Gradient fill beneath line chart using matplotlib?

Question

I'm trying to product a line chart with a gradient fill beneath the line. I have searched for hours online for a solution, but none of them deal specifically with what I'm looking for.

ma = average_rate(t[0], window=900, interval=60)
fig = Figure(figsize=(8.5, 1.5), dpi=100)
canvas = FigureCanvasAgg(fig)

col = '#4f81b3'
ax = fig.add_axes([0.076, 0.11, 0.88, 0.74])

dts, vals = zip(*ma)
ax.fill(dts, vals, color=col)
fig.savefig(b, format='png')

This produces the following chart:

I have tried to use colormaps, contourf, fill_between etc with code I found online, but couldn't get it working and I'm really hoping someone has a simple solution to this problem.

With much help from @Ajean, my latest code is as follows:

    # dtseries contains a list of datetime.datetime values
    # yvalues contains a corresponding list of y-axis values
    # len(dtseries) == len(yvalues)

    import numpy as np

    # Need dpi for png generation
    fig = Figure(figsize=(8.5, 2), dpi=100)
    # Create axes directly on figure [left, bottom, width, height]
    ax = fig.add_axes([0.076, 0.11, 0.88, 0.74])

    xlims = mdates.date2num([dtseries[0], dtseries[-1]])

    # Construct an image linearly increasing in y
    xv, yv = np.meshgrid(np.linspace(0,1,50), np.linspace(0,1,50))
    zv = yv

    ax.imshow(zv, cmap='PuBu', origin='lower',
              extent=[xlims[0], xlims[1], min(yvalues), max(yvalues)])

    # Erase above the data by filling with white
    ax.fill_between(dtseries, yvalues, max(yvalues), color='w')

    # Make the line plot over the top
    colr = '#325272'
    ax.plot(dtseries, yvalues, color=colr, linewidth=0.5)

    ax.set_ylim(min(yvalues), max(yvalues))

    # Render chart as png to memory
    b = BytesIO()
    fig.savefig(b, format='png')
    return b.getvalue()

This is what I get:

Answer 1

There is actually a fairly nice answer to this at this SO question, and the following borrows the main idea, but I've substituted a call to imshow rather than contourf because I think it looks smoother. I borrow the key element, which is to place the gradient over the whole image and then 'erase' above the data using fill_between .

import numpy as np
import matplotlib.pyplot as plt
import datetime
import matplotlib.dates as mdates

# Fake data using dates as requested
xdata = np.array([datetime.datetime.today()+
                 datetime.timedelta(days=1)*i for i in range(15)])
ydata = np.cumsum(np.random.uniform(size=len(xdata)))
xlims = mdates.date2num([xdata[0], xdata[-1]])

# Construct an image linearly increasing in y
xv, yv = np.meshgrid(np.linspace(0,1,50), np.linspace(0,1,50))
zv = yv

# Draw the image over the whole plot area
fig, ax = plt.subplots(figsize=(5,3))
ax.imshow(zv, cmap='YlGnBu_r', origin='lower',
          extent=[xlims[0], xlims[1], ydata.min(), ydata.max()])

# Erase above the data by filling with white
ax.fill_between(xdata, ydata, ydata.max(), color='w')

# Make the line plot over the top
ax.plot(xdata, ydata, 'b-', linewidth=2)

ax.set_ylim(ydata.min(), ydata.max())
fig.autofmt_xdate()

plt.show()

This gives me this plot: