多面板时间序列的线条和填充轮廓使用matplotlib？

Question

如果我想制作如下所示的组合图像(原始来源),你能指出我需要组装的matplotlib对象吗？我一直在尝试使用AxesImage对象,我也下载了SciKits Timeseries - 但我是否需要这个,或者可以从时间模块中轻松使用strptime,mktime和strftime并滚动我自己的自定义轴？谢谢〜

Answer 1

您不应该需要任何自定义轴.Timeseries Scikit很棒,但你根本不需要它来使用matplotlib中的日期......

你可能想要使用各种函数matplotlib.dates,plot_date绘制你的值,imshow(和/或pcolor在某些情况下)来绘制各种类型的specgrams,并使用matplotlib.mlab.specgram来计算它们.

对于子图,你需要sharex在创建它时使用kwarg,以便它们共享相同的x轴.要在图形之间共享x轴时禁用某些轴上的x轴标签,您需要使用类似的东西matplotlib.pyplot.setp(ax1.get_xticklabels(), visible=False).(这是一个稍微粗糙的黑客,但它是在所有子图之间共享相同的x轴时仅在底部子图上显示x轴标签的唯一方法)要调整子图之间的间距,请参阅subplots_adjust.

希望所有这些都是有道理的...我会在今天晚些时候有时候添加一个使用所有这些的快速示例...

编辑: 所以这是一个粗略的例子.您的示例中显示的某些内容(例如多色轴标签)在matplotlib中很难做到.(不是不​​可能,但我在这里跳过它们......)

import datetime

import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt

from matplotlib import mlab

from mpl_toolkits.axes_grid1 import make_axes_locatable

def main():
    #-- Make a series of dates
    start = datetime.datetime(2010,9,15,8,0)
    end = datetime.datetime(2010,9,15,18,0)
    delta = datetime.timedelta(seconds=1)

    # Note: "time" is now an array of floats, where 1.0 corresponds
    # to one day, and 0.0 corresponds to 1900 (I think...)
    # It's _not_ an array of datetime objects!
    time = mpl.dates.drange(start, end, delta)

    num = time.size

    #-- Generate some data
    x = brownian_noise(num) 
    y = brownian_noise(num)
    z = brownian_noise(num)

    plot(x, y, z, time)
    plt.show()

def plot(x, y, z, time):
    fig = plt.figure()

    #-- Panel 1
    ax1 = fig.add_subplot(311)
    im, cbar = specgram(x, time, ax1, fig)
    ax1.set_ylabel('X Freq. (Hz)')
    ax1.set_title('Fake Analysis of Something')

    #-- Panel 2
    ax2 = fig.add_subplot(312, sharex=ax1)
    im, cbar = specgram(y, time, ax2, fig)
    ax2.set_ylabel('Y Freq. (Hz)')

    #-- Panel 3
    ax3 = fig.add_subplot(313, sharex=ax1)
    # Plot the 3 source datasets
    xline = ax3.plot_date(time, x, 'r-')
    yline = ax3.plot_date(time, y, 'b-')
    zline = ax3.plot_date(time, z, 'g-')
    ax3.set_ylabel(r'Units $(\mu \phi)$')

    # Make an invisible spacer...
    cax = make_legend_axes(ax3)
    plt.setp(cax, visible=False)

    # Make a legend
    ax3.legend((xline, yline, zline), ('X', 'Y', 'Z'), loc='center left', 
            bbox_to_anchor=(1.0, 0.5), frameon=False)

    # Set the labels to be rotated at 20 deg and aligned left to use less space
    plt.setp(ax3.get_xticklabels(), rotation=-20, horizontalalignment='left')

    # Remove space between subplots
    plt.subplots_adjust(hspace=0.0)

def specgram(x, time, ax, fig):
    """Make and plot a log-scaled spectrogram"""
    dt = np.diff(time)[0] # In days...
    fs = dt * (3600 * 24) # Samples per second

    spec_img, freq, _ = mlab.specgram(x, Fs=fs, noverlap=200)
    t = np.linspace(time.min(), time.max(), spec_img.shape[1])

    # Log scaling for amplitude values
    spec_img = np.log10(spec_img)

    # Log scaling for frequency values (y-axis)
    ax.set_yscale('log')

    # Plot amplitudes
    im = ax.pcolormesh(t, freq, spec_img)

    # Add the colorbar in a seperate axis
    cax = make_legend_axes(ax)
    cbar = fig.colorbar(im, cax=cax, format=r'$10^{%0.1f}$')
    cbar.set_label('Amplitude', rotation=-90)

    ax.set_ylim([freq[1], freq.max()])

    # Hide x-axis tick labels
    plt.setp(ax.get_xticklabels(), visible=False)

    return im, cbar

def make_legend_axes(ax):
    divider = make_axes_locatable(ax)
    legend_ax = divider.append_axes('right', 0.4, pad=0.2)
    return legend_ax

def brownian_noise(num):
    x = np.random.random(num) - 0.5
    x = np.cumsum(x)
    return x


if __name__ == '__main__':
    main()