matplotlib和numpy - 直方图条颜色和标准化

Question

So I have 2 problems:

1- I have a 2D histogram w/ 1D histograms along the x & y axes. These histograms total up their respective x and y values, while the main histogram totals the values in logarithmic xy bins. The code is below. I've used pcolormesh to generate the 2D histogram... and I've generated a color bar in the range vmin=1, vmax=14 ... I make these constant since I'm generating a set of these plots over a wide data range -- and I want the colors to be consistent across them.

I also want to color the 1D histogram bars according to the same normalization. I've setup a function to do the mapping, but it is stubbornly linear -- even tho I specify LogNorm for the mapping.

I've attached a couple plots that show what I think is a linear scale for the 1D histograms. Look at x-axis histogram values around 10^4 (or 10^6)... They are colored at the 1/2 way point in the color bar, not at the log scale point.

What am I doing wrong?

2- I also want to eventually normalize the 1D histograms by the bin width (xrange or yrange). However, I don't think I can do it directly in the matplotlib.hist. Maybe I should use np hist, but then I don't know how to do a matplotlib.bar plot with log scale and colored bars (again, mapping the colors I use for the 2D hist).

Here's the code:

#
# 20 Oct 2015
# Rick Sarmento
#
# Purpose:
#  Reads star particle data and creates phase plots
#  Place histograms of x and y axis along axes
#  Uses pcolormesh norm=LogNorm(vmin=1,vmax=8)
#
# Method:
#  Main plot uses np.hist2d then takes log of result
#
# Revision history
#

# ##########################################################
# Generate colors for histogram bars based on height
# This is not right! 
# ##########################################################
def colorHistOnHeight(N, patches):
    # we need to normalize the data to 0..1 for the full
    # range of the colormap
    print("N max: %.2lf"%N.max())
    fracs = np.log10(N.astype(float))/9.0 # normalize colors to the top of our scale
    print("fracs max: %.2lf"%fracs.max())
    norm = mpl.colors.LogNorm(2.0, 9.0)
    # NOTE this color mapping is different from the one below.
    for thisfrac, thispatch in zip(fracs, patches):
        color = mpl.cm.jet(thisfrac)
        thispatch.set_facecolor(color)

    return

# ##########################################################
# Generate a combo contour/density plot
# ##########################################################
def genDensityPlot(x, y, mass, pf, z, filename, xaxislabel):
    """

    :rtype : none
    """
    nullfmt = NullFormatter()

    # Plot location and size
    fig = plt.figure(figsize=(20, 20))
    ax2dhist = plt.axes(rect_2dhist)
    axHistx = plt.axes(rect_histx)
    axHisty = plt.axes(rect_histy)

    # Fix any "log10(0)" points...
    x[x == np.inf] = 0.0
    y[y == np.inf] = 0.0
    y[y > 1.0] = 1.0 # Fix any minor numerical errors that could result in y>1

    # Bin data in log-space
    xrange = np.logspace(minX,maxX,xbins)
    yrange = np.logspace(minY,maxY,ybins)
    # Note axis order: y then x
    # H is the binned data... counts normalized by star particle mass
    # TODO -- if we're looking at x = log Z, don't weight by mass * f_p... just mass!
    H, xedges, yedges = np.histogram2d(y, x, weights=mass * (1.0 - pf), # We have log bins, so we take 
                                        bins=(yrange,xrange))

    # Use the bins to find the extent of our plot
    extent = [yedges[0], yedges[-1], xedges[0], xedges[-1]]

    # levels = (5, 4, 3) # Needed for contours only... 

    X,Y=np.meshgrid(xrange,yrange) # Create a mess over our range of bins

    # Take log of the bin data
    H = np.log10(H)
    masked_array = np.ma.array(H, mask=np.isnan(H))  # mask out all nan, i.e. log10(0.0)

    # Fix colors -- white for values of 1.0. 
    cmap = copy.copy(mpl.cm.jet)
    cmap.set_bad('w', 1.)  # w is color, for values of 1.0

    # Create a plot of the binned
    cax = (ax2dhist.pcolormesh(X,Y,masked_array, cmap=cmap, norm=LogNorm(vmin=1,vmax=8)))
    print("Normalized H max %.2lf"%masked_array.max())

    # Setup the color bar
    cbar = fig.colorbar(cax, ticks=[1, 2, 4, 6, 8])
    cbar.ax.set_yticklabels(['1', '2', '4', '6', '8'], size=24)
    cbar.set_label('$log\, M_{sp, pol,\odot}$', size=30)

    ax2dhist.tick_params(axis='x', labelsize=22)
    ax2dhist.tick_params(axis='y', labelsize=22)
    ax2dhist.set_xlabel(xaxislabel, size=30)
    ax2dhist.set_ylabel('$log\, Z_{pri}/Z$', size=30)

    ax2dhist.set_xlim([10**minX,10**maxX])
    ax2dhist.set_ylim([10**minY,10**maxY])
    ax2dhist.set_xscale('log')
    ax2dhist.set_yscale('log')
    ax2dhist.grid(color='0.75', linestyle=':', linewidth=2)

    # Generate the xy axes histograms
    ylims = ax2dhist.get_ylim()
    xlims = ax2dhist.get_xlim()

    ##########################################################
    # Create the axes histograms
    ##########################################################
    # Note that even with log=True, the array N is NOT log of the weighted counts
    # Eventually we want to normalize these value (in N) by binwidth and overall
    # simulation volume... but I don't know how to do that.
    N, bins, patches = axHistx.hist(x, bins=xrange, log=True, weights=mass * (1.0 - pf))
    axHistx.set_xscale("log")
    colorHistOnHeight(N, patches)
    N, bins, patches = axHisty.hist(y, bins=yrange, log=True, weights=mass * (1.0 - pf),
                                    orientation='horizontal')
    axHisty.set_yscale('log')
    colorHistOnHeight(N, patches)

    # Setup format of the histograms
    axHistx.set_xlim(ax2dhist.get_xlim())  # Match the x range on the horiz hist
    axHistx.set_ylim([100.0,10.0**9])       # Constant range for all histograms
    axHistx.tick_params(labelsize=22)
    axHistx.yaxis.set_ticks([1e2,1e4,1e6,1e8])
    axHistx.grid(color='0.75', linestyle=':', linewidth=2)

    axHisty.set_xlim([100.0,10.0**9])       # We're rotated, so x axis is the value
    axHisty.set_ylim([10**minY,10**maxY])  # Match the y range on the vert hist
    axHisty.tick_params(labelsize=22)
    axHisty.xaxis.set_ticks([1e2,1e4,1e6,1e8])
    axHisty.grid(color='0.75', linestyle=':', linewidth=2)

    # no labels
    axHistx.xaxis.set_major_formatter(nullfmt)
    axHisty.yaxis.set_major_formatter(nullfmt)

    if z[0] == '0': z = z[1:]
    axHistx.set_title('z=' + z, size=40)

    plt.savefig(filename + "-z_" + z + ".png", dpi=fig.dpi)
    #    plt.show()
    plt.close(fig) # Release memory assoc'd with the plot
    return


# ##########################################################
# ##########################################################
##
## Main program
##
# ##########################################################
# ##########################################################
import matplotlib as mpl
import matplotlib.pyplot as plt
#import matplotlib.colors as colors # For the colored 1d histogram routine
from matplotlib.ticker import NullFormatter
from matplotlib.colors import LogNorm
from matplotlib.ticker import LogFormatterMathtext
import numpy as np
import copy as copy

files = [
    "18.00",
    "17.00",
    "16.00",
    "15.00",
    "14.00",
    "13.00",
    "12.00",
    "11.00",
    "10.00",
    "09.00",
    "08.50",
    "08.00",
    "07.50",
    "07.00",
    "06.50",
    "06.00",
    "05.50",
    "05.09"
]
# Plot parameters - global
left, width = 0.1, 0.63
bottom, height = 0.1, 0.63
bottom_h = left_h = left + width + 0.01

xbins = ybins = 100

rect_2dhist = [left, bottom, width, height]
rect_histx = [left, bottom_h, width, 0.15]
rect_histy = [left_h, bottom, 0.2, height]

prefix = "./"
# prefix="20Sep-BIG/"
for indx, z in enumerate(files):
    spZ = np.loadtxt(prefix + "spZ_" + z + ".txt", skiprows=1)
    spPZ = np.loadtxt(prefix + "spPZ_" + z + ".txt", skiprows=1)
    spPF = np.loadtxt(prefix + "spPPF_" + z + ".txt", skiprows=1)
    spMass = np.loadtxt(prefix + "spMass_" + z + ".txt", skiprows=1)

    print ("Generating phase diagram for z=%s" % z)
    minY = -4.0
    maxY = 0.5
    minX = -8.0
    maxX = 0.5
    genDensityPlot(spZ, spPZ / spZ, spMass, spPF, z,
                   "Z_PMassZ-MassHistLogNorm", "$log\, Z_{\odot}$")
    minX = -5.0
    genDensityPlot((spZ) / (1.0 - spPF), spPZ / spZ, spMass, spPF, z,
                   "Z_PMassZ1-PGF-MassHistLogNorm", "$log\, Z_{\odot}/f_{pol}$")

Here are a couple plots showing the problem with the coloring of the 1D axis histograms

Answer 1

0) Your code is very nicely (and helpfully!) documented, but it would be very helpful for you to trim down to a minimal working example.
1) the fracs array in colorHistOnHeight does not include the lower bound of 1e2.
2) The bounds of your different LogNorm colormaps are changing throughout your code (eg [1,8] vs. [2,9]). Set these parameters to variables, and pass those variables as needed.
3) Create a scalar mappable object matplotlib.cm.ScalarMappable object to convert a scalar value directly into a color using the to_rgba method.

Hope one of those helps!

Answer 2

I figured out how to use the above suggestion: matplotlib.sm.ScalarMappable . This does it! The mapping matches my colorbar scale.

# ##########################################################
# Generate colors for histogram bars based on height
# Method:
#  Take log of the histogram values (weighted counts)..
#  Create a LogNorm mapping between 1->9
#  Use the norm to map scalar values between 1 & 9 to rgb
# ##########################################################
def colorHistOnHeight(N, patches):
    cleanN = np.ma.masked_where(N == 0.0, N)
    fracs  = np.log10(cleanN) # normalize colors to the top of our scale
    norm   = mpl.colors.LogNorm(vmin=1.0, vmax=9.0)
    sm     = mpl.cm.ScalarMappable(norm=norm, cmap=mpl.cm.jet)
    sm.set_clim([1.0,9.0])
    for thisfrac, thispatch in zip(fracs, patches):
        color = sm.to_rgba(thisfrac)
        thispatch.set_facecolor(color)
    return 

Answer 3

And here's how I was able to normalize the matplotlib hist by the logarithmic bin width (and the volume of my simulation -- a scalar value). Someone please check my solution however.

yrange = np.logspace(minY,maxY,ybins)
N, bins, patches = axHisty.hist(y, bins=yrange, log=True, weights=mass * (1.0 - pf))

widths = np.diff(bins)
for item,dbx in zip(patches,widths):
    item.set_height(item.get_height()/dbx/cmvol)

I normalize the heights of the histogram rectangles by the bin width (dbx) and the comoving volume of my simulation (cmvol). I think this does it!