Last modified on 01 Oct 2021.
In this note, I use df as DataFrame, s as Series.
Libraries
import pandas as pd # import pandas package
import numpy as np
Import and have a look
df = pd.read_csv('filename.csv', na_values=['none']) # "none" is missing data
df.head() # read first 5 rows
df.tail() # last 5 rows
df.head(10) # first 10 rows
Get general infos
df.info() # show dtype of dataframe
df.describe() # numerical features
df.describe(include=['O']) # categorical features
df.describe(include='all') # all types
df.shape # dataframe's shape
df.dtypes # type of each column
df.get_dtype_counts() # count the number of data types
Check distribution of values using KDE (Kernel Density Estimation),
plt.figure(figsize=(20, 5))
df['value'].plot.kde()
Get columns’ info
# LIST OF COLUMNS
df.columns
len(df.columns) # #cols?
# UNIQUE VALUES IN COL
df['col'].unique()
df['col'].unique().size #unique vals
df['col'].nunique() # number of unique vals
Counting
# Counting #elements of each class in df
df.Classes.value_counts() # give number of each 0 and 1
# count #elements each unique values in a col/series
df[col].value_counts()
Missing values
# total number of nans in df
df.isnull().sum().sum()
# #nans in each col (including zeros)
df.isnull().sum()
# #not-nans in each col
df.count()
# each row
df.count(axis=1)
# columns having the nulls (any nan)
null_columns = df.columns[df.isna().any()].tolist()
# how many?
df[null_columns].isnull().sum()
# number of rows having ALL nans
df.isna().all(axis=1).sum()
# number of columns having ALL nans
df.isna().all(axis=0).sum()
# find index of rows having ALL nans
df.index[df.isna().all(axis=1)].to_list()
# number of nans in df
df.isnull().sum().sort_values(ascending=False)
# find % of null values
(df.isnull().sum()/df.isnull().count()*100).sort_values(ascending=False)
# Visualize the locations of missing values,
import seaborn as sns
df = df.set_index('YEAR') # y-axis is YEAR
sns.heatmap(df.isnull(), cbar=False) # x-axis is columns' name
# Plot the percentage of nans w.r.t. each column (feature)
df_tmp = (df.isnull().sum()/df.isnull().count()*100).sort_values(ascending=False).to_frame(name='percentage')
df_tmp.reset_index().plot(kind='bar', x='index', y='percentage', figsize=(20,5))
plt.xlabel('features', fontsize=14)
plt.ylabel('% of nans', fontsize=14)
Duplicates
# Check if there are duplicated values?
df['col'].duplicated().any() # returns True/False
# How many duplicates? (only count the first occurs)
df['col'].duplicated().sum()
# How many (including the repeated occurs)
df['col'].duplicated(keep=False).sum()
# List all duplicated values (LONG EXECUTING!!!)
pd.concat( g for _, g in df.groupby('col') if len(g)>1 )