在不同颜色的熊猫中绘制平行坐标

Question

I have a data frame like this:

    0   1   2   3   4   5   6   7   8   9   Cluster
0   0.018420    0.003357    0.002626    -0.015603   -0.009005   -0.023671   -0.016316   0.066504    -0.039526   0.037820    For
1   0.017684    0.003434    -0.003338   -0.003904   -0.021871   -0.009454   -0.013772   -0.004610   -0.006150   -0.005746   For
2   0.018857    0.003987    0.001749    -0.019840   0.011184    -0.020451   0.082434    -0.008789   0.000449    0.005445    Against
3   0.020454    0.026437    0.036899    0.027168    -0.018483   -0.001076   0.005831    -0.002117   -0.011288   0.007491    For
4   0.018006    0.005365    0.001298    -0.006953   0.017034    0.006931    0.000268    0.001615    0.016707    -0.017798   Against

Df.columns
Index([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, u'Cluster'], dtype='object')

The last column 'cluster' specifies whether an observation belong to "For" cluster or "Against" cluster.

I want to do a plot such that all observations which are for "For" are same coloured while which are for "Against" are same colour. There are 2740 obs, so some transparency in lines would also be required to give better visualisation.

I did the following , but it is plotting the same colour red to both the classes even when I am specifying the colour option.

parallel_coordinates(Y_embed,'Cluster',color=["r" if c=="For" else "g" for c in Y_embed.Cluster])

Where Y_embed is my data frame.

If I don't mention colour option it does plot two colours by default. But I want to put my own colour option.

Any advice?

Answer 1

The color argument is just a list of colors with one color for each cluster, not one color for each row. You can make a plot with green for one cluster and magenta for the other cluster like this:

parallel_coordinates(data,'Cluster',color=['g','m'])

To make the lines transparent you can use rgba values for the colors .

parallel_coordinates(data,'Cluster',color=[[1,0,0,0.2],[0,1,0,0.9]])

Here the first set is red and partially transparent and the second is green and mostly opaque.

Answer 2

You can control the transparency of the lines by repeating your calls to parallel_coordinates with a different value in each case, eg

lowColorList=["k","k","y","y"]
midColorList=["c","b","g"]
topColorList=["r"]

plt.close()
plt.gcf().clear()    
fig, ax = plt.subplots()

parallel_coordinates(lowDf, "Cat",color=lowColorList, alpha=0.1)

parallel_coordinates(midDf, "Cat", color=midColorList, alpha=0.4)

parallel_coordinates(topDf, "Cat", color=topColorList, alpha=0.9)
# remove the pandas legend
plt.gca().legend_.remove()

plt.xlabel("Each Component of X is Shown on its Own Vertical Axis")
plt.ylabel("Values")
plt.title("Finding the Optimal Value of X")
# add new legend
topHandle =    mlines.Line2D([],[], color='red',   ls="-", label="Best")
midHandleOne = mlines.Line2D([],[], color='blue',  ls="-", label="Next Best")
lowHandle =    mlines.Line2D([],[], color='black', ls="-", label="Worst")
plt.legend(handles=[topHandle, midHandleOne,lowHandle],loc=1, prop={'size':10})

In the plot, there are actually 8 categories, but to keep the legend manageable I have some of them mapped to the same colors.

Also, be aware that pandas assigns the colors from the color list in the order in which it encounters the categories, so if you have more than one color in a group of lines, you will need to use pandas.DataFrame.sort_values(...) on your category variable.