Seaborn scatterplot does not color correctly 'hue'

I am having some trouble with coloring my scatterplot markers. I have a simple dataframe with a value "pos" and two other values, "af_min" and "af_max". I want to color the markers based on some conditions for af_x and af_y, but since I don't any column to exploit as a hue, I created my own column "color".

       pos      af_x      af_y  color 
0  3671023  0.200000  0.333333    2.0
1  4492071  0.176471  0.333333    2.0
2  4492302  0.222222  0.285714    2.0
3  4525905  0.298246  0.234043    2.0
4  4520905  0.003334  0.234043    1.0
5  4520905  0.400098  0.000221    0.0
6  4520905  0.001134  0.714043    1.0
7  4520905  0.559008  0.010221    0.0

Now, i create a scatterplot using seaborn and a seaborn palette in this way:

sns.scatterplot(data = df, x="af_x", y="af_y", hue="color", palette = "hsv", s=40, legend=False)

But the result is the following: as you can see, one hue does not get colored, as there are only two colors, blue and red. .

Now something VERY weird happens: to circumnavigate the problem, I built my own palette ad added it to the seaborn istance. But the scatterplot instead of getting colored with the shades I picked, it gets colored with some colors I used in another script some time ago, and there's no way to change them. Here the plot: and here's the code:

           #violet      #green      #orange
 colors = ['#747FE3', '#8EE35D', '#E37346']
 sns.set_palette(sns.color_palette(colors))

 sns.scatterplot(data = df,  x="af_x", y="af_y", hue="color", s=40, legend=False)

Here I put the entire script so that you can replicate it:

import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns

lst = [[3671023, 0.200000, 0.333333], [4492071, 0.176471, 0.333333],
      [4492302, 0.222222, 0.285714], [4525905, 0.298246, 0.234043],
      [4520905, 0.003334, 0.234043], [4520905, 0.400098, 0.000221], 
      [4520905, 0.001134, 0.714043], [4520905, 0.559008, 0.010221]
      ]
df = pd.DataFrame(lst, columns =['pos', 'af_x', 'af_y'])

afMin=0.1
afMax=0.9

df['color']=np.nan
for index in df.index:
  afx=df.loc[index, "af_x"]
  afy=df.loc[index, "af_y"]
  if ((afx >= afMin and afx <= afMax) and (afy < afMin or afy > afMax)):
      df.loc[index, "color"] = 0
  elif ((afy >= afMin and afy <= afMax) and (afx < afMin or afx > afMax)):
      df.loc[index, "color"] = 1
  elif ((afy >= afMin and afy <= afMax) and (afx >= afMin or afx <= afMax)):
      df.loc[index, "color"] = 2

sns.scatterplot(data = df,  x="af_x", y="af_y", hue="color", palette = "hsv", s=40, 
legend=False)

plt.savefig("stack_why_hsv.png")

           #violet      #green      #orange
colors = ['#747FE3', '#8EE35D', '#E37346']
sns.set_palette(sns.color_palette(colors))

sns.scatterplot(data = df,  x="af_x", y="af_y", hue="color", s=40, legend=False)
plt.savefig("stack_why_personal.png")

Thanks to anyone that can help!

Solution

The problem with your first example, is that the hsv palette has the same color at its start and at its end. This is because "h" in "hsv" is a circular variable, going from 0 to 360 degrees. Matplotlib default uses 3 colors, uniformly spaced over the range of colors, so using the red from the start, the cyan from the center and again the red from the end. So, hsv isn't the most adequate color scheme in this case. See matplotlib's available colormaps and seaborn's extensions.

The hsv palette:

For your second example, sns.set_palette() sets matplotlib's color cycle, but seaborn itself doesn't always use it. When a numeric hue is given, seaborn default chooses the rocket colormap by default. From the documentation:

The default treatment of the hue (and to a lesser extent, size) semantic, if present, depends on whether the variable is inferred to represent “numeric” or “categorical” data. In particular, numeric variables are represented with a sequential colormap by default, and the legend entries show regular “ticks” with values that may or may not exist in the data.

The easiest way to use a custom palette, is directly providing it to the function (there is no need to call sns.color_palette() as seaborn palettes internally are just lists of colors):

colors = ['#747FE3', '#8EE35D', '#E37346']
sns.scatterplot(data = df,  x="af_x", y="af_y", hue="color", palette=colors, s=40)

PS: set_palette is used by scatterplot when the hue is categorical. Here is an example. I also added the preferred way to set values to a selection of rows; this would be important for large data frames. Note that the boolean operations on arrays need quite a lot of brackets here.

afMin = 0.1
afMax = 0.9

df['color'] = ""
afx = df["af_x"]
afy = df["af_y"]
df.loc[((afx >= afMin) & (afx <= afMax) & ((afy < afMin) | (afy > afMax))), "color"] = "a"
df.loc[((afy >= afMin) & (afy <= afMax) & ((afx < afMin) | (afx > afMax))), "color"] = "b"
df.loc[((afy >= afMin) & (afy <= afMax) & (afx >= afMin) & (afx <= afMax)), "color"] = "c"

colors = ['#747FE3', '#8EE35D', '#E37346']
sns.set_palette(sns.color_palette(colors))

sns.scatterplot(data=df, x="af_x", y="af_y", hue="color", s=40)