0

I'm trying to plot a seaborn heatmap centered on 0. So I can see the divergence of the values on the positive and negative range. The problem is that near 0 the values change too smoothly, on the other side the absolute values, either for the positive or the negative range are really large.

I'm using the code 

params = {'font.size': 18, 'xtick.labelsize':16, ytick.labelsize':16}
plt.rcParams.update(params)
plt.figure(figsize=(24,8))
mask = dfs.isnull()
cmap = sns.diverging_palette(255, 0, s=99, sep=1, as_cmap=True)
sns.heatmap(dfs,cmap=cmap, mask=mask, center=0, annot=labelS, fmt='')
plt.show()

So, I expected that positive values to be red, and the negatives to be blue. I created a 2D numpy array so I can write the signal of the data. The problem is that, when I plot it, near 0, are negative values that are still red.

The heatmap result from the code above is shown here:

Edit1: I tryed this answer: Defining the midpoint of a colormap in matplotlib, with the midpoint suggested (1 - vmax / (vmax + abs(vmin)) = 0.9481981715613719) And neither the option of shifted or shrunk cmap works, maybe if I know the best star and end point of the function it will work better, but as I don't know how to choose it different than 0 and 1, the look the same. I plot the heatmap with the actual values in the figures, so it will be better to see. enter image description here

I think because the range of values are large, and there are some between 0 and 1, while vmax=217 and vmin=-3981, the divergence isn't fast enough. And the divergence of the positive color is different from the divergence of the negative range.

Maybe there is a best value for start and end point, the problem is that how can I do it from the data?

@DaveTheScientist said that this solution fails when the midpoint is equal to 0 or 1, and he proposes a new function, my midpoint is really near 1, but his solution still didn't work for me, I got the same result as the original answer from @Paul H.

And if the value of each cell is not showed, i'm not able to differentiate the missing values, from the near 0 values.

Edit 2: Now I tryed the options 1 and 2 from here Shifted colorbar matplotlib. And there are even more problems, because of the missing values, they are getting colors as the other values. And the problems with the values near 0 still appears.

My dataset is a pandas pivot table(dfs) with missing values(nan), so it's not regular, (maybe this is the problem?)

0.2     0.3     0.4     0.5     0.6     0.7     0.8     0.85    0.9
0.59    nan     0.409447    0.403747    0.220221    0.198353    0.112916    1.70769     -188.474    -1803.68
0.57    nan     nan     0.402517    0.393347    0.206475    0.182728    2.3221  -180.016    -1814.15
0.55    nan     nan     0.403747    0.229443    0.211342    0.114802    3.02297     -203.211    -1787.95
0.53    nan     0.410742    0.404906    0.226998    0.129954    0.113135    -0.759108   -183.026    -1837.94
0.51    nan     nan     0.401211    0.39512     0.20654     0.189979    1.02238     -178.454    -1584.25
0.49    nan     nan     0.404906    0.396786    0.209138    0.117177    -0.472419   -173.616    -1750.06
0.47    nan     0.410139    0.405985    0.228938    0.128596    0.119513    2.03246     -241.663    -1751.43
0.45    nan     0.411318    0.404906    0.398352    0.206402    0.120057    -5.65621    -234.121    -1669.31
0.43    nan     nan     0.405985    0.398352    0.221434    0.203799    -3.45825    -247.245    -1540.45
0.41    nan     nan     0.409447    0.402516    0.213089    0.191273    -8.74121    -290.459    -1501.15
0.39    nan     nan     0.406968    0.402516    0.226998    0.12801     -3.31603    -260.139    -1460.85
0.37    nan     nan     0.406968    0.403747    0.227574    0.132975    -4.60131    -339.783    -1423.44
0.35    nan     nan     0.410139    0.403747    0.393347    0.209181    -10.0318    -309.951    -1264.69
0.33    nan     nan     0.410741    0.403747    0.230069    0.141903    -15.315     -366.672    -1106.8
0.31    nan     nan     0.413946    0.406968    0.24464     0.219125    -22.3067    -371.801    -3981.65
0.29    nan     nan     0.413761    0.408701    0.401211    0.223775    -19.8598    -429.873    -812.076
0.27    nan     nan     0.412663    0.405985    0.396786    0.221357    -19.9852    -420.288    -661.54
0.25    nan     nan     0.413762    0.410139    0.24378     0.220389    -45.1834    -407.948    -561.944
0.23    nan     nan     nan     0.412253    0.402516    0.229078    -83.3562    -384.158    -222.658
0.21    nan     nan     0.41405     0.412986    0.403747    0.401211    -129.443    -407.312    -8.51213
0.19    nan     nan     nan     0.414211    0.4087  0.80159     -128.428    -226.756    204.561
0.17    nan     nan     nan     0.413975    0.412986    -2.14317    -107.155    106.968     217.525
0.15    nan     nan     nan     nan     0.414136    -6.2662     107.339     154.627     128.412
0.13    nan     nan     -4.84622    16.165  31.0498     29.0702     22.9361     23.9982     18.0343
0.125   nan     -8.68298    17.7774     21.5409     21.1447     23.1349     12.5357     19.238  13.954
0.12    5.19241     20.5002     25.1215     18.5761     21.3382     15.3316     2.84265     2.37559     11.0637
0.115   26.9932     35.8498     30.6433     10.9662     -0.579587   4.75105     1.82187     5.57865     1.94192
0.11    33.4479     34.3506     22.2485     13.1617     12.6101     12.3581     1.8572  2.91308     2.91308
0.105   39.4335     37.4019     37.4019     26.5404     15.0971     15.0971     -871.345    -871.345    1.14922
0.1     30.1349     30.1349     27.678  27.678  19.0532     19.0532     8.53414     8.53414     8.53414
0.095   34.9027     34.9027     16.1035     16.1035     16.1035     14.4113     14.4113     14.4113     14.4113
0.09    32.3556     32.3556     32.3556     5.51217     5.51217     5.51217     5.51217     5.51217     5.51217
0.085   -479.572    -479.572    -479.572    1.9069  1.9069  1.9069  1.9069  1.9069  1.9069
0.08    -0.14278    -0.14278    -0.14278    -0.14278    -0.14278    -0.14278    -0.14278    -0.14278    -0.14278
0.075   4.3636  4.3636  4.3636  4.3636  4.3636  4.3636  4.3636  4.3636  4.3636
0.07    -0.103964   -0.103964   -0.103964   -0.103964   -0.103964   -0.103964   -0.103964   -0.103964   -0.103964
0.065   -3.51647    -3.51647    -3.51647    -3.51647    -3.51647    -3.51647    -3.51647    -3.51647    -3.51647
0.06    -5.33304    -5.33304    -5.33304    -5.33304    -5.33304    -5.33304    -5.33304    -5.33304    -5.33304
0.055   -1.01983    -1.01983    -1.01983    -1.01983    -1.01983    -1.01983    -1.01983    -1.01983    -1.01983
0.05    -0.0490618  -0.0490618  -0.0490618  -0.0490618  -0.0490618  -0.0490618  -0.0490618  -0.0490618  -0.0490618
0.045   -3.03683    -3.03683    -3.03683    -3.03683    -3.03683    -3.03683    -3.03683    -3.03683    -3.03683
0.04    -2.75252    -2.75252    -2.75252    -2.75252    -2.75252    -2.75252    -2.75252    -2.75252    -2.75252
0.035   -3.31855    -3.31855    -3.31855    -3.31855    -3.31855    -3.31855    -3.31855    -3.31855    -3.31855
0.03    0.0711071   0.0711071   0.0711071   0.0711071   0.0711071   0.0711071   0.0711071   0.0711071   0.0711071
0.025   0.0669509   0.0669509   0.0669509   0.0669509   0.0669509   0.0669509   0.0669509   0.0669509   0.0669509
0.02    0.0587947   0.0587947   0.0587947   0.0587947   0.0587947   0.0587947   0.0587947   0.0587947   0.0587947
0.015   -4.69897    -4.69897    -4.69897    -4.69897    -4.69897    -4.69897    -4.69897    -4.69897    -4.69897
0.01    -4.44004    -4.44004    -4.44004    -4.44004    -4.44004    -4.44004    -4.44004    -4.44004    -4.44004
0.005   -3.66247    -3.66247    -3.66247    -3.66247    -3.66247    -3.66247    -3.66247    -3.66247    -3.66247

As you can see, the divergence it's not centered on 0.

Jheni Gonsalves
  • 41
  • 7
  • Could you show the image result? – Mad Physicist Jul 18 '19 at 22:01
  • It's on the hiperlink 'The heatmap result from the code above is shown here' above the data, I didn't find a way to put it along the question – Jheni Gonsalves Jul 19 '19 at 00:01
  • I tryed again, and it says that i'm not alowed to put an image in a question until i get 100 points in reputation. So, the stackoverflow puts a hiperlink. Sorry. Here is the link: https://i.stack.imgur.com/1Yptq.png – Jheni Gonsalves Jul 19 '19 at 00:08
  • No worries, I fixed it for you – Mad Physicist Jul 19 '19 at 01:49
  • Here's something that might help, if not an outright duplicate: https://stackoverflow.com/q/7404116/2988730 – Mad Physicist Jul 19 '19 at 02:45
  • I tryed this solution https://stackoverflow.com/questions/7404116/defining-the-midpoint-of-a-colormap-in-matplotlib, for both options, shifted and shrunk cmaps, it's a little bit better, but still didn't work, i think its because the range of the values that I have. Because, in the data, there are some values between -1 and 1, and they look the same. I think the divergence of the cmap its not fast enough. I will edit the question to add this part – Jheni Gonsalves Jul 19 '19 at 03:28
  • @MadPhysicist, I tryed both solutions, that are duplicated, but both of them don't work for me – Jheni Gonsalves Jul 19 '19 at 04:49
  • Your right. This is indeed very strange. I'll have a look at the seaborn code later to see what's happening. This isn't normal behavior. – Mad Physicist Jul 19 '19 at 12:08

0 Answers0