event.y)['index'] IPlot.plot(ds.getRowData(index)) def summaryPlot(event): g = event.widget pos = "@" +str(event.x) +"," +str(event.y) legend = g.legend_get(pos) # get the selected cluster print legend cluster = subsetByLabeling(ds, labeling, legend) clusterAnnots = Labeling(cluster) clusterAnnotes.labelFrom(annotes) g = IPlot.plot(cluster.getData(), setLabel=clusterAnnotes.getLabelByRows()) g.legend_configure(hide=1) IPlot.hold(0) ds.removeView(cluster) del(cluster) pcaDs = RowPCAView(ds) pcaLabs = Labeling(pcaDs) pcaLabs.labelFrom(labeling) if labeling is not None: for label in labeling.getLabels(): #cluster = pcaDs.subsetRows(labeling.getRowsByLabel( label)) cluster = subsetByLabeling(pcaDs, pcaLabs, label) g = IPlot.plot(cluster.getColData(0), cluster.getColData(1), pointLabels = annotes.getLabelByRows(), setType='scatter', function2=drawLine ) pcaDs.removeView(cluster) IPlot.hold(1) else: g = IPlot.plot (pcaDs.getColData (0), pcaDs.getColData (1), setType='scatter', pointLabels = annotes.getLabelByRows(), function2=drawLine ) g.legend_bind("all", "", summaryPlot) IPlot.hold(0) if removeAnnotes: ds.removeLabeling(annotes) pcaDs.removeLabeling(pcaLabs) ds.removeView(pcaDs) return(g) def tragectorySummary(ds, labeling, annotes=None, clusters=None, computeROC=1): """ This creates a set of thumbnail size trajectories """ def summaryPlot(event, labels): global ev ev= event try: g = Pmw.Blt.Graph() except: reload(Pmw.Blt) g = Pmw.Blt.Graph() row = int(event.widget.grid_info()['row']) col = int(event.widget.grid_info()['column']) label = re.sub('^label ', '', event.widget.winfo_name()) print "working on plotting cluster %s..."%(label) cluster = subsetByLabeling(ds, labeling, label) if annotes is not None: clusterAnnotes = Labeling(cluster) clusterAnnotes.labelFrom(annotes) g = IPlot.plot(cluster.getData(), setLabel=clusterAnnotes.getLabelByRows()) else: g = IPlot.plot(cluster.getData()) g.legend_configure(hide=1) g.configure(title='Cluster %s'%(label)) IPlot.hold(0) ds.removeView(cluster) del(cluster) # this seems like a reasonable default settings numCols = 2 plotSize = (200,200) # this sets up the scrolled frame r = Tkinter.Tk() master = Pmw.ScrolledFrame(r) master.pack(expand='y', fill='both') # this sets up the dataset/labeling stuff if clusters is None: labels = labeling.getLabels() else: labels = clusters tmp = zip(map(lambda x: len(labeling.getRowsByLabel(x)), labels), labels) tmp.sort() tmp.reverse() labels = map(lambda x: x[1], tmp) k= len(labels) numRows = MLab.ceil(float(k)/numCols) for count in range(k): col = count%numCols row = int(MLab.ceil(count/numCols)) label = labels[count] print "working on cluster %s"%(label) print "\t Subseting Cluster" cluster = subsetByLabeling(ds, labeling, label) clusterData = cluster.getData() print "\t Caclulating Mean/Std Cluster" dataMean = MLab.mean(clusterData) try: dataStd = MLab.std(clusterData) except: dataStd = MLab.zeros(cluster.getNumCols()) print "\t Generating Plots" frm = Tkinter.Frame(master.interior()) frm.grid(row=row, col=col) #g = IPlot.plot(Numeric.array([dataMean+dataStd, # dataMean, # dataMean-dataStd]), # master = frm) try: g = Pmw.Blt.Graph(master=frm) except: reload(Pmw.Blt) g= Pmw.Blt.Graph(master=frm) g.line_create('+std', xdata=tuple(range(dataMean.shape[0])), ydata=tuple(dataMean+dataStd), color='red', pixels = "0.04i", dashes=(5,1)) g.line_create('mean', xdata=tuple(range(dataMean.shape[0])), ydata=tuple(dataMean), color='blue', pixels = "0.04i" ) g.line_create('-std', xdata=tuple(range(dataMean.shape[0])), ydata=tuple(dataMean-dataStd), color='red', pixels = "0.04i", dashes=(5,1)) g.legend_configure(hide=1) g.configure(width=plotSize[0],height=plotSize[1]) g.grid_on() g.grid(row=0, col=0) if computeROC and cluster.getNumRows > 3: print "\t Computing ROC" try: rocArea = roc.computeRocForLabel(label, labeling, ds)['area'] except: rocArea = NaN.nan else: rocArea = NaN.nan summaryText = "Cluster %s\n\n Size: %i\n ROC Area: %3.2f\n"%(label, cluster.getNumRows(), rocArea) button = Tkinter.Button(frm, text=summaryText, width=20, name="label "+label) button.grid(row=0,col=1) #IPlot.hold(0) button.bind('', lambda e: summaryPlot(e, labels)) count +=1 del(clusterData) cluster.detatch() del(cluster) return(master) def fitnessTable(fitnessTable, k='k'): """ summarizeFitnessTable(fitnessTable, k) where k is either 'k' or 'k-prime' ie) mccvAlgo.getFitnessTable() """ # fitnessTable = MCCVAlgo.getFitnessTable() if type(fitnessTable) == types.StringType: f = open(fitnessTable,'r') best = f.readline() data=[] for line in f.readlines(): data.append(map(float, line.split())) fitnessTable = Numeric.array(data) if k=='k': KIndex=0 else: KIndex=1 data = {} for row in fitnessTable: if data.has_key(row[KIndex]): data[row[KIndex]].append(row[2]) else: data[row[KIndex]]=[row[2]] X = data.keys() X.sort() Y = map(data.get, X) yMean = MLab.mean(Y,1) try: yStd = MLab.std(Y,1) except: yStd = MLab.zeros(len(X)) g = IPlot.IPlot() X = tuple(X) g.line_create('mean', xdata=X, ydata=tuple(yMean), yerror = tuple(yStd), showerrorbars = 'y', color='blue', pixels = "0.04i" ) g.marker_create("text", name="bestK") try: g.marker_configure("bestK", coords=(MCCVAlgo.getBestParam(), MLab.mean(data[MCCVAlgo.getBestParam()])), text="%i"%(MCCVAlgo.getBestParam()), background="lightblue") except: pass g.grid_on() g.pack(expand=1, fill='both') return(g)