观察数据，从整体上了解数据类型，分布，相关性

常用包：pandas，numpy,  scipy(统计学包，可完成matlab诸多功能),  seaborn(丰富的数据可视化包), matplotlib，missingno(缺失值可视化)

常用语句：

整体观察：.nunique()  .value_counts()   .info()   .describe()   .isnull().sum()   .skew()   .kurt()

缺失值可视化：missingno.matrix(data.sample())       missingno.bar(data.sample())

分布可视化：

按类别特征和数字特征分类，对数字特征分析相关性，对类别特征看分布状态

数字特征：热力图，散点图，分布，回归

f,ax = plt.subplots(figsize = (7,7))

sns.heatmap(correlation, square = True, vmax = 0.8)

f = pd.melt(data, value_vars = num)

#all distributions

g = sns.FacetGrid(f,col = 'variable',col_wrap=2,sharex=False,sharey=False)

g = g.map(sns.distplot,'value')

#散点图

sns.set()

sns.pairplot(data[['price','v_1','v_2']],diag_kind='kde', kind = 'scatter')

plt.show()

#回归

fig, (ax1,ax2) = plt.subplots(nrows = 1, ncols = 2, figsize = (10,5))

sns.regplot(x = 'v_1',y = 'price', data = data[['v_1','price']],scatter = True, fit_reg = True,ax = ax1)

sns.regplot(x = 'v_2',y = 'price', data = data[['v_2','price']],scatter = True, fit_reg = True,ax = ax2)

#类别特征箱线图，小提琴图，柱状图

for c in ['brand','gearbox']: #填补缺失值

    data[c] = data[c].astype('category')

    if data[c].isnull().any():

        data[c] = data[c].cat.add_categories(['MISSING']) #增加missing类

        data[c] = data[c].fillna('MISSING')

def boxplot(x, y, **kwargs):

    sns.boxplot(x=x, y=y)

    x=plt.xticks(rotation=90)

f = pd.melt(data, id_vars=['price'], value_vars=['brand','gearbox'])

g = sns.FacetGrid(f, col="variable",  col_wrap=2, sharex=False, sharey=False, height=5)

g = g.map(boxplot, "value", "price")

柱状图

def count_plot(x, **kwargs):

    sns.countplot(x=x)

    x=plt.xticks(rotation=90)

f = pd.melt(data,  value_vars=['brand','gearbox'])

g = sns.FacetGrid(f, col="variable",  col_wrap=2, sharex=False, sharey=False, height=5)

g = g.map(count_plot, "value")