4

thanks for taking time to click on this..!

Right I'm trying to make a heatmap..

The problem I have is I have data like this:

please see screen show of DataFrame:

g.head(11)
Out = :

           profit
a    b    
1.04 1.09  0.886
1.08 1.03  0.03
     1.05  0.17
     1.09  0.39
     1.16  2.85
1.1  1.12 -0.346
     1.14  0.34
     1.29  1.23
1.11 1.06  0.87
     1.08  1.23
     1.09 -2.86

  • These values are all unique, meaning that there is only one profit for each combination of a & b

  • I want to : plot a heatmap of x,y and the colour is the z value..

My x-coordinate is a list that goes from 1.01 to 2.0 at 0.01 intervals

My y-coordinate is a list that does the same e.g. [1.01,1.02,1.03.. 1.99, 2.0] So I make a meshgrid of this x & y

My z coordinate I want to be 'profit' from the column above, when x = a and b = y, plot the profit

But I'm having problems vectorising, extracting profit using x & y with a meshgrid and looking up profit when a = x and b = y.. I'll explain some more

Make a meshgrid with x & y For all combinations of x & y, find a & b and lookup profit, using a = x and b = y If no such combination exists, return 0 If the combination exists, if profit is +, plot a green point If profit is -, plot a red point. The bigger the profit, the brighter the green The bigger the loss, the brighter the red, Zero can be represented by eg blue

Here was an attempt as some code I did.. I've changed it round so much trying different things that it prob wont make much sense, but you will get a gist of what I've been trying to do..

def z_func(x,y):
    #print 'x = ',x
    q = plotDF[(plotDF.x == x) & (plotDF.y == y)].profit.values
    return q
    #return x + y
    #return x.astype(float) + y.astype(float)


X,Y = meshgrid(x, y) # grid of point
Z = z_func(X, Y) # evaluation of the function on the grid
im = imshow(Z,cmap=cm.RdBu) # drawing the function
colorbar(im)
show()

Guys, I've been playing round with this all day, reading stuff on line trying lookup(), also taking out the hierarahical index and just having a dataframe with x, y and profit columns.. but I keep getting silly errors, err.. I don't actually know what I'm doing!

Can anyone shed some light on this? Would be very much appreciated!!!!!

P.s. Happy New Year 2014 guys!

user3087320
  • 101
  • 8

1 Answers1

7

Instead of setting up a meshgrid, just unstack. That positions b as the columns and leaves a as the rows, with each value in its correct place and NaN in an where the a/b combination does not exist. You can fill those with 0.

Then, instead of imshow, use pcolor.

plt.pcolor(df.unstack().fillna(0))

You might need to use sort_index to get the columns and rows in sequential order, depending on df is set up.

This answer is related but not identical.

Community
  • 1
  • 1
Dan Allan
  • 34,073
  • 6
  • 70
  • 63
  • Hi thanks for the quick reply Dan, I tried this out.. yes it is much simpler! The thing is, my plot looks like a bag of shit now :) the zero mark is a cyan colour, the other colours are veryhard to see.. I have been playing around with the different colormaps.. but cant seems to find something that fits, would you know how I can make -ve values red and +values green, the thing is, the rang may be distorted, eg on + side, may range from 0.001 to 1872 and on negative side may range from - 0.26 to - 73.82 etc.. I'd like the brightness of the greens to scale between 0.001 and 1872 etc – user3087320 Jan 03 '14 at 20:06
  • usually Matplotlib does all that for you.. but with pcolor does I have to do myself? or write my own mapping function? – user3087320 Jan 03 '14 at 20:10
  • Haha, yeah, I understand the problem. Google two projects that are addressing this: seaborn and prettyplotlib. The former is an especially active project; I know for sure that the latter has good palettes for handling positive / negative data in a smart way. – Dan Allan Jan 03 '14 at 20:35
  • @user3087320 see if this answer hepls you center your colormap where it should be: http://stackoverflow.com/questions/7404116/defining-the-midpoint-of-a-colormap-in-matplotlib/20528097#20528097 – Paul H Jan 04 '14 at 00:05
  • Hi Dan: Thanks for info, I'm in process of installing prettyplotlib now, I think will deal with negative values better – user3087320 Jan 04 '14 at 12:00
  • Hi Paul: Thankyou for this link, Id actually read it before, but because is used imshow, and I had problems with the multindexing (as mentioned above) when Dan suggested to use pcolor I then ignored the post, but on 2nd read I think I could use the colour stretching technique mentioned in here.. thanks mate – user3087320 Jan 04 '14 at 12:01