Exemplo n.º 1
0
def get_correlation(x, y, test=True):
  corr = r.cor(x, y)
  return corr
print r.sum(valores)[0]
print r.prod(valores)[0]
print r.sort(valores)
print r.mean(valores)[0]
print r.median(valores)[0]
print r.sd(valores)[0]
print r.var(valores)[0]

valores_python = list(valores)

he = IntVector([10, 2, 23, 11, 14, 35, 46, 32, 13, 51, 27, 49])
ha = he
print r.var(he)[0]
print r.cov(ha, he)[0]
print r.cor(ha, he)[0]

#  funções

sqr = robjects.r('function(x) x^2')
print(sqr)
print(sqr(2))
print(sqr(IntVector([4])))
print(sqr(IntVector([4,4])))

eleva3 = robjects.r('function(a){ return(a*a*a); }')
print(eleva3)
print(eleva3(2))
print(eleva3(IntVector([4])))
print(eleva3(IntVector([4,4])))
Exemplo n.º 3
0
def main(argv=None):

    if argv is None:
        argv = sys.argv

    parser = E.OptionParser(
        version="%prog version: $Id: r_table2scatter.py 2782 2009-09-10 11:40:29Z andreas $")

    parser.add_option("-c", "--columns", dest="columns", type="string",
                      help="columns to take from table. Choices are 'all', 'all-but-first' or a ','-separated list of columns.")

    parser.add_option("--logscale", dest="logscale", type="string",
                      help="log-transform one or both axes [default=%Default].")

    parser.add_option("-a", "--hardcopy", dest="hardcopy", type="string",
                      help="write hardcopy to file [default=%default].",
                      metavar="FILE")

    parser.add_option("-f", "--file", dest="input_filename", type="string",
                      help="filename with table data [default=%default].",
                      metavar="FILE")

    parser.add_option("-2", "--file2", dest="input_filename2", type="string",
                      help="additional data file [default=%default].",
                      metavar="FILE")

    parser.add_option("-s", "--stats", dest="statistics", type="choice",
                      choices=("correlation", "spearman", "pearson", "count"),
                      help="statistical quantities to compute [default=%default]",
                      action="append")

    parser.add_option("-p", "--plot", dest="plot", type="choice",
                      choices=("scatter", "pairs", "panel", "bar", "bar-stacked",
                               "bar-besides", "1_vs_x", "matched", "boxplot", "scatter+marginal",
                               "scatter-regression"),
                      help="plots to plot [default=%default]",
                      action="append")

    parser.add_option("-t", "--threshold", dest="threshold", type="float",
                      help="min threshold to use for counting method [default=%default].")

    parser.add_option("-o", "--colours", dest="colours", type="int",
                      help="column with colour information [default=%default].")

    parser.add_option("-l", "--plot-labels", dest="labels", type="string",
                      help="column labels for x and y in matched plots [default=%default].")

    parser.add_option("-d", "--add-diagonal", dest="add_diagonal", action="store_true",
                      help="add diagonal to plot [default=%default].")

    parser.add_option("-e", "--plot-legend", dest="legend", type="int",
                      help="column with legend [default=%default].")

    parser.add_option("-r", "--options", dest="r_options", type="string",
                      help="R plotting options [default=%default].")

    parser.add_option("--format", dest="format", type="choice",
                      choices=("full", "sparse"),
                      help="output format [default=%default].")

    parser.add_option("--title", dest="title", type="string",
                      help="""plot title [default=%default].""")

    parser.add_option("", "--xrange", dest="xrange", type="string",
                      help="x viewing range of plot [default=%default].")

    parser.add_option("", "--yrange", dest="yrange", type="string",
                      help="y viewing range of plot[default=%default].")

    parser.add_option("--allow-empty-file", dest="fail_on_empty", action="store_false",
                      help="do not fail on empty input [default=%default].")

    parser.add_option("--fail-on-empty", dest="fail_on_empty", action="store_true",
                      help="fail on empty input [default=%default].")

    parser.set_defaults(
        hardcopy=None,
        input_filename="",
        input_filename2=None,
        columns="all",
        logscale=None,
        statistics=[],
        plot=[],
        threshold=0.0,
        labels="x,y",
        colours=None,
        diagonal=False,
        legend=None,
        title=None,
        xrange=None,
        yrange=None,
        r_options="",
        fail_on_empty=True,
        format="full")

    (options, args) = E.Start(parser)

    if len(args) == 1 and not options.input_filename:
        options.input_filename = args[0]

    if options.columns not in ("all", "all-but-first"):
        options.columns = [int(x) - 1 for x in options.columns.split(",")]

    if options.colours:
        options.colours -= 1
    if options.legend:
        options.legend -= 1

    table = {}
    headers = []

    # read data matrix
    if options.input_filename:
        lines = IOTools.openFile(options.input_filename, "r").readlines()
    else:
        # note: this will not work for interactive viewing, but
        # creating hardcopy plots works.
        lines = sys.stdin.readlines()

    lines = [x for x in lines if x[0] != "#"]

    if len(lines) == 0:
        if options.fail_on_empty:
            raise IOError("no input")
        E.warn("empty input")
        E.Stop()
        return

    matrix, headers, colours, legend = readTable(lines,
                                                 "matrix",
                                                 take_columns=options.columns,
                                                 headers=True,
                                                 colours=options.colours,
                                                 row_names=options.legend)

    if options.input_filename2:
        # read another matrix (should be of the same format.
        matrix2, headers2, colours2, legend2 = readTable(
            lines,
            "matrix2",
            take_columns=options.columns,
            headers=True,
            colours=options.colours,
            row_names=options.legend)

    R.assign("headers", headers)

    ndata = R("""length( matrix[,1] )""")[0]

    if options.loglevel >= 1:
        options.stdlog.write("# read matrix: %ix%i\n" % (len(headers), ndata))

    if colours:
        R.assign("colours", colours)

    for method in options.statistics:

        if method == "correlation":
            cor = R.cor(matrix, use="pairwise.complete.obs")
            writeMatrix(sys.stdout, cor, headers=headers, format="%5.2f")

        elif method == "pearson":
            options.stdout.write("\t".join(("var1",
                                            "var2",
                                            "coeff",
                                            "passed",
                                            "pvalue",
                                            "n",
                                            "method",
                                            "alternative")) + "\n")
            for x in range(len(headers) - 1):
                for y in range(x + 1, len(headers)):
                    try:
                        result = R(
                            """cor.test( matrix[,%i], matrix[,%i] )""" % (x + 1, y + 1))
                    except rpy.RPyException as msg:
                        E.warn("correlation not computed for columns %i(%s) and %i(%s): %s" % (
                            x, headers[x], y, headers[y], msg))
                        options.stdout.write("%s\t%s\t%s\t%s\t%s\t%i\t%s\t%s\n" %
                                             (headers[x], headers[y],
                                              "na",
                                              "na",
                                              "na",
                                              0,
                                              "na",
                                              "na"))

                    else:
                        options.stdout.write(
                            "%s\t%s\t%6.4f\t%s\t%e\t%i\t%s\t%s\n" %
                            (headers[x], headers[y],
                             result.rx2('estimate').rx2(
                                 'cor')[0],
                             Stats.getSignificance(
                                 float(result.rx2('p.value')[0])),
                             result.rx2('p.value')[0],
                             result.rx2('parameter').rx2(
                                 'df')[0],
                             result.rx2('method')[0],
                             result.rx2('alternative')[0]))

        elif method == "spearman":
            options.stdout.write("\t".join(("var1", "var2",
                                            "coeff",
                                            "passed",
                                            "pvalue",
                                            "method",
                                            "alternative")) + "\n")
            for x in range(len(headers) - 1):
                for y in range(x + 1, len(headers)):
                    result = R(
                        """cor.test( matrix[,%i], matrix[,%i], method='spearman')""" % (x + 1, y + 1))
                    options.stdout.write(
                        "%s\t%s\t%6.4f\t%s\t%e\t%i\t%s\t%s\n" %
                        (headers[x], headers[y],
                         result['estimate']['rho'],
                         Stats.getSignificance(float(result['p.value'])),
                         result['p.value'],
                         result['parameter']['df'],
                         result['method'],
                         result['alternative']))

        elif method == "count":
            # number of shared elements > threshold
            m, r, c = MatlabTools.ReadMatrix(open(options.input_filename, "r"),
                                             take=options.columns,
                                             headers=True)
            mask = numpy.greater(m, options.threshold)
            counts = numpy.dot(numpy.transpose(mask), mask)
            writeMatrix(options.stdout, counts, headers=c, format="%i")

    if options.plot:

        # remove columns that are completely empty
        if "pairs" in options.plot:
            colsums = R('''colSums( is.na(matrix ))''')
            take = [x for x in range(len(colsums)) if colsums[x] != ndata]
            if take:
                E.warn("removing empty columns %s before plotting" % str(take))
                matrix = R.subset(matrix, select=[x + 1 for x in take])
                R.assign("""matrix""", matrix)
                headers = [headers[x] for x in take]
                if legend:
                    legend = [headers[x] for x in take]

        if options.r_options:
            extra_options = ", %s" % options.r_options
        else:
            extra_options = ""

        if options.legend is not None and len(legend):
            extra_options += ", legend=c('%s')" % "','".join(legend)

        if options.labels:
            xlabel, ylabel = options.labels.split(",")
            extra_options += ", xlab='%s', ylab='%s'" % (xlabel, ylabel)
        else:
            xlabel, ylabel = "", ""

        if options.colours:
            extra_options += ", col=colours"

        if options.logscale:
            extra_options += ", log='%s'" % options.logscale

        if options.xrange:
            extra_options += ", xlim=c(%f,%f)" % tuple(
                map(float, options.xrange.split(",")))

        if options.yrange:
            extra_options += ", ylim=c(%f,%f)" % tuple(
                map(float, options.yrange.split(",")))

        if options.hardcopy:
            if options.hardcopy.endswith(".eps"):
                R.postscript(options.hardcopy)
            elif options.hardcopy.endswith(".png"):
                R.png(options.hardcopy, width=1024, height=768, type="cairo")
            elif options.hardcopy.endswith(".jpg"):
                R.jpg(options.hardcopy, width=1024, height=768, type="cairo")

        for method in options.plot:

            if ndata < 100:
                point_size = "1"
                pch = "o"
            elif ndata < 1000:
                point_size = "1"
                pch = "o"
            else:
                point_size = "0.5"
                pch = "."

            if method == "scatter":
                R("""plot( matrix[,1], matrix[,2], cex=%s, pch="o" %s)""" % (
                    point_size, extra_options))

            if method == "scatter-regression":
                R("""plot( matrix[,1], matrix[,2], cex=%s, pch="o" %s)""" % (
                    point_size, extra_options))
                dat = R(
                    """dat <- data.frame(x = matrix[,1], y = matrix[,2])""")
                R(
                    """new <- data.frame(x = seq( min(matrix[,1]), max(matrix[,1]), (max(matrix[,1]) - min(matrix[,1])) / 100))""")
                mod = R("""mod <- lm( y ~ x, dat)""")
                R("""predict(mod, new, se.fit = TRUE)""")
                R("""pred.w.plim <- predict(mod, new, interval="prediction")""")
                R("""pred.w.clim <- predict(mod, new, interval="confidence")""")
                R(
                    """matpoints(new$x,cbind(pred.w.clim, pred.w.plim[,-1]), lty=c(1,2,2,3,3), type="l")""")
                R.mtext(
                    "y = %f * x + %f, r=%6.4f, n=%i" % (mod["coefficients"]["x"],
                                                        mod["coefficients"][
                                                            "(Intercept)"],
                                                        R("""cor( dat )[2]"""),
                                                        ndata),
                    3,
                    cex=1.0)

            elif method == "pairs":
                if options.add_diagonal:
                    R(
                        """panel.hist <- function( x,y,...  ) { points(x,y,...); abline(0,1); }""")
                else:
                    R(
                        """panel.hist <- function( x,y,...  ) { points(x,y,...); }""")

                # There used to be a argument na_action="na.omit", but
                # removed this as there appeared error messages saying
                # "na.action is not a graphical parameter" and the
                # plots showed occasionally the wrong scale.
                # cex=point_size also caused trouble (error message:
                # "X11 used font size 8 when 2 was requested" or
                # similar)
                if options.colours:
                    R.pairs(matrix,
                            pch=pch,
                            col=colours,
                            main=options.title,
                            panel="panel.hist",
                            labels=headers,
                            cex_labels=2.0)
                else:
                    R.pairs(matrix,
                            pch=pch,
                            panel="panel.hist",
                            main=options.title,
                            labels=headers,
                            cex_labels=2.0)

            elif method == "boxplot":
                extra_options += ",main='%s'" % options.title

                # set vertical orientation
                if max([len(x) for x in headers]) > 40 / len(headers):
                    # remove xlabel:
                    extra_options = re.sub(", xlab='[^']+'", "", extra_options)
                    extra_options += ", names.arg=headers, las=2"
                    R(
                        """op <- par(mar=c(11,4,4,2))""")  # the 10 allows the names.arg below the barplot

                R("""boxplot( matrix %s)""" % extra_options)

            elif method == "bar" or method == "bar-stacked":
                if not options.colours:
                    extra_options += ", col=rainbow(5)"

                # set vertical orientation
                if max([len(x) for x in headers]) > 40 / len(headers):
                    # remove xlabel:
                    extra_options = re.sub(", xlab='[^']+'", "", extra_options)
                    extra_options += ", names.arg=headers, las=2"
                    R(
                        """op <- par(mar=c(11,4,4,2))""")  # the 10 allows the names.arg below the barplot

                R("""barplot(as.matrix(matrix), %s)""" % extra_options)

            elif method == "bar-besides":
                if not options.colours:
                    extra_options += ", col=rainbow(%i)" % ndata

                # set vertical orientation
                if max([len(x) for x in headers]) > 40 / len(headers):
                    # remove xlabel:
                    extra_options = re.sub(", xlab='[^']+'", "", extra_options)
                    extra_options += ", names.arg=headers, las=2"
                    R(
                        """op <- par(mar=c(11,4,4,2))""")  # the 10 allows the names.arg below the barplot

                R("""barplot(as.matrix(matrix), beside=TRUE %s)""" %
                  extra_options)

            elif method == "scatter+marginal":

                if options.title:
                    # set the size of the outer margins - the title needs to be added at the end
                    # after plots have been created
                    R.par(oma=R.c(0, 0, 4, 0))

                R("""matrix""")
                R("""
x <- matrix[,1];
y <- matrix[,2];
xhist <- hist(x, breaks=20, plot=FALSE);
yhist <- hist(y, breaks=20, plot=FALSE);
top <- max(c(xhist$counts, yhist$counts));
nf <- layout(matrix(c(2,0,1,3),2,2,byrow=TRUE), c(3,1), c(1,3), respect=TRUE );
par(mar=c(3,3,1,1)) ;
plot(x, y, cex=%s, pch="o" %s) ;
par(mar=c(0,3,1,1)) ;
barplot(xhist$counts, axes=FALSE, ylim=c(0, top), space=0 ) ;
par(mar=c(3,0,1,1)) ;
title(main='%s');
barplot(yhist$counts, axes=FALSE, xlim=c(0, top), space=0, horiz=TRUE ) ;
title(main='%s');
""" % (point_size, extra_options, xlabel, ylabel))

                if options.title:
                    R.mtext(options.title, 3, outer=True, line=1, cex=1.5)

            elif method in ("panel", "1_vs_x", "matched"):

                if method == "panel":
                    pairs = []
                    for x in range(len(headers) - 1):
                        for y in range(x + 1, len(headers)):
                            pairs.append((x, y))

                elif method == "1_vs_x":
                    pairs = []
                    for x in range(1, len(headers)):
                        pairs.append((0, x))

                # print matching columns
                elif method == "matched":
                    pairs = []
                    for x in range(len(headers) - 1):
                        for y in range(x + 1, len(headers)):
                            if headers[x] == headers[y]:
                                pairs.append((x, y))
                                break

                w = int(math.ceil(math.sqrt(len(pairs))))
                h = int(math.ceil(float(len(pairs)) / w))

                PosInf = 1e300000
                NegInf = -1e300000

                xlabel, ylabel = options.labels.split(",")

                R("""layout(matrix(seq(1,%i), %i, %i, byrow = TRUE))""" %
                  (w * h, w, h))
                for a, b in pairs:
                    new_matrix = [x for x in zip(
                        list(matrix[a].values())[0],
                        list(matrix[b].values())[0])
                                  if x[0] not in (float("nan"), PosInf, NegInf) and
                                  x[1] not in (float("nan"), PosInf, NegInf)]
                    try:
                        R("""plot(matrix[,%i], matrix[,%i], main='%s versus %s', cex=0.5, pch=".", xlab='%s', ylab='%s' )""" % (
                            a + 1, b + 1, headers[b], headers[a], xlabel, ylabel))
                    except rpy.RException as msg:
                        print("could not plot %s versus %s: %s" % (headers[b], headers[a], msg))

        if options.hardcopy:
            R['dev.off']()

    E.info("matrix added as >matrix< in R.")

    if not options.hardcopy:
        if options.input_filename:
            interpreter = code.InteractiveConsole(globals())
            interpreter.interact()
        else:
            E.info(
                "can not start new interactive session as input has come from stdin.")

    E.Stop()
Exemplo n.º 4
0
def main( argv = None ):
    
    if argv == None: argv = sys.argv

    parser = E.OptionParser( version = "%prog version: $Id: r_table2scatter.py 2782 2009-09-10 11:40:29Z andreas $")

    parser.add_option("-c", "--columns", dest="columns", type="string",
                      help="columns to take from table. Choices are 'all', 'all-but-first' or a ','-separated list of columns." )
    
    parser.add_option( "--logscale", dest="logscale", type="string",
                      help="log-transform one or both axes [default=%Default]."  )

    parser.add_option("-a", "--hardcopy", dest="hardcopy", type="string",
                      help="write hardcopy to file [default=%default].", 
                      metavar = "FILE" )

    parser.add_option("-f", "--file", dest="input_filename", type="string",
                      help="filename with table data [default=%default].",
                      metavar = "FILE")

    parser.add_option("-2", "--file2", dest="input_filename2", type="string",
                      help="additional data file [default=%default].",
                      metavar = "FILE")

    parser.add_option("-s", "--stats", dest="statistics", type="choice",
                      choices=("correlation", "spearman", "pearson", "count"),
                      help="statistical quantities to compute [default=%default]",
                      action = "append")
    
    parser.add_option("-p", "--plot", dest="plot", type="choice",
                      choices=("scatter", "pairs", "panel", "bar", "bar-stacked", 
                               "bar-besides", "1_vs_x", "matched", "boxplot", "scatter+marginal",
                               "scatter-regression" ),
                      help="plots to plot [default=%default]",
                      action = "append")

    parser.add_option("-t", "--threshold", dest="threshold", type="float",
                      help="min threshold to use for counting method [default=%default].")

    parser.add_option("-o", "--colours", dest="colours", type="int",
                      help="column with colour information [default=%default].")

    parser.add_option("-l", "--labels", dest="labels", type="string",
                      help="column labels for x and y in matched plots [default=%default].")

    parser.add_option("-d", "--add-diagonal", dest="add_diagonal", action="store_true",
                      help="add diagonal to plot [default=%default].")
    
    parser.add_option("-e", "--legend", dest="legend", type="int",
                      help="column with legend [default=%default].")

    parser.add_option("-r", "--options", dest="r_options", type="string",
                      help="R plotting options [default=%default].")
    
    parser.add_option("--format", dest="format", type="choice",
                      choices=("full", "sparse"),
                      help="output format [default=%default]." )

    parser.add_option( "--title", dest="title", type="string",
                       help="""plot title [default=%default].""")

    parser.add_option("", "--xrange", dest="xrange", type="string",
                      help="x viewing range of plot [default=%default]."  )

    parser.add_option("", "--yrange", dest="yrange", type="string",
                      help="y viewing range of plot[default=%default]."  )

    parser.add_option( "--allow-empty", dest="fail_on_empty", action="store_false",
                      help="do not fail on empty input [default=%default].")

    parser.add_option( "--fail-on-empty", dest="fail_on_empty", action="store_true",
                      help="fail on empty input [default=%default].")

    parser.set_defaults( \
        hardcopy = None,
        input_filename = "",
        input_filename2 = None,
        columns = "all",
        logscale = None,
        statistics = [],
        plot=[],
        threshold=0.0,
        labels = "x,y",
        colours= None,
        diagonal = False,
        legend = None,
        title = None,
        xrange = None,
        yrange = None,        
        r_options = "",
        fail_on_empty = True,
        format = "full")

    (options, args) = E.Start( parser )

    if len(args) == 1 and not options.input_filename:
        options.input_filename = args[0]

    if options.columns not in ("all", "all-but-first"):
        options.columns = map( lambda x: int(x)-1, options.columns.split(","))
        
    if options.colours: options.colours -= 1
    if options.legend: options.legend -= 1
    
    table ={}
    headers = []

    ## read data matrix
    if options.input_filename:
        lines = open(options.input_filename, "r").readlines()
    else:
        ## note: this will not work for interactive viewing, but
        ## creating hardcopy plots works.
        lines = sys.stdin.readlines()

    lines = filter( lambda x: x[0] != "#", lines)
    
    if len(lines) == 0:
        if options.fail_on_empty:
            raise IOError ( "no input" )
        E.warn( "empty input" )
        E.Stop()
        return

    matrix, headers, colours, legend = readTable( lines,
                                                  "matrix",
                                                  take_columns = options.columns,
                                                  headers=True,
                                                  colours=options.colours,
                                                  row_names = options.legend )

    if options.input_filename2:
        ## read another matrix (should be of the same format.
        matrix2, headers2, colours2, legend2 = readTable( lines,
                                                 "matrix2",
                                                 take_columns = options.columns,
                                                 headers=True,
                                                 colours=options.colours,
                                                 row_names = options.legend )
    R.assign("headers", headers)

    ndata = R( """length( matrix[,1] )""" )[0]

    if options.loglevel >=1:
        options.stdlog.write("# read matrix: %ix%i\n" % (len(headers),ndata) )

    if colours:
        R.assign("colours", colours)

    for method in options.statistics:

        if method == "correlation":
            cor = R.cor(matrix, use="pairwise.complete.obs" )
            writeMatrix( sys.stdout, cor, headers=headers, format = "%5.2f" )

        elif method == "pearson":
            options.stdout.write( "\t".join( ("var1", 
                                              "var2",
                                              "coeff",
                                              "passed",
                                              "pvalue",
                                              "n",
                                              "method",
                                              "alternative" )) + "\n" )
            for x in range(len(headers)-1):
                for y in range( x+1, len(headers)):
                    try:
                        result = R("""cor.test( matrix[,%i], matrix[,%i] )""" % (x + 1, y + 1))
                    except rpy.RPyException, msg:
                        E.warn( "correlation not computed for columns %i(%s) and %i(%s): %s" % (x, headers[x], y, headers[y], msg) )
                        options.stdout.write( "%s\t%s\t%s\t%s\t%s\t%i\t%s\t%s\n" % \
                                                  (headers[x], headers[y],
                                                   "na",
                                                   "na",
                                                   "na",
                                                   0,
                                                   "na",
                                                   "na" ))

                    else:
                        options.stdout.write( "%s\t%s\t%6.4f\t%s\t%e\t%i\t%s\t%s\n" % \
                                                  (headers[x], headers[y],
                                                   result.rx2('estimate').rx2('cor')[0], 
                                                   Stats.getSignificance( float(result.rx2('p.value')[0]) ),
                                                   result.rx2('p.value')[0],
                                                   result.rx2('parameter').rx2('df')[0],
                                                   result.rx2('method')[0], 
                                                   result.rx2('alternative')[0]) )

        elif method == "spearman":
            options.stdout.write( "\t".join( ("var1", "var2",
                                              "coeff",
                                              "passed",
                                              "pvalue",
                                              "method",
                                              "alternative" )) + "\n" )
            for x in range(len(headers)-1):
                for y in range( x+1, len(headers)):
                    result = R("""cor.test( matrix[,%i], matrix[,%i], method='spearman' )""" % (x + 1, y + 1))
                    options.stdout.write( "%s\t%s\t%6.4f\t%s\t%e\t%i\t%s\t%s\n" % \
                                              (headers[x], headers[y],
                                               result['estimate']['rho'], 
                                               Stats.getSignificance( float(result['p.value']) ),
                                               result['p.value'],
                                               result['parameter']['df'],
                                               result['method'], 
                                               result['alternative']))
Exemplo n.º 5
0
def main(argv=None):

    if argv is None:
        argv = sys.argv

    parser = E.OptionParser(
        version=
        "%prog version: $Id: r_table2scatter.py 2782 2009-09-10 11:40:29Z andreas $"
    )

    parser.add_option(
        "-c",
        "--columns",
        dest="columns",
        type="string",
        help=
        "columns to take from table. Choices are 'all', 'all-but-first' or a ','-separated list of columns."
    )

    parser.add_option(
        "--logscale",
        dest="logscale",
        type="string",
        help="log-transform one or both axes [default=%Default].")

    parser.add_option("-a",
                      "--hardcopy",
                      dest="hardcopy",
                      type="string",
                      help="write hardcopy to file [default=%default].",
                      metavar="FILE")

    parser.add_option("-f",
                      "--file",
                      dest="input_filename",
                      type="string",
                      help="filename with table data [default=%default].",
                      metavar="FILE")

    parser.add_option("-2",
                      "--file2",
                      dest="input_filename2",
                      type="string",
                      help="additional data file [default=%default].",
                      metavar="FILE")

    parser.add_option(
        "-s",
        "--stats",
        dest="statistics",
        type="choice",
        choices=("correlation", "spearman", "pearson", "count"),
        help="statistical quantities to compute [default=%default]",
        action="append")

    parser.add_option("-p",
                      "--plot",
                      dest="plot",
                      type="choice",
                      choices=("scatter", "pairs", "panel", "bar",
                               "bar-stacked", "bar-besides", "1_vs_x",
                               "matched", "boxplot", "scatter+marginal",
                               "scatter-regression"),
                      help="plots to plot [default=%default]",
                      action="append")

    parser.add_option(
        "-t",
        "--threshold",
        dest="threshold",
        type="float",
        help="min threshold to use for counting method [default=%default].")

    parser.add_option(
        "-o",
        "--colours",
        dest="colours",
        type="int",
        help="column with colour information [default=%default].")

    parser.add_option(
        "-l",
        "--plot-labels",
        dest="labels",
        type="string",
        help="column labels for x and y in matched plots [default=%default].")

    parser.add_option("-d",
                      "--add-diagonal",
                      dest="add_diagonal",
                      action="store_true",
                      help="add diagonal to plot [default=%default].")

    parser.add_option("-e",
                      "--plot-legend",
                      dest="legend",
                      type="int",
                      help="column with legend [default=%default].")

    parser.add_option("-r",
                      "--options",
                      dest="r_options",
                      type="string",
                      help="R plotting options [default=%default].")

    parser.add_option("--format",
                      dest="format",
                      type="choice",
                      choices=("full", "sparse"),
                      help="output format [default=%default].")

    parser.add_option("--title",
                      dest="title",
                      type="string",
                      help="""plot title [default=%default].""")

    parser.add_option("",
                      "--xrange",
                      dest="xrange",
                      type="string",
                      help="x viewing range of plot [default=%default].")

    parser.add_option("",
                      "--yrange",
                      dest="yrange",
                      type="string",
                      help="y viewing range of plot[default=%default].")

    parser.add_option("--allow-empty-file",
                      dest="fail_on_empty",
                      action="store_false",
                      help="do not fail on empty input [default=%default].")

    parser.add_option("--fail-on-empty",
                      dest="fail_on_empty",
                      action="store_true",
                      help="fail on empty input [default=%default].")

    parser.set_defaults(hardcopy=None,
                        input_filename="",
                        input_filename2=None,
                        columns="all",
                        logscale=None,
                        statistics=[],
                        plot=[],
                        threshold=0.0,
                        labels="x,y",
                        colours=None,
                        diagonal=False,
                        legend=None,
                        title=None,
                        xrange=None,
                        yrange=None,
                        r_options="",
                        fail_on_empty=True,
                        format="full")

    (options, args) = E.Start(parser)

    if len(args) == 1 and not options.input_filename:
        options.input_filename = args[0]

    if options.columns not in ("all", "all-but-first"):
        options.columns = [int(x) - 1 for x in options.columns.split(",")]

    if options.colours:
        options.colours -= 1
    if options.legend:
        options.legend -= 1

    table = {}
    headers = []

    # read data matrix
    if options.input_filename:
        lines = IOTools.openFile(options.input_filename, "r").readlines()
    else:
        # note: this will not work for interactive viewing, but
        # creating hardcopy plots works.
        lines = sys.stdin.readlines()

    lines = [x for x in lines if x[0] != "#"]

    if len(lines) == 0:
        if options.fail_on_empty:
            raise IOError("no input")
        E.warn("empty input")
        E.Stop()
        return

    matrix, headers, colours, legend = readTable(lines,
                                                 "matrix",
                                                 take_columns=options.columns,
                                                 headers=True,
                                                 colours=options.colours,
                                                 row_names=options.legend)

    if options.input_filename2:
        # read another matrix (should be of the same format.
        matrix2, headers2, colours2, legend2 = readTable(
            lines,
            "matrix2",
            take_columns=options.columns,
            headers=True,
            colours=options.colours,
            row_names=options.legend)

    R.assign("headers", headers)

    ndata = R("""length( matrix[,1] )""")[0]

    if options.loglevel >= 1:
        options.stdlog.write("# read matrix: %ix%i\n" % (len(headers), ndata))

    if colours:
        R.assign("colours", colours)

    for method in options.statistics:

        if method == "correlation":
            cor = R.cor(matrix, use="pairwise.complete.obs")
            writeMatrix(sys.stdout, cor, headers=headers, format="%5.2f")

        elif method == "pearson":
            options.stdout.write("\t".join(("var1", "var2", "coeff", "passed",
                                            "pvalue", "n", "method",
                                            "alternative")) + "\n")
            for x in range(len(headers) - 1):
                for y in range(x + 1, len(headers)):
                    try:
                        result = R("""cor.test( matrix[,%i], matrix[,%i] )""" %
                                   (x + 1, y + 1))
                    except rpy.RPyException as msg:
                        E.warn(
                            "correlation not computed for columns %i(%s) and %i(%s): %s"
                            % (x, headers[x], y, headers[y], msg))
                        options.stdout.write(
                            "%s\t%s\t%s\t%s\t%s\t%i\t%s\t%s\n" %
                            (headers[x], headers[y], "na", "na", "na", 0, "na",
                             "na"))

                    else:
                        options.stdout.write(
                            "%s\t%s\t%6.4f\t%s\t%e\t%i\t%s\t%s\n" %
                            (headers[x], headers[y],
                             result.rx2('estimate').rx2('cor')[0],
                             Stats.getSignificance(
                                 float(result.rx2('p.value')[0])),
                             result.rx2('p.value')[0],
                             result.rx2('parameter').rx2('df')[0],
                             result.rx2('method')[0],
                             result.rx2('alternative')[0]))

        elif method == "spearman":
            options.stdout.write("\t".join(("var1", "var2", "coeff", "passed",
                                            "pvalue", "method",
                                            "alternative")) + "\n")
            for x in range(len(headers) - 1):
                for y in range(x + 1, len(headers)):
                    result = R(
                        """cor.test( matrix[,%i], matrix[,%i], method='spearman')"""
                        % (x + 1, y + 1))
                    options.stdout.write(
                        "%s\t%s\t%6.4f\t%s\t%e\t%i\t%s\t%s\n" %
                        (headers[x], headers[y], result['estimate']['rho'],
                         Stats.getSignificance(float(result['p.value'])),
                         result['p.value'], result['parameter']['df'],
                         result['method'], result['alternative']))

        elif method == "count":
            # number of shared elements > threshold
            m, r, c = MatlabTools.ReadMatrix(open(options.input_filename, "r"),
                                             take=options.columns,
                                             headers=True)
            mask = numpy.greater(m, options.threshold)
            counts = numpy.dot(numpy.transpose(mask), mask)
            writeMatrix(options.stdout, counts, headers=c, format="%i")

    if options.plot:

        # remove columns that are completely empty
        if "pairs" in options.plot:
            colsums = R('''colSums( is.na(matrix ))''')
            take = [x for x in range(len(colsums)) if colsums[x] != ndata]
            if take:
                E.warn("removing empty columns %s before plotting" % str(take))
                matrix = R.subset(matrix, select=[x + 1 for x in take])
                R.assign("""matrix""", matrix)
                headers = [headers[x] for x in take]
                if legend:
                    legend = [headers[x] for x in take]

        if options.r_options:
            extra_options = ", %s" % options.r_options
        else:
            extra_options = ""

        if options.legend is not None and len(legend):
            extra_options += ", legend=c('%s')" % "','".join(legend)

        if options.labels:
            xlabel, ylabel = options.labels.split(",")
            extra_options += ", xlab='%s', ylab='%s'" % (xlabel, ylabel)
        else:
            xlabel, ylabel = "", ""

        if options.colours:
            extra_options += ", col=colours"

        if options.logscale:
            extra_options += ", log='%s'" % options.logscale

        if options.xrange:
            extra_options += ", xlim=c(%f,%f)" % tuple(
                map(float, options.xrange.split(",")))

        if options.yrange:
            extra_options += ", ylim=c(%f,%f)" % tuple(
                map(float, options.yrange.split(",")))

        if options.hardcopy:
            if options.hardcopy.endswith(".eps"):
                R.postscript(options.hardcopy)
            elif options.hardcopy.endswith(".png"):
                R.png(options.hardcopy, width=1024, height=768, type="cairo")
            elif options.hardcopy.endswith(".jpg"):
                R.jpg(options.hardcopy, width=1024, height=768, type="cairo")

        for method in options.plot:

            if ndata < 100:
                point_size = "1"
                pch = "o"
            elif ndata < 1000:
                point_size = "1"
                pch = "o"
            else:
                point_size = "0.5"
                pch = "."

            if method == "scatter":
                R("""plot( matrix[,1], matrix[,2], cex=%s, pch="o" %s)""" %
                  (point_size, extra_options))

            if method == "scatter-regression":
                R("""plot( matrix[,1], matrix[,2], cex=%s, pch="o" %s)""" %
                  (point_size, extra_options))
                dat = R(
                    """dat <- data.frame(x = matrix[,1], y = matrix[,2])""")
                R("""new <- data.frame(x = seq( min(matrix[,1]), max(matrix[,1]), (max(matrix[,1]) - min(matrix[,1])) / 100))"""
                  )
                mod = R("""mod <- lm( y ~ x, dat)""")
                R("""predict(mod, new, se.fit = TRUE)""")
                R("""pred.w.plim <- predict(mod, new, interval="prediction")"""
                  )
                R("""pred.w.clim <- predict(mod, new, interval="confidence")"""
                  )
                R("""matpoints(new$x,cbind(pred.w.clim, pred.w.plim[,-1]), lty=c(1,2,2,3,3), type="l")"""
                  )
                R.mtext("y = %f * x + %f, r=%6.4f, n=%i" %
                        (mod["coefficients"]["x"],
                         mod["coefficients"]["(Intercept)"],
                         R("""cor( dat )[2]"""), ndata),
                        3,
                        cex=1.0)

            elif method == "pairs":
                if options.add_diagonal:
                    R("""panel.hist <- function( x,y,...  ) { points(x,y,...); abline(0,1); }"""
                      )
                else:
                    R("""panel.hist <- function( x,y,...  ) { points(x,y,...); }"""
                      )

                # There used to be a argument na_action="na.omit", but
                # removed this as there appeared error messages saying
                # "na.action is not a graphical parameter" and the
                # plots showed occasionally the wrong scale.
                # cex=point_size also caused trouble (error message:
                # "X11 used font size 8 when 2 was requested" or
                # similar)
                if options.colours:
                    R.pairs(matrix,
                            pch=pch,
                            col=colours,
                            main=options.title,
                            panel="panel.hist",
                            labels=headers,
                            cex_labels=2.0)
                else:
                    R.pairs(matrix,
                            pch=pch,
                            panel="panel.hist",
                            main=options.title,
                            labels=headers,
                            cex_labels=2.0)

            elif method == "boxplot":
                extra_options += ",main='%s'" % options.title

                # set vertical orientation
                if max([len(x) for x in headers]) > 40 / len(headers):
                    # remove xlabel:
                    extra_options = re.sub(", xlab='[^']+'", "", extra_options)
                    extra_options += ", names.arg=headers, las=2"
                    R("""op <- par(mar=c(11,4,4,2))"""
                      )  # the 10 allows the names.arg below the barplot

                R("""boxplot( matrix %s)""" % extra_options)

            elif method == "bar" or method == "bar-stacked":
                if not options.colours:
                    extra_options += ", col=rainbow(5)"

                # set vertical orientation
                if max([len(x) for x in headers]) > 40 / len(headers):
                    # remove xlabel:
                    extra_options = re.sub(", xlab='[^']+'", "", extra_options)
                    extra_options += ", names.arg=headers, las=2"
                    R("""op <- par(mar=c(11,4,4,2))"""
                      )  # the 10 allows the names.arg below the barplot

                R("""barplot(as.matrix(matrix), %s)""" % extra_options)

            elif method == "bar-besides":
                if not options.colours:
                    extra_options += ", col=rainbow(%i)" % ndata

                # set vertical orientation
                if max([len(x) for x in headers]) > 40 / len(headers):
                    # remove xlabel:
                    extra_options = re.sub(", xlab='[^']+'", "", extra_options)
                    extra_options += ", names.arg=headers, las=2"
                    R("""op <- par(mar=c(11,4,4,2))"""
                      )  # the 10 allows the names.arg below the barplot

                R("""barplot(as.matrix(matrix), beside=TRUE %s)""" %
                  extra_options)

            elif method == "scatter+marginal":

                if options.title:
                    # set the size of the outer margins - the title needs to be added at the end
                    # after plots have been created
                    R.par(oma=R.c(0, 0, 4, 0))

                R("""matrix""")
                R("""
x <- matrix[,1];
y <- matrix[,2];
xhist <- hist(x, breaks=20, plot=FALSE);
yhist <- hist(y, breaks=20, plot=FALSE);
top <- max(c(xhist$counts, yhist$counts));
nf <- layout(matrix(c(2,0,1,3),2,2,byrow=TRUE), c(3,1), c(1,3), respect=TRUE );
par(mar=c(3,3,1,1)) ;
plot(x, y, cex=%s, pch="o" %s) ;
par(mar=c(0,3,1,1)) ;
barplot(xhist$counts, axes=FALSE, ylim=c(0, top), space=0 ) ;
par(mar=c(3,0,1,1)) ;
title(main='%s');
barplot(yhist$counts, axes=FALSE, xlim=c(0, top), space=0, horiz=TRUE ) ;
title(main='%s');
""" % (point_size, extra_options, xlabel, ylabel))

                if options.title:
                    R.mtext(options.title, 3, outer=True, line=1, cex=1.5)

            elif method in ("panel", "1_vs_x", "matched"):

                if method == "panel":
                    pairs = []
                    for x in range(len(headers) - 1):
                        for y in range(x + 1, len(headers)):
                            pairs.append((x, y))

                elif method == "1_vs_x":
                    pairs = []
                    for x in range(1, len(headers)):
                        pairs.append((0, x))

                # print matching columns
                elif method == "matched":
                    pairs = []
                    for x in range(len(headers) - 1):
                        for y in range(x + 1, len(headers)):
                            if headers[x] == headers[y]:
                                pairs.append((x, y))
                                break

                w = int(math.ceil(math.sqrt(len(pairs))))
                h = int(math.ceil(float(len(pairs)) / w))

                PosInf = 1e300000
                NegInf = -1e300000

                xlabel, ylabel = options.labels.split(",")

                R("""layout(matrix(seq(1,%i), %i, %i, byrow = TRUE))""" %
                  (w * h, w, h))
                for a, b in pairs:
                    new_matrix = [
                        x for x in zip(
                            list(matrix[a].values())[0],
                            list(matrix[b].values())[0])
                        if x[0] not in (float("nan"), PosInf, NegInf)
                        and x[1] not in (float("nan"), PosInf, NegInf)
                    ]
                    try:
                        R("""plot(matrix[,%i], matrix[,%i], main='%s versus %s', cex=0.5, pch=".", xlab='%s', ylab='%s' )"""
                          % (a + 1, b + 1, headers[b], headers[a], xlabel,
                             ylabel))
                    except rpy.RException as msg:
                        print("could not plot %s versus %s: %s" %
                              (headers[b], headers[a], msg))

        if options.hardcopy:
            R['dev.off']()

    E.info("matrix added as >matrix< in R.")

    if not options.hardcopy:
        if options.input_filename:
            interpreter = code.InteractiveConsole(globals())
            interpreter.interact()
        else:
            E.info(
                "can not start new interactive session as input has come from stdin."
            )

    E.Stop()
Exemplo n.º 6
0
def main(argv=None):

    if argv is None:
        argv = sys.argv

    parser = E.OptionParser(
        version=
        "%prog version: $Id: r_table2scatter.py 2782 2009-09-10 11:40:29Z andreas $"
    )

    parser.add_option(
        "-c",
        "--columns",
        dest="columns",
        type="string",
        help=
        "columns to take from table. Choices are 'all', 'all-but-first' or a ','-separated list of columns."
    )

    parser.add_option(
        "--logscale",
        dest="logscale",
        type="string",
        help="log-transform one or both axes [default=%Default].")

    parser.add_option("-a",
                      "--hardcopy",
                      dest="hardcopy",
                      type="string",
                      help="write hardcopy to file [default=%default].",
                      metavar="FILE")

    parser.add_option("-f",
                      "--file",
                      dest="input_filename",
                      type="string",
                      help="filename with table data [default=%default].",
                      metavar="FILE")

    parser.add_option("-2",
                      "--file2",
                      dest="input_filename2",
                      type="string",
                      help="additional data file [default=%default].",
                      metavar="FILE")

    parser.add_option(
        "-s",
        "--stats",
        dest="statistics",
        type="choice",
        choices=("correlation", "spearman", "pearson", "count"),
        help="statistical quantities to compute [default=%default]",
        action="append")

    parser.add_option("-p",
                      "--plot",
                      dest="plot",
                      type="choice",
                      choices=("scatter", "pairs", "panel", "bar",
                               "bar-stacked", "bar-besides", "1_vs_x",
                               "matched", "boxplot", "scatter+marginal",
                               "scatter-regression"),
                      help="plots to plot [default=%default]",
                      action="append")

    parser.add_option(
        "-t",
        "--threshold",
        dest="threshold",
        type="float",
        help="min threshold to use for counting method [default=%default].")

    parser.add_option(
        "-o",
        "--colours",
        dest="colours",
        type="int",
        help="column with colour information [default=%default].")

    parser.add_option(
        "-l",
        "--plot-labels",
        dest="labels",
        type="string",
        help="column labels for x and y in matched plots [default=%default].")

    parser.add_option("-d",
                      "--add-diagonal",
                      dest="add_diagonal",
                      action="store_true",
                      help="add diagonal to plot [default=%default].")

    parser.add_option("-e",
                      "--plot-legend",
                      dest="legend",
                      type="int",
                      help="column with legend [default=%default].")

    parser.add_option("-r",
                      "--options",
                      dest="r_options",
                      type="string",
                      help="R plotting options [default=%default].")

    parser.add_option("--format",
                      dest="format",
                      type="choice",
                      choices=("full", "sparse"),
                      help="output format [default=%default].")

    parser.add_option("--title",
                      dest="title",
                      type="string",
                      help="""plot title [default=%default].""")

    parser.add_option("",
                      "--xrange",
                      dest="xrange",
                      type="string",
                      help="x viewing range of plot [default=%default].")

    parser.add_option("",
                      "--yrange",
                      dest="yrange",
                      type="string",
                      help="y viewing range of plot[default=%default].")

    parser.add_option("--allow-empty-file",
                      dest="fail_on_empty",
                      action="store_false",
                      help="do not fail on empty input [default=%default].")

    parser.add_option("--fail-on-empty",
                      dest="fail_on_empty",
                      action="store_true",
                      help="fail on empty input [default=%default].")

    parser.set_defaults(hardcopy=None,
                        input_filename="",
                        input_filename2=None,
                        columns="all",
                        logscale=None,
                        statistics=[],
                        plot=[],
                        threshold=0.0,
                        labels="x,y",
                        colours=None,
                        diagonal=False,
                        legend=None,
                        title=None,
                        xrange=None,
                        yrange=None,
                        r_options="",
                        fail_on_empty=True,
                        format="full")

    (options, args) = E.Start(parser)

    if len(args) == 1 and not options.input_filename:
        options.input_filename = args[0]

    if options.columns not in ("all", "all-but-first"):
        options.columns = map(lambda x: int(x) - 1, options.columns.split(","))

    if options.colours:
        options.colours -= 1
    if options.legend:
        options.legend -= 1

    table = {}
    headers = []

    # read data matrix
    if options.input_filename:
        lines = open(options.input_filename, "r").readlines()
    else:
        # note: this will not work for interactive viewing, but
        # creating hardcopy plots works.
        lines = sys.stdin.readlines()

    lines = filter(lambda x: x[0] != "#", lines)

    if len(lines) == 0:
        if options.fail_on_empty:
            raise IOError("no input")
        E.warn("empty input")
        E.Stop()
        return

    matrix, headers, colours, legend = readTable(lines,
                                                 "matrix",
                                                 take_columns=options.columns,
                                                 headers=True,
                                                 colours=options.colours,
                                                 row_names=options.legend)

    if options.input_filename2:
        # read another matrix (should be of the same format.
        matrix2, headers2, colours2, legend2 = readTable(
            lines,
            "matrix2",
            take_columns=options.columns,
            headers=True,
            colours=options.colours,
            row_names=options.legend)

    R.assign("headers", headers)

    ndata = R("""length( matrix[,1] )""")[0]

    if options.loglevel >= 1:
        options.stdlog.write("# read matrix: %ix%i\n" % (len(headers), ndata))

    if colours:
        R.assign("colours", colours)

    for method in options.statistics:

        if method == "correlation":
            cor = R.cor(matrix, use="pairwise.complete.obs")
            writeMatrix(sys.stdout, cor, headers=headers, format="%5.2f")

        elif method == "pearson":
            options.stdout.write("\t".join(("var1", "var2", "coeff", "passed",
                                            "pvalue", "n", "method",
                                            "alternative")) + "\n")
            for x in range(len(headers) - 1):
                for y in range(x + 1, len(headers)):
                    try:
                        result = R("""cor.test( matrix[,%i], matrix[,%i] )""" %
                                   (x + 1, y + 1))
                    except rpy.RPyException, msg:
                        E.warn(
                            "correlation not computed for columns %i(%s) and %i(%s): %s"
                            % (x, headers[x], y, headers[y], msg))
                        options.stdout.write(
                            "%s\t%s\t%s\t%s\t%s\t%i\t%s\t%s\n" %
                            (headers[x], headers[y], "na", "na", "na", 0, "na",
                             "na"))

                    else:
                        options.stdout.write(
                            "%s\t%s\t%6.4f\t%s\t%e\t%i\t%s\t%s\n" %
                            (headers[x], headers[y],
                             result.rx2('estimate').rx2('cor')[0],
                             Stats.getSignificance(
                                 float(result.rx2('p.value')[0])),
                             result.rx2('p.value')[0],
                             result.rx2('parameter').rx2('df')[0],
                             result.rx2('method')[0],
                             result.rx2('alternative')[0]))

        elif method == "spearman":
            options.stdout.write("\t".join(("var1", "var2", "coeff", "passed",
                                            "pvalue", "method",
                                            "alternative")) + "\n")
            for x in range(len(headers) - 1):
                for y in range(x + 1, len(headers)):
                    result = R(
                        """cor.test( matrix[,%i], matrix[,%i], method='spearman')"""
                        % (x + 1, y + 1))
                    options.stdout.write(
                        "%s\t%s\t%6.4f\t%s\t%e\t%i\t%s\t%s\n" %
                        (headers[x], headers[y], result['estimate']['rho'],
                         Stats.getSignificance(float(result['p.value'])),
                         result['p.value'], result['parameter']['df'],
                         result['method'], result['alternative']))
Exemplo n.º 7
0
def get_correlation(x, y, test=True):
    corr = r.cor(x, y)
    #r('gc()')
    #gc.collect()
    return corr
Exemplo n.º 8
0
def get_correlation(x, y, test=True):
  corr = r.cor(x, y)
  #r('gc()')
  #gc.collect()
  return corr
Exemplo n.º 9
0
print r.sum(valores)[0]
print r.prod(valores)[0]
print r.sort(valores)
print r.mean(valores)[0]
print r.median(valores)[0]
print r.sd(valores)[0]
print r.var(valores)[0]

valores_python = list(valores)

he = IntVector([10, 2, 23, 11, 14, 35, 46, 32, 13, 51, 27, 49])
ha = he
print r.var(he)[0]
print r.cov(ha, he)[0]
print r.cor(ha, he)[0]

#  funções

sqr = robjects.r('function(x) x^2')
print(sqr)
print(sqr(2))
print(sqr(IntVector([4])))
print(sqr(IntVector([4, 4])))

eleva3 = robjects.r('function(a){ return(a*a*a); }')
print(eleva3)
print(eleva3(2))
print(eleva3(IntVector([4])))
print(eleva3(IntVector([4, 4])))
Exemplo n.º 10
0
def get_correlation(x, y, test=True):
    corr = r.cor(x, y)
    return corr