sns.reset_defaults() 
sns.set(
 rc={'figure.figsize':(7,5)},
 style="white" # nicer layout )

繪制單變量分布

如前所述，我非常喜歡分布。 直方圖和核密度分布都是可視化特定變量的關(guān)鍵特征的有效方法。 讓我們看看如何在一個(gè)圖表中為單個(gè)變量或多個(gè)變量分配生成分布。

Left chart: Histogram and kernel density estimation of “Life Ladder” for Asian countries in 2018; Ri

繪制雙變量分布

每當(dāng)我想直觀地探索兩個(gè)或多個(gè)變量之間的關(guān)系時(shí)，通常都會(huì)歸結(jié)為某種形式的散點(diǎn)圖和分布評(píng)估。 概念上相似的圖有三種變體。 在每個(gè)圖中，中心圖（散點(diǎn)圖，雙變量KDE和hexbin）有助于理解兩個(gè)變量之間的聯(lián)合頻率分布。 此外，在中心圖的右邊界和上邊界，描繪了各個(gè)變量的邊際單變量分布（作為KDE或直方圖）。

sns.jointplot(
 x='Log GDP per capita',
 y='Life Ladder',
 data=data,
 kind='scatter' # or 'kde' or 'hex' )

Seaborn jointplot with scatter, bivariate kde, and hexbin in the center graph and marginal distribut

散點(diǎn)圖

散點(diǎn)圖是一種可視化兩個(gè)變量的聯(lián)合密度分布的方法。 我們可以通過(guò)添加色相來(lái)添加第三個(gè)變量，并通過(guò)添加size參數(shù)來(lái)可視化第四個(gè)變量。

sns.scatterplot(
 x='Log GDP per capita',
 y='Life Ladder',
 data=data[data['Year'] == 2018],
 hue='Continent',
 size='Gapminder Population' ) 

# both, hue and size are optional
sns.despine() # prettier layout

Log GDP per capita against Life Ladder, colors based on the continent and size on population

小提琴圖

小提琴圖是箱形圖和籽粒密度估計(jì)值的組合。 它起著箱形圖的作用。 它顯示了跨類別變量的定量數(shù)據(jù)分布，以便可以比較那些分布。

sns.set(
 rc={'figure.figsize':(18,6)},
 style="white" ) 
sns.violinplot(
 x='Continent',
 y='Life Ladder',
 hue='Mean Log GDP per capita',
 data=data
 ) 
sns.despine()

Violin plot where we plot continents against Life Ladder, we use the Mean Log GDP per capita to grou

配對(duì)圖

Seaborn對(duì)圖在一個(gè)大網(wǎng)格中繪制了兩個(gè)變量散點(diǎn)圖的所有組合。 我通常感覺(jué)這有點(diǎn)信息過(guò)載，但是它可以幫助發(fā)現(xiàn)模式。

sns.set(
 style="white",
 palette="muted",
 color_codes=True
 ) 
sns.pairplot(
 data[data.Year == 2018][[ 'Life Ladder','Log GDP per capita', 'Social support','Healthy life expectancy at birth', 'Freedom to make life choices','Generosity', 'Perceptions of corruption', 'Positive affect', 'Negative affect','Confidence in national government', 'Mean Log GDP per capita' ]].dropna(),
 hue='Mean Log GDP per capita'
 )

Seaborn scatterplot grid where all selected variables a scattered against every other variable in th

FacetGrid

對(duì)我而言，Seaborn的FacetGrid是使用Seaborn的最令人信服的論點(diǎn)之一，因?yàn)樗箘?chuàng)建多圖變得輕而易舉。 通過(guò)對(duì)圖，我們已經(jīng)看到了FacetGrid的示例。 FacetGrid允許創(chuàng)建按變量分段的多個(gè)圖表。 例如，行可以是一個(gè)變量（人均GDP類別），列可以是另一個(gè)變量（大陸）。

它確實(shí)比我個(gè)人需要更多的自定義（即使用matplotlib），但這仍然很吸引人。

FacetGrid —折線圖

g = sns.FacetGrid(
 data.groupby(['Mean Log GDP per capita','Year','Continent'])['Life Ladder'].mean().reset_index(),
 row='Mean Log GDP per capita',
 col='Continent',
 margin_titles=True ) 
g = (g.map(plt.plot, 'Year','Life Ladder'))

Life Ladder on the Y-axis, Year on the X-axis. The grid’s columns are the continent, and the grid’s rows are the different levels of Mean Log GDP per capita. Overall things seem to be getting better for the countries with a Low Mean Log GDP per Capita in North America and the countries with a Medium or High Mean Log GDP per Capita in Europe

FacetGrid —直方圖

g = sns.FacetGrid(data, col="Continent", col_wrap=3,height=4) 
g = (g.map(plt.hist, "Life Ladder",bins=np.arange(2,9,0.5)))

FacetGrid with a histogram of LifeLadder by continent

FacetGrid —帶注釋的KDE圖

也可以向網(wǎng)格中的每個(gè)圖表添加構(gòu)面特定的符號(hào)。 在下面的示例中，我們添加平均值和標(biāo)準(zhǔn)偏差，并在該平均值處繪制一條垂直線（下面的代碼）。

Life Ladder kernel density estimation based on the continent, annotated with a mean and standard deviation

def vertical_mean_line(x, **kwargs):
    plt.axvline(x.mean(), linestyle ="--", 
                color = kwargs.get("color", "r"))
    txkw = dict(size=15, color = kwargs.get("color", "r"))

    label_x_pos_adjustment = 0.08 # this needs customization based on your data
    label_y_pos_adjustment = 5 # this needs customization based on your data
    if x.mean() < 6: # this needs customization based on your data
        tx = "mean: {:.2f}\n(std: {:.2f})".format(x.mean(),x.std())
        plt.text(x.mean() + label_x_pos_adjustment, label_y_pos_adjustment, tx, **txkw)
    else:
        tx = "mean: {:.2f}\n  (std: {:.2f})".format(x.mean(),x.std())
        plt.text(x.mean() -1.4, label_y_pos_adjustment, tx, **txkw)

_ = data.groupby(['Continent','Year'])['Life Ladder'].mean().reset_index()

g = sns.FacetGrid(_, col="Continent", height=4, aspect=0.9, col_wrap=3, margin_titles=True)
g.map(sns.kdeplot, "Life Ladder", shade=True, color='royalblue')
g.map(vertical_mean_line, "Life Ladder")

FacetGrid —熱圖

我最喜歡的繪圖類型之一是熱圖FacetGrid，即網(wǎng)格每個(gè)面中的熱圖。 這種類型的繪圖對(duì)于在一個(gè)繪圖中可視化四個(gè)維度和一個(gè)度量很有用。 該代碼有點(diǎn)麻煩，但可以根據(jù)需要快速進(jìn)行調(diào)整。 值得注意的是，這種圖表需要相對(duì)大量的數(shù)據(jù)或適當(dāng)?shù)募?xì)分，因?yàn)樗荒芎芎玫靥幚砣笔е怠?

Facet heatmap, visualizing on the outer rows a year range, outer columns the GDP per Capita, on the inner rows the level of perceived corruption and the inner columns the continents. We see that happiness increases towards the top right (i.e., high GDP per Capita and low perceived corruption). The effect of time is not definite, and some continents (Europe and North America) seem to be happier than others (Africa).

def draw_heatmap(data,inner_row, inner_col, outer_row, outer_col, values, vmin,vmax):
    sns.set(font_scale=1)
    fg = sns.FacetGrid(
        data, 
        row=outer_row,
        col=outer_col, 
        margin_titles=True
    )

    position = left, bottom, width, height = 1.4, .2, .1, .6
    cbar_ax = fg.fig.add_axes(position) 

    fg.map_dataframe(
        draw_heatmap_facet, 
        x_col=inner_col,
        y_col=inner_row, 
        values=values, 
        cbar_ax=cbar_ax,
        vmin=vmin, 
        vmax=vmax
    )

    fg.fig.subplots_adjust(right=1.3)  
    plt.show()

def draw_heatmap_facet(*args, **kwargs):
    data = kwargs.pop('data')
    x_col = kwargs.pop('x_col')
    y_col = kwargs.pop('y_col')
    values = kwargs.pop('values')
    d = data.pivot(index=y_col, columns=x_col, values=values)
    annot = round(d,4).values
    cmap = sns.color_palette("Blues",30) + sns.color_palette("Blues",30)[0::2]
    #cmap = sns.color_palette("Blues",30)
    sns.heatmap(
        d, 
        **kwargs,
        annot=annot, 
        center=0, 
        cmap=cmap, 
        linewidth=.5
    )

# Data preparation
_ = data.copy()
_['Year'] = pd.cut(_['Year'],bins=[2006,2008,2012,2018])

_['GDP per Capita'] = _.groupby(['Continent','Year'])['Log GDP per capita'].transform(
    pd.qcut,
    q=3,
    labels=(['Low','Medium','High'])
).fillna('Low')

_['Corruption'] = _.groupby(['Continent','GDP per Capita'])['Perceptions of corruption'].transform(
    pd.qcut,
    q=3,
    labels=(['Low','Medium','High'])
)

_ = _[_['Continent'] != 'Oceania'].groupby(['Year','Continent','GDP per Capita','Corruption'])['Life Ladder'].mean().reset_index()
_['Life Ladder'] = _['Life Ladder'].fillna(-10)

draw_heatmap(
    data=_,
    outer_row='Corruption',
    outer_col='GDP per Capita',
    inner_row='Year',
    inner_col='Continent',
    values='Life Ladder',
    vmin=3,
    vmax=8,
)