Identify name of default color palette being used by seaborn or matplotlib

In your first graph the cylinders is a continuous variable of type int64 and seaborn is using a single color, in this case purple, and shading it to indicate the scale of the value, so that 8 cylinders would be darker than 4. This is done on purpose so you can easily tell what is what by the shade of the color.

Once you convert to categorical halfway through there is no longer such a relationship between the cylinder values, i.e. 8 cylinders is not twice as much a 4 cylinders anymore, they are essentially two totally different categories. To avoid associating the shade of color with the scale of the variable (since the values are no longer continuous and the relationship doesn't exist) a categorical color palette will be used by default, such that every color is distinct from the other.

In order to solve your problem, you will need to cast cylinders back to int64 prior to running your final chart with

df.cylinders = df.cylinders.astype('int64')

This will restore the variable to a continuous one and will allow seaborn to use gradients of the same color to represent the size of the values and your final plot will look just like the first one.

import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
import plotly.express as px
import warnings
warnings.filterwarnings("ignore")

# download data
df = pd.read_csv("https://www.statlearning.com/s/Auto.csv")
df.head()

# remove rows with "?"
df.drop(df.index[~df.horsepower.str.isnumeric()], axis=0, inplace=True)
df['horsepower'] = pd.to_numeric(df.horsepower, errors='coerce')

# plot 1 (gives the desired color-palette)
fig = sns.PairGrid(df, vars=df.columns[~df.columns.isin(['cylinders','origin','name'])].tolist(), hue='cylinders')
plt.gcf().set_size_inches(17,15)
fig.map_diag(sns.histplot)
fig.map_upper(sns.scatterplot)
fig.map_lower(sns.kdeplot)
fig.add_legend(ncol=5, loc=1, bbox_to_anchor=(0.5, 1.05), borderaxespad=0, frameon=False);

# plot 2
# Converting column cylinder to factor before using for 'color'
df.cylinders = df.cylinders.astype('category')

# Scatter plot - Cylinders as hue
pal = ['#fdc086','#386cb0','#beaed4','#33a02c','#f0027f']
col_map = dict(zip(sorted(df.cylinders.unique()), pal))
fig = px.scatter(df, y='mpg', x='year', color='cylinders', 
                 color_discrete_map=col_map, 
                 hover_data=['name','origin'])
fig.update_layout(width=800, height=400, plot_bgcolor='#fff')
fig.update_traces(marker=dict(size=8, line=dict(width=0.2,color='DarkSlateGrey')),
                  selector=dict(mode='markers'))
fig.show()

# plot 1 run again
df.cylinders = df.cylinders.astype('int64')
fig = sns.PairGrid(df, vars=df.columns[~df.columns.isin(['cylinders','origin','name'])].tolist(), hue='cylinders')
plt.gcf().set_size_inches(17,15)
fig.map_diag(sns.histplot)
fig.map_upper(sns.scatterplot)
fig.map_lower(sns.kdeplot)
fig.add_legend(ncol=5, loc=1, bbox_to_anchor=(0.5, 1.05), borderaxespad=0, frameon=False);

Output

The specific palette you have mentioned is the cubehelix and you can get it using:

sns.cubehelix_palette()

You can get the colours using indexing:

sns.cubehelix_palette()[:]

# [[0.9312692223325372, 0.8201921796082118, 0.7971480974663592],
#  [0.8559578605899612, 0.6418993116910497, 0.6754191211563135],
#  [0.739734329496642, 0.4765280683170713, 0.5959617419736206],
#  [0.57916573903086, 0.33934576125314425, 0.5219003947563425],
#  [0.37894937987025, 0.2224702044652721, 0.41140014301575434],
#  [0.1750865648952205, 0.11840023306916837, 0.24215989137836502]]

In general, checking the official docs or in the case when you need to check some defaults of seaborn (which is the only case when you don't know what the palette is, otherwise you know because you're the one defining it), you can always check the github code (eg. here or here).

One way to get it back is with sns.set(). But that doesn't tell us the name of the color scheme.