Cheat Sheet: The pandas DataFrame Object

Preliminaries Get your data into a DataFrame

Always start by importing these Python modules Instantiate an empty DataFrame

import numpy as np df = DataFrame()
import matplotlib.pyplot as plt
import pandas as pd Load a DataFrame from a CSV file
from pandas import DataFrame, Series df = pd.read_csv('file.csv')# often works
Note: these are the recommended import aliases df = pd.read_csv('file.csv', header=0,
index_col=0, quotechar='"',sep=':',
na_values = ['na', '-', '.', ''])
Note: refer to pandas docs for all arguments
Cheat sheet conventions
Get data from inline CSV text to a DataFrame
Code examples from io import StringIO
# Code examples are found in yellow boxes data = """, Animal, Cuteness, Desirable
# These are designed to be cut and paste row-1, dog, 8.7, True
row-2, cat, 9.5, True
row-3, bat, 2.6, False"""
In the code examples, typically I use: df = pd.read_csv(StringIO(data),
• s to represent a pandas Series object; header=0, index_col=0,
• df to represent a pandas DataFrame object; skipinitialspace=True)
• idx to represent a pandas Index object. Note: skipinitialspace=True allows a pretty layout
• Also: t – tuple, l – list, b – Boolean, i – integer,
a – numpy array, st – string, d – dictionary, etc. Load DataFrames from a Microsoft Excel file
# Each Excel sheet in a Python dictionary
workbook = pd.ExcelFile('file.xlsx')
d = {} # start with an empty dictionary
The conceptual model for sheet_name in workbook.sheet_names:
df = workbook.parse(sheet_name)
DataFrame object: The pandas DataFrame is a two- d[sheet_name] = df
dimensional table of data with column and row indexes Note: the parse() method takes many arguments like
(something like a spread sheet). The columns are made read_csv() above. Refer to the pandas documentation.
up of pandas Series objects (more below).
Load a DataFrame from a MySQL database
Column index (df.columns) import pymysql
from sqlalchemy import create_engine
engine = create_engine('mysql+pymysql://'
Series of data

Series of data

Series of data

Series of data

Series of data

Series of data

Series of data

+'USER:PASSWORD@HOST/DATABASE')
(df.index)
Row index

df = pd.read_sql_table('table', engine)

Data in Series then combine into a DataFrame

# Example 1 ...
s1 = Series(range(6))
s2 = s1 * s1
s2.index = s2.index + 2# misalign indexes
df = pd.concat([s1, s2], axis=1)
A DataFrame has two Indexes:
• Typically, the column index (df.columns) is a list of # Example 2 ...
strings (observed variable names) or (less s3 = Series({'Tom':1, 'Dick':4, 'Har':9})
commonly) integers s4 = Series({'Tom':3, 'Dick':2, 'Mar':5})
• Typically, the row index (df.index) might be: df = pd.concat({'A':s3, 'B':s4 }, axis=1)
o Integers - for case or row numbers; Note: 1st method has in integer column labels
o Strings – for case names; or Note: 2nd method does not guarantee col order
o DatetimeIndex or PeriodIndex – for time series Note: index alignment on DataFrame creation

Series object: an ordered, one-dimensional array of Get a DataFrame from a Python dictionary
data with an index. All the data in a Series is of the # default --- assume data is in columns
same data type. Series arithmetic is vectorised after first df = DataFrame({
aligning the Series index for each of the operands. 'col0' : [1.0, 2.0, 3.0, 4.0],
'col1' : [100, 200, 300, 400]
s1 = Series(range(0,4)) # -> 0, 1, 2, 3 })
s2 = Series(range(1,5)) # -> 1, 2, 3, 4
s3 = s1 + s2 # -> 1, 3, 5, 7
Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
1
Get a DataFrame from data in a Python dictionary
# --- use helper method for data in rows Working with the whole DataFrame
df = DataFrame.from_dict({ # data by row
# rows as python dictionaries Peek at the DataFrame contents/structure
'row0' : {'col0':0, 'col1':'A'},
df.info() # index & data types
'row1' : {'col0':1, 'col1':'B'}
dfh = df.head(i) # get first i rows
}, orient='index')
dft = df.tail(i) # get last i rows
dfs = df.describe() # summary stats cols
df = DataFrame.from_dict({ # data by row
top_left_corner_df = df.iloc[:4, :4]
# rows as python lists
'row0' : [1, 1+1j, 'A'],
'row1' : [2, 2+2j, 'B'] DataFrame non-indexing attributes
}, orient='index') dfT = df.T # transpose rows and cols
l = df.axes # list row and col indexes
Create play/fake data (useful for testing) (r, c) = df.axes # from above
s = df.dtypes # Series column data types
# --- simple - default integer indexes
b = df.empty # True for empty DataFrame
df = DataFrame(np.random.rand(50,5))
i = df.ndim # number of axes (it is 2)
t = df.shape # (row-count, column-count)
# --- with a time-stamp row index:
i = df.size # row-count * column-count
df = DataFrame(np.random.rand(500,5))
a = df.values # get a numpy array for df
df.index = pd.date_range('1/1/2005',
periods=len(df), freq='M')
DataFrame utility methods
# --- with alphabetic row and col indexes df = df.copy() # copy a DataFrame
# and a "groupable" variable df = df.rank() # rank each col (default)
import string df = df.sort_values(by=col)
import random df = df.sort_values(by=[col1, col2])
r = 52 # note: min r is 1; max r is 52 df = df.sort_index()
c = 5 df = df.astype(dtype) # type conversion
df = DataFrame(np.random.randn(r, c),
columns = ['col'+str(i) for i in DataFrame iteration methods
range(c)], df.iteritems()# (col-index, Series) pairs
index = list((string. ascii_uppercase+ df.iterrows() # (row-index, Series) pairs
string.ascii_lowercase)[0:r]))
df['group'] = list( # example ... iterating over columns
''.join(random.choice('abcde') for (name, series) in df.iteritems():
for _ in range(r)) ) print('Col name: ' + str(name))
print('First value: ' +
str(series.iat[0]) + '\n')

Saving a DataFrame Maths on the whole DataFrame (not a complete list)

df = df.abs() # absolute values
Saving a DataFrame to a CSV file df = df.add(o) # add df, Series or value
df.to_csv('name.csv', encoding='utf-8') s = df.count() # non NA/null values
df = df.cummax() # (cols default axis)
df = df.cummin() # (cols default axis)
Saving DataFrames to an Excel Workbook
df = df.cumsum() # (cols default axis)
from pandas import ExcelWriter df = df.diff() # 1st diff (col def axis)
writer = ExcelWriter('filename.xlsx') df = df.div(o) # div by df, Series, value
df1.to_excel(writer,'Sheet1') df = df.dot(o) # matrix dot product
df2.to_excel(writer,'Sheet2') s = df.max() # max of axis (col def)
writer.save() s = df.mean() # mean (col default axis)
s = df.median()# median (col default)
Saving a DataFrame to MySQL s = df.min() # min of axis (col def)
import pymysql df = df.mul(o) # mul by df Series val
from sqlalchemy import create_engine s = df.sum() # sum axis (cols default)
e = create_engine('mysql+pymysql://' + df = df.where(df > 0.5, other=np.nan)
'USER:PASSWORD@HOST/DATABASE') Note: The methods that return a series default to
df.to_sql('TABLE',e, if_exists='replace') working on columns.
Note: if_exists à 'fail', 'replace', 'append'
DataFrame select/filter rows/cols on label values
Saving to Python objects df = df.filter(items=['a', 'b']) # by col
d = df.to_dict() # to dictionary df = df.filter(items=[5], axis=0) #by row
str = df.to_string() # to string df = df.filter(like='x') # keep x in col
m = df.as_matrix() # to numpy matrix df = df.filter(regex='x') # regex in col
df = df.select(lambda x: not x%5)#5th rows
Note: select takes a Boolean function, for cols: axis=1
Note: filter defaults to cols; select defaults to rows
Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
2
 Set column values set based on criteria
Working with Columns df['b']=df['a'].where(df['a']>0,other=0)
df['d']=df['a'].where(df.b!=0,other=df.c)
Each DataFrame column is a pandas Series object Note: where other can be a Series or a scalar

Data type conversions

Get column index and labels
st = df['col'].astype(str)# Series dtype
idx = df.columns # get col index
a = df['col'].values # numpy array
label = df.columns[0] # first col label
pl = df['col'].tolist() # python list
l = df.columns.tolist() # list col labels
Note: useful dtypes for Series conversion: int, float, str
Trap: index lost in conversion from Series to array or list
Change column labels
df.rename(columns={'old1':'new1',
Common column-wide methods/attributes
'old2':'new2'}, inplace=True)
value = df['col'].dtype # type of data
Note: can rename multiple columns at once.
value = df['col'].size # col dimensions
value = df['col'].count()# non-NA count
Selecting columns value = df['col'].sum()
s = df['colName'] # select col to Series value = df['col'].prod()
df = df[['colName']] # select col to df value = df['col'].min()
df = df[['a','b']] # select 2 or more value = df['col'].max()
df = df[['c','a','b']]# change col order value = df['col'].mean() # also median()
s = df[df.columns[0]] # select by number value = df['col'].cov(df['col2'])
df = df[df.columns[[0, 3, 4]] # by number s = df['col'].describe()
s = df.pop('c') # get col & drop from df s = df['col'].value_counts()

Selecting columns with Python attributes Find index label for min/max values in column
s = df.a # same as s = df['a'] label = df['col1'].idxmin()
# cannot create new columns by attribute label = df['col1'].idxmax()
df.existing_column = df.a / df.b
df['new_column'] = df.a / df.b
Common column element-wise methods
Trap: column names must be valid identifiers. s = df['col'].isnull()
s = df['col'].notnull() # not isnull()
Adding new columns to a DataFrame s = df['col'].astype(float)
df['new_col'] = range(len(df)) s = df['col'].abs()
df['new_col'] = np.repeat(np.nan,len(df)) s = df['col'].round(decimals=0)
df['random'] = np.random.rand(len(df)) s = df['col'].diff(periods=1)
df['index_as_col'] = df.index s = df['col'].shift(periods=1)
df1[['b','c']] = df2[['e','f']] s = df['col'].to_datetime()
df3 = df1.append(other=df2) s = df['col'].fillna(0) # replace NaN w 0
Trap: When adding an indexed pandas object as a new s = df['col'].cumsum()
column, only items from the new series that have a s = df['col'].cumprod()
corresponding index in the DataFrame will be added. s = df['col'].pct_change(periods=4)
The receiving DataFrame is not extended to s = df['col'].rolling_sum(periods=4,
accommodate the new series. To merge, see below. window=4)
Trap: when adding a python list or numpy array, the Note: also rolling_min(), rolling_max(), and many more.
column will be added by integer position.
Append a column of row sums to a DataFrame
Swap column contents – change column order df['Total'] = df.sum(axis=1)
df[['B', 'A']] = df[['A', 'B']] Note: also means, mins, maxs, etc.

Dropping (deleting) columns (mostly by label) Multiply every column in DataFrame by Series
df = df.drop('col1', axis=1) df = df.mul(s, axis=0) # on matched rows
df.drop('col1', axis=1, inplace=True) Note: also add, sub, div, etc.
df = df.drop(['col1','col2'], axis=1)
s = df.pop('col') # drops from frame Selecting columns with .loc, .iloc and .ix
del df['col'] # even classic python works df = df.loc[:, 'col1':'col2'] # inclusive
df.drop(df.columns[0], inplace=True) df = df.iloc[:, 0:2] # exclusive

Vectorised arithmetic on columns Get the integer position of a column index label
df['proportion']=df['count']/df['total'] j = df.columns.get_loc('col_name')
df['percent'] = df['proportion'] * 100.0
Test if column index values are unique/monotonic
Apply numpy mathematical functions to columns
if df.columns.is_unique: pass # ...
df['log_data'] = np.log(df['col1']) b = df.columns.is_monotonic_increasing
Note: Many more numpy mathematical functions. b = df.columns.is_monotonic_decreasing
Hint: Prefer pandas math over numpy where you can.

Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
3
 Select a slice of rows by label/index
Working with rows [inclusive-from : inclusive–to [ : step]]
df = df['a':'c'] # rows 'a' through 'c'
Get the row index and labels Trap: doesn't work on integer labelled rows
idx = df.index # get row index
label = df.index[0] # 1st row label Append a row of column totals to a DataFrame
lst = df.index.tolist() # get as a list # Option 1: use dictionary comprehension
sums = {col: df[col].sum() for col in df}
Change the (row) index sums_df = DataFrame(sums,index=['Total'])
df.index = idx # new ad hoc index df = df.append(sums_df)
df = df.set_index('A')# col A new index
df = df.set_index(['A', 'B'])# MultiIndex # Option 2: All done with pandas
df = df.reset_index() # replace old w new df = df.append(DataFrame(df.sum(),
# note: old index stored as a col in df columns=['Total']).T)
df.index = range(len(df)) # set with list
df = df.reindex(index=range(len(df))) Iterating over DataFrame rows
df = df.set_index(keys=['r1','r2','etc']) for (index, row) in df.iterrows(): # pass
df.rename(index={'old':'new'}, Trap: row data type may be coerced.
inplace=True)
Sorting DataFrame rows values
Adding rows df = df.sort(df.columns[0],
df = original_df.append(more_rows_in_df) ascending=False)
Hint: convert to a DataFrame and then append. Both df.sort(['col1', 'col2'], inplace=True)
DataFrames should have same column labels.
Sort DataFrame by its row index
Dropping rows (by name) df.sort_index(inplace=True) # sort by row
df = df.drop('row_label') df = df.sort_index(ascending=False)
df = df.drop(['row1','row2']) # multi-row
Random selection of rows
Boolean row selection by values in a column import random as r
df = df[df['col2'] >= 0.0] k = 20 # pick a number
df = df[(df['col3']>=1.0) | selection = r.sample(range(len(df)), k)
(df['col1']<0.0)] df_sample = df.iloc[selection, :]
df = df[df['col'].isin([1,2,5,7,11])] Note: this sample is not sorted
df = df[~df['col'].isin([1,2,5,7,11])]
df = df[df['col'].str.contains('hello')] Drop duplicates in the row index
Trap: bitwise "or", "and" “not; (ie. | & ~) co-opted to be df['index'] = df.index # 1 create new col
Boolean operators on a Series of Boolean df = df.drop_duplicates(cols='index',
Trap: need parentheses around comparisons. take_last=True)# 2 use new col
del df['index'] # 3 del the col
Selecting rows using isin over multiple columns df.sort_index(inplace=True)# 4 tidy up
# fake up some data
data = {1:[1,2,3], 2:[1,4,9], 3:[1,8,27]} Test if two DataFrames have same row index
df = DataFrame(data) len(a)==len(b) and all(a.index==b.index)

# multi-column isin Get the integer position of a row or col index label
lf = {1:[1, 3], 3:[8, 27]} # look for
i = df.index.get_loc('row_label')
f = df[df[list(lf)].isin(lf).all(axis=1)]
Trap: index.get_loc() returns an integer for a unique
match. If not a unique match, may return a slice or
Selecting rows using an index
mask.
idx = df[df['col'] >= 2].index
print(df.ix[idx])
Get integer position of rows that meet condition
a = np.where(df['col'] >= 2) #numpy array
Select a slice of rows by integer position
[inclusive-from : exclusive-to [: step]]
default start is 0; default end is len(df) Test if the row index values are unique/monotonic
df = df[:] # copy DataFrame if df.index.is_unique: pass # ...
df = df[0:2] # rows 0 and 1 b = df.index.is_monotonic_increasing
df = df[-1:] # the last row b = df.index.is_monotonic_decreasing
df = df[2:3] # row 2 (the third row)
df = df[:-1] # all but the last row
df = df[::2] # every 2nd row (0 2 ..)
Trap: a single integer without a colon is a column label
for integer numbered columns.

Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
4
 Working with cells Summary: selecting using the Index

Selecting a cell by row and column labels Using the DataFrame index to select columns
value = df.at['row', 'col'] s = df['col_label'] # returns Series
value = df.loc['row', 'col'] df = df[['col_label']]# return DataFrame
value = df['col'].at['row'] # tricky df = df[['L1', 'L2']] # select with list
Note: .at[] fastest label based scalar lookup df = df[index] # select with index
df = df[s] #select with Series
Setting a cell by row and column labels Note: the difference in return type with the first two
df.at['row', 'col'] = value examples above based on argument type (scalar vs list).
df.loc['row', 'col'] = value
df['col'].at['row'] = value # tricky Using the DataFrame index to select rows
df = df['from':'inc_to']# label slice
Selecting and slicing on labels df = df[3:7] # integer slice
df = df.loc['row1':'row3', 'col1':'col3'] df = df[df['col'] > 0.5]# Boolean Series
df = df.loc['label'] # single label
Note: the "to" on this slice is inclusive.
df = df.loc[container] # lab list/Series
df = df.loc['from':'to']# inclusive slice
Setting a cross-section by labels df = df.loc[bs] # Boolean Series
df.loc['A':'C', 'col1':'col3'] = np.nan df = df.iloc[0] # single integer
df.loc[1:2,'col1':'col2']=np.zeros((2,2)) df = df.iloc[container] # int list/Series
df.loc[1:2,'A':'C']=othr.loc[1:2,'A':'C'] df = df.iloc[0:5] # exclusive slice
Remember: inclusive "to" in the slice df = df.ix[x] # loc then iloc

Selecting a cell by integer position Using the DataFrame index to select a cross-section
value = df.iat[9, 3] # [row, col] # r and c can be scalar, list, slice
value = df.iloc[0, 0] # [row, col] df.loc[r, c] # label accessor (row, col)
value = df.iloc[len(df)-1, df.iloc[r, c]# integer accessor
len(df.columns)-1] df.ix[r, c] # label access int fallback
df[c].iloc[r]# chained – also for .loc
Selecting a range of cells by int position
df = df.iloc[2:4, 2:4] # subset of the df Using the DataFrame index to select a cell
df = df.iloc[:5, :5] # top left corner # r and c must be label or integer
s = df.iloc[5, :] # returns row as Series df.at[r, c] # fast scalar label accessor
df = df.iloc[5:6, :] # returns row as row df.iat[r, c] # fast scalar int accessor
Note: exclusive "to" – same as python list slicing. df[c].iat[r] # chained – also for .at

Setting cell by integer position DataFrame indexing methods

df.iloc[0, 0] = value # [row, col] v = df.get_value(r, c) # get by row, col
df.iat[7, 8] = value df = df.set_value(r,c,v)# set by row, col
df = df.xs(key, axis) # get cross-section
Setting cell range by integer position df = df.filter(items, like, regex, axis)
df.iloc[0:3, 0:5] = value df = df.select(crit, axis)
df.iloc[1:3, 1:4] = np.ones((2, 3))
df.iloc[1:3, 1:4] = np.zeros((2, 3)) Note: the indexing attributes (.loc, .iloc, .ix, .at .iat) can
df.iloc[1:3, 1:4] = np.array([[1, 1, 1], be used to get and set values in the DataFrame.
[2, 2, 2]]) Note: the .loc, iloc and .ix indexing attributes can accept
Remember: exclusive-to in the slice python slice objects. But .at and .iat do not.
Note: .loc can also accept Boolean Series arguments
.ix for mixed label and integer position indexing Avoid: chaining in the form df[col_indexer][row_indexer]
value = df.ix[5, 'col1'] Trap: label slices are inclusive, integer slices exclusive.
df = df.ix[1:5, 'col1':'col3']
Some index attributes and methods
Views and copies # --- some Index attributes
From the manual: Setting a copy can cause subtle b = idx.is_monotonic_decreasing
errors. The rules about when a view on the data is b = idx.is_monotonic_increasing
returned are dependent on NumPy. Whenever an array b = idx.has_duplicates
of labels or a Boolean vector are involved in the indexing i = idx.nlevels # num of index levels
operation, the result will be a copy. # --- some Index methods
idx = idx.astype(dtype)# change data type
b = idx.equals(o) # check for equality
idx = idx.union(o) # union of two indexes
i = idx.nunique() # number unique labels
label = idx.min() # minimum label
label = idx.max() # maximum label

Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
5
 Grouping
Joining/Combining DataFrames gb = df.groupby('cat') # by one columns
gb = df.groupby(['c1','c2']) # by 2 cols
Three ways to join two DataFrames: gb = df.groupby(level=0) # multi-index gb
gb = df.groupby(level=['a','b']) # mi gb
• merge (a database/SQL-like join operation)
print(gb.groups)
• concat (stack side by side or one on top of the other)
Note: groupby() returns a pandas groupby object
• combine_first (splice the two together, choosing
Note: the groupby object attribute .groups contains a
values from one over the other)
dictionary mapping of the groups.
Trap: NaN values in the group key are automatically
Merge on indexes
dropped – there will never be a NA group.
df_new = pd.merge(left=df1, right=df2,
how='outer', left_index=True,
Iterating groups – usually not needed
right_index=True)
for name, group in gb:
How: 'left', 'right', 'outer', 'inner'
print (name)
How: outer=union/all; inner=intersection print (group)
Merge on columns
Selecting a group
df_new = pd.merge(left=df1, right=df2,
dfa = df.groupby('cat').get_group('a')
how='left', left_on='col1',
dfb = df.groupby('cat').get_group('b')
right_on='col2')
Trap: When joining on columns, the indexes on the
passed DataFrames are ignored. Applying an aggregating function
Trap: many-to-many merges on a column can result in # apply to a column ...
an explosion of associated data. s = df.groupby('cat')['col1'].sum()
s = df.groupby('cat')['col1'].agg(np.sum)
# apply to the every column in DataFrame
Join on indexes (another way of merging)
s = df.groupby('cat').agg(np.sum)
df_new = df1.join(other=df2, on='col1', df_summary = df.groupby('cat').describe()
how='outer') df_row_1s = df.groupby('cat').head(1)
df_new = df1.join(other=df2,on=['a','b'],
how='outer') Note: aggregating functions reduce the dimension by
one – they include: mean, sum, size, count, std, var,
Note: DataFrame.join() joins on indexes by default.
sem, describe, first, last, min, max
DataFrame.merge() joins on common columns by
default.
Applying multiple aggregating functions
gb = df.groupby('cat')
Simple concatenation is often the best
df=pd.concat([df1,df2],axis=0)#top/bottom # apply multiple functions to one column
df = df1.append([df2, df3]) #top/bottom dfx = gb['col2'].agg([np.sum, np.mean])
df=pd.concat([df1,df2],axis=1)#left/right # apply to multiple fns to multiple cols
Trap: can end up with duplicate rows or cols dfy = gb.agg({
Note: concat has an ignore_index parameter 'cat': np.count_nonzero,
'col1': [np.sum, np.mean, np.std],
Combine_first 'col2': [np.min, np.max]
df = df1.combine_first(other=df2) })
Note: gb['col2'] above is shorthand for
# multi-combine with python reduce() df.groupby('cat')['col2'], without the need for regrouping.
df = reduce(lambda x, y:
x.combine_first(y), Transforming functions
[df1, df2, df3, df4, df5]) # transform to group z-scores, which have
Uses the non-null values from df1. The index of the # a group mean of 0, and a std dev of 1.
combined DataFrame will be the union of the indexes zscore = lambda x: (x-x.mean())/x.std()
from df1 and df2. dfz = df.groupby('cat').transform(zscore)

# replace missing data with group mean

mean_r = lambda x: x.fillna(x.mean())
Groupby: Split-Apply-Combine dfm = df.groupby('cat').transform(mean_r)
Note: can apply multiple transforming functions in a
The pandas "groupby" mechanism allows us to split the manner similar to multiple aggregating functions above,
data into groups, apply a function to each group
independently and then combine the results. Applying filtering functions
Filtering functions allow you to make selections based
on whether each group meets specified criteria
# select groups with more than 10 members
eleven = lambda x: (len(x['col1']) >= 11)
df11 = df.groupby('cat').filter(eleven)

Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
6
Group by a row index (non-hierarchical index)
df = df.set_index(keys='cat') Working with dates, times and their indexes
s = df.groupby(level=0)['col1'].sum()
dfg = df.groupby(level=0).sum() Dates and time – points and spans
With its focus on time-series data, pandas has a suite of
tools for managing dates and time: either as a point in
time (a Timestamp) or as a span of time (a Period).
Pivot Tables: working with long and wide data
t = pd.Timestamp('2013-01-01')
t = pd.Timestamp('2013-01-01 21:15:06')
These features work with and often create t = pd.Timestamp('2013-01-01 21:15:06.7')
hierarchical or multi-level Indexes; p = pd.Period('2013-01-01', freq='M')
(the pandas MultiIndex is powerful and complex). Note: Timestamps should be in range 1678 and 2261
years. (Check Timestamp.max and Timestamp.min).
Pivot, unstack, stack and melt
Pivot tables move from long format to wide format data A Series of Timestamps or Periods
# Let's start with data in long format ts = ['2015-04-01 13:17:27',
from StringIO import StringIO # python2.7 '2014-04-02 13:17:29']
#from io import StringIO # python 3
data = """Date,Pollster,State,Party,Est # Series of Timestamps (good)
13/03/2014, Newspoll, NSW, red, 25 s = pd.to_datetime(pd.Series(ts))
13/03/2014, Newspoll, NSW, blue, 28
13/03/2014, Newspoll, Vic, red, 24 # Series of Periods (often not so good)
13/03/2014, Newspoll, Vic, blue, 23 s = pd.Series( [pd.Period(x, freq='M')
13/03/2014, Galaxy, NSW, red, 23 for x in ts] )
13/03/2014, Galaxy, NSW, blue, 24 s = pd.Series(
13/03/2014, Galaxy, Vic, red, 26 pd.PeriodIndex(ts,freq='S'))
13/03/2014, Galaxy, Vic, blue, 25 Note: While Periods make a very useful index; they may
13/03/2014, Galaxy, Qld, red, 21 be less useful in a Series.
13/03/2014, Galaxy, Qld, blue, 27"""
df = pd.read_csv(StringIO(data), From non-standard strings to Timestamps
header=0, skipinitialspace=True) t = ['09:08:55.7654-JAN092002',
'15:42:02.6589-FEB082016']
# pivot to wide format on 'Party' column s = pd.Series(pd.to_datetime(t,
# 1st: set up a MultiIndex for other cols format="%H:%M:%S.%f-%b%d%Y"))
df1 = df.set_index(['Date', 'Pollster', Also: %B = full month name; %m = numeric month;
'State']) %y = year without century; and more …
# 2nd: do the pivot
wide1 = df1.pivot(columns='Party') Dates and time – stamps and spans as indexes
An index of Timestamps is a DatetimeIndex.
# unstack to wide format on State / Party
An index of Periods is a PeriodIndex.
# 1st: MultiIndex all but the Values col
df2 = df.set_index(['Date', 'Pollster', date_strs = ['2014-01-01', '2014-04-01',
'State', 'Party']) '2014-07-01', '2014-10-01']
# 2nd: unstack a column to go wide on it
wide2 = df2.unstack('State') dti = pd.DatetimeIndex(date_strs)
wide3 = df2.unstack() # pop last index
pid = pd.PeriodIndex(date_strs, freq='D')
# Use stack() to get back to long format pim = pd.PeriodIndex(date_strs, freq='M')
long1 = wide1.stack() piq = pd.PeriodIndex(date_strs, freq='Q')
# Then use reset_index() to remove the
# MultiIndex. print (pid[1] - pid[0]) # 90 days
long2 = long1.reset_index() print (pim[1] - pim[0]) # 3 months
print (piq[1] - piq[0]) # 1 quarter
# Or melt() back to long format
# 1st: flatten the column index time_strs = ['2015-01-01 02:10:40.12345',
wide1.columns = ['_'.join(col).strip() '2015-01-01 02:10:50.67890']
for col in wide1.columns.values] pis = pd.PeriodIndex(time_strs, freq='U')
# 2nd: remove the MultiIndex
wdf = wide1.reset_index() df.index = pd.period_range('2015-01',
# 3rd: melt away periods=len(df), freq='M')
long3 = pd.melt(wdf, value_vars=
['Est_blue', 'Est_red'], dti = pd.to_datetime(['04-01-2012'],
var_name='Party', id_vars=['Date', dayfirst=True) # Australian date format
'Pollster', 'State']) pi = pd.period_range('1960-01-01',
'2015-12-31', freq='M')
Note: See documentation, there are many arguments to
these methods. Hint: unless you are working in less than seconds,
prefer PeriodIndex over DateTimeImdex.

Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
7
Period frequency constants (not a complete list) Upsampling and downsampling
Name Description # upsample from quarterly to monthly
U Microsecond pi = pd.period_range('1960Q1',
L Millisecond periods=220, freq='Q')
S Second df = DataFrame(np.random.rand(len(pi),5),
index=pi)
T Minute
dfm = df.resample('M', convention='end')
H Hour # use ffill or bfill to fill with values
D Calendar day
B Business day # downsample from monthly to quarterly
W-{MON, TUE, …} Week ending on … dfq = dfm.resample('Q', how='sum')
MS Calendar start of month
M Calendar end of month Time zones
QS-{JAN, FEB, …} Quarter start with year starting t = ['2015-06-30 00:00:00',
(QS – December) '2015-12-31 00:00:00']
Q-{JAN, FEB, …} Quarter end with year ending (Q dti = pd.to_datetime(t
– December) ).tz_localize('Australia/Canberra')
AS-{JAN, FEB, …} Year start (AS - December) dti = dti.tz_convert('UTC')
ts = pd.Timestamp('now',
A-{JAN, FEB, …} Year end (A - December) tz='Europe/London')
From DatetimeIndex to Python datetime objects # get a list of all time zones
dti = pd.DatetimeIndex(pd.date_range( import pyzt
start='1/1/2011', periods=4, freq='M')) for tz in pytz.all_timezones:
s = Series([1,2,3,4], index=dti) print tz
na = dti.to_pydatetime() #numpy array Note: by default, Timestamps are created without time
na = s.index.to_pydatetime() #numpy array zone information.

Frome Timestamps to Python dates or times Row selection with a time-series index
df['date'] = [x.date() for x in df['TS']] # start with the play data above
df['time'] = [x.time() for x in df['TS']] idx = pd.period_range('2015-01',
Note: converts to datatime.date or datetime.time. But periods=len(df), freq='M')
does not convert to datetime.datetime. df.index = idx

From DatetimeIndex to PeriodIndex and back february_selector = (df.index.month == 2)

df = DataFrame(np.random.randn(20,3)) february_data = df[february_selector]
df.index = pd.date_range('2015-01-01',
periods=len(df), freq='M') q1_data = df[(df.index.month >= 1) &
dfp = df.to_period(freq='M') (df.index.month <= 3)]
dft = dfp.to_timestamp()
Note: from period to timestamp defaults to the point in mayornov_data = df[(df.index.month == 5)
| (df.index.month == 11)]
time at the start of the period.
totals = df.groupby(df.index.year).sum()
Working with a PeriodIndex
Also: year, month, day [of month], hour, minute, second,
pi = pd.period_range('1960-01','2015-12',
dayofweek [Mon=0 .. Sun=6], weekofmonth, weekofyear
freq='M')
na = pi.values # numpy array of integers [numbered from 1], week starts on Monday], dayofyear
lp = pi.tolist() # python list of Periods [from 1], …
sp = Series(pi)# pandas Series of Periods
ss = Series(pi).astype(str) # S of strs The Series.dt accessor attribute
ls = Series(pi).astype(str).tolist() DataFrame columns that contain datetime-like objects
can be manipulated with the .dt accessor attribute
Get a range of Timestamps t = ['2012-04-14 04:06:56.307000',
dr = pd.date_range('2013-01-01', '2011-05-14 06:14:24.457000',
'2013-12-31', freq='D') '2010-06-14 08:23:07.520000']

# a Series of time stamps

Error handling with dates s = pd.Series(pd.to_datetime(t))
# 1st example returns string not Timestamp print(s.dtype) # datetime64[ns]
t = pd.to_datetime('2014-02-30') print(s.dt.second) # 56, 24, 7
# 2nd example returns NaT (not a time) print(s.dt.month) # 4, 5, 6
t = pd.to_datetime('2014-02-30',
coerce=True) # a Series of time periods
# NaT like NaN tests True for isnull() s = pd.Series(pd.PeriodIndex(t,freq='Q'))
b = pd.isnull(t) # --> True print(s.dtype) # datetime64[ns]
print(s.dt.quarter) # 2, 2, 2
The tail of a time-series DataFrame print(s.dt.year) # 2012, 2011, 2010
df = df.last("5M") # the last five months
Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
8
 Working with missing and non-finite data Working with Categorical Data

Working with missing data Categorical data

Pandas uses the not-a-number construct (np.nan and The pandas Series has an R factors-like data type for
float('nan')) to indicate missing data. The Python None encoding categorical data.
can arise in data as well. It is also treated as missing s = Series(['a','b','a','c','b','d','a'],
data; as is the pandas not-a-time construct dtype='category')
(pandas.NaT). df['B'] = df['A'].astype('category')
Note: the key here is to specify the "category" data type.
Missing data in a Series Note: categories will be ordered on creation if they are
s = Series( [8,None,float('nan'),np.nan]) sortable. This can be turned off. See ordering below.
#[8, NaN, NaN, NaN]
s.isnull() #[False, True, True, True] Convert back to the original data type
s.notnull()#[True, False, False, False] s = Series(['a','b','a','c','b','d','a'],
s.fillna(0)#[8, 0, 0, 0] dtype='category')
s = s.astype('string')
Missing data in a DataFrame
df = df.dropna() # drop all rows with NaN Ordering, reordering and sorting
df = df.dropna(axis=1) # same for cols s = Series(list('abc'), dtype='category')
df=df.dropna(how='all') #drop all NaN row print (s.cat.ordered)
df=df.dropna(thresh=2) # drop 2+ NaN in r s=s.cat.reorder_categories(['b','c','a'])
# only drop row if NaN in a specified col s = s.sort()
df = df.dropna(df['col'].notnull()) s.cat.ordered = False
Trap: category must be ordered for it to be sorted
Recoding missing data
df.fillna(0, inplace=True) # np.nan à 0 Renaming categories
s = df['col'].fillna(0) # np.nan à 0 s = Series(list('abc'), dtype='category')
df = df.replace(r'\s+', np.nan, s.cat.categories = [1, 2, 3] # in place
regex=True) # white space à np.nan s = s.cat.rename_categories([4,5,6])
# using a comprehension ...
Non-finite numbers s.cat.categories = ['Group ' + str(i)
With floating point numbers, pandas provides for for i in s.cat.categories]
positive and negative infinity. Trap: categories must be uniquely named
s = Series([float('inf'), float('-inf'),
np.inf, -np.inf]) Adding new categories
Pandas treats integer comparisons with plus or minus s = s.cat.add_categories([4])
infinity as expected.
Removing categories
Testing for finite numbers s = s.cat.remove_categories([4])
(using the data from the previous example) s.cat.remove_unused_categories() #inplace
b = np.isfinite(s)

Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
9
 Working with strings Basic Statistics

Working with strings Summary statistics

# assume that df['col'] is series of s = df['col1'].describe()
strings df1 = df.describe()
s = df['col'].str.lower()
s = df['col'].str.upper() DataFrame – key stats methods
s = df['col'].str.len() df.corr() # pairwise correlation cols
df.cov() # pairwise covariance cols
# the next set work like Python df.kurt() # kurtosis over cols (def)
df['col'] += 'suffix' # append df.mad() # mean absolute deviation
df['col'] *= 2 # duplicate df.sem() # standard error of mean
s = df['col1'] + df['col2'] # concatenate df.var() # variance over cols (def)
Most python string functions are replicated in the pandas
DataFrame and Series objects. Value counts
s = df['col1'].value_counts()
Regular expressions
s = df['col'].str.contains('regex')
Cross-tabulation (frequency count)
s = df['col'].str.startswith('regex')
s = df['col'].str.endswith('regex') ct = pd.crosstab(index=df['a'],
s = df['col'].str.replace('old', 'new') cols=df['b'])
df['b'] = df.a.str.extract('(pattern)')
Note: pandas has many more regex methods. Quantiles and ranking
quants = [0.05, 0.25, 0.5, 0.75, 0.95]
q = df.quantile(quants)
r = df.rank()

Histogram binning
count, bins = np.histogram(df['col1'])
count, bins = np.histogram(df['col'],
bins=5)
count, bins = np.histogram(df['col1'],
bins=[-3,-2,-1,0,1,2,3,4])

Regression
import statsmodels.formula.api as sm
result = sm.ols(formula="col1 ~ col2 +
col3", data=df).fit()
print (result.params)
print (result.summary())

Smoothing example using rolling_apply

k3x5 = np.array([1,2,3,3,3,2,1]) / 15.0
s = pd.rolling_apply(df['col1'],
window=7,
func=lambda x: (x * k3x5).sum(),
min_periods=7, center=True)

Cautionary note

This cheat sheet was cobbled together by bots roaming

the dark recesses of the Internet seeking ursine and
pythonic myths. There is no guarantee the narratives
were captured and transcribed accurately. You use
these notes at your own risk. You have been warned.

Version: This cheat sheet was last updated with

Python 3.5 and pandas 0.18.0 in mind.

Errors: If you find any errors, please email me at

[email protected]; (but please do not correct
my use of Australian-English spelling conventions).

Version 14 May 2016 - [Draft – Mark Graph – mark dot the dot graph at gmail dot com – @Mark_Graph on twitter]
10