Skip to content

Instantly share code, notes, and snippets.

@zufanka

zufanka/The_turbo_data_analysis_cheatsheet.md

Last active April 8, 2024 12:33
Show Gist options
  • Star 1 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save zufanka/39b8a55d707b3b4a2a4d369694739561 to your computer and use it in GitHub Desktop.
Save zufanka/39b8a55d707b3b4a2a4d369694739561 to your computer and use it in GitHub Desktop.
Download ZIP
The turbo data analysis cheatsheet
Raw
The_turbo_data_analysis_cheatsheet.md

Contents

SQL

Update

Update a column

postgres: update a column based on another column

UPDATE income
SET currency = 'HRK'
WHERE income.date < '2022-01-01';

Update with a join

UPDATE income
SET currency = 'HRK'
FROM parties
WHERE income.party_id = parties.id
AND parties.country = 'HR'
AND income.date < '2022-01-01';

Python

Lists

Cycling through a list

To cycle through the list, you can use the modulo operator: L[i % n]

computer_rooms = ["2.08","2.09","2.1"]
[computer_rooms[i % len(computer_rooms)] for i in range(10)]
# ['2.08', '2.09', '2.1', '2.08', '2.09', '2.1', '2.08', '2.09', '2.1', '2.08']

Split on a pattern

The re.findall(pattern, string) method scans string from left to right, searching for all non-overlapping matches of the pattern. It returns a list of strings in the matching order when scanning the string from left to right.

import re
s = '111A222B333C'
res = re.findall('(\d+|[A-Za-z]+)', s)
print(res)
# ['111', 'A', '222', 'B', '333', 'C']

Flatten list of lists

[item for sublist in listoflists for item in sublist]

Find intersection of two lists

has_tail = ["cat", "dog", "chameleon"]
has_earlobes = ["human", "cat", "dog"]

list(filter(set(has_tail).__contains__, has_earlobes))

>> ['cat', 'dog']

Dictionaries

Find parameters of an object

object.as_dict()

Swap keys and values of a dictionary

{y:x for x,y in stations.items()}

Using dictionaries instead of if-elif-else

# unPythonic
if season == 'Winter':
    holiday = 'Christmas'
elif season == 'Spring':
    holiday = 'Easter'
elif season == 'Summer':
    holiday = 'American Independence Day'
elif season == 'Fall':
    holiday = 'Halloween'
else:
    holiday = 'Personal day off'

# Pythonic
season = "Winter"
holiday = {'Winter': 'Christmas',
           'Spring': 'Easter',
           'Summer': 'American Independence Day',
           'Fall':   'Halloween'}.get(season, 'Personal day off')
season
"Christmas"

Using dictionaries to prevent a key error

workDetails = {}

# unPythonic
if 'hours' in spam:
    hoursWorked = workDetails['hours']
else:
    hoursWorked = 0 # Default to 0 if the 'hours' key doesn't exist.
    
# Pythonic
hoursWorked = spam.get('hours', 0)

# Pythonic 2
workDetails.setdefault('hours', 0)  # Does nothing if 'hours' exists.

two lists into a dictionary

keys = ['a', 'b', 'c']
values = [1, 2, 3]
dictionary = dict(zip(keys, values))

Dates

Make datetime object from isoformat

datetime.date.fromisoformat('string')

Put a datetime object to isoformat

datetimeObject.isoformat()

BeautifulSoup

replace br with a different character

for br in soup.find_all('br'):
   br.replace_with("|")

Pandas

Jupyter notebook pandas settings with gspread

import pandas as pd

from IPython.display import Markdown, display

def printmd(string):
    display(Markdown(string))
    
import gspread
import gspread_dataframe as gd
from oauth2client.service_account import ServiceAccountCredentials
    
pd.set_option("max_colwidth", 500)
pd.options.display.max_rows = 100
pd.options.display.max_columns = 100
pd.options.display.float_format = '{:,.2f}'.format

Check if a cell is NaN

Not a column, but just one value

pd.isna(df["colname"][0])

Hence filling in a column only where there is NaN values based on a dictionary

dict = {"key" : "value"}
x["different_col"] * dict[x["somecolthat_holds_the_keys"]] if pd.isna(x["colname"]) else x["colname"]

Change type of Series

Series.astype(int)

Check what percentage of a df is (not) empty

returns a Series source

df.count() / len(df)
df.notnull().mean().round(4) * 100

Filter out columns that are N% empty

returns a list of columns

cutoff = 50 # percent of not null
filter_columns = df.notnull().mean().round(4) * 100 
filter_columns.where(filter_columns > cutoff).dropna().index

Set value based on a value of another column (based on a list)

If an item from somecol is in the list, assign value 1 to a new column in the DataFrame, if not assign a 0

list = [items, items, items]
df["newcol"] = [1 if y in list else 0 for y in df["somecol"]]

Count frequency: how many time the same row appears in the dataframe

result is a matrix

Group           Size

Short          Small
Short          Small
Moderate       Medium
Moderate       Small
Tall           Large

pd.crosstab(df.Group,df.Size)

sum strings across rows

df.groupby("col")["jointhis"].agg(lambda col: ', '.join(col))
df.groupby("col")["jointhis"].apply(list)

Cummulative count = making ID's, counting occurences of a column value

df.groupby("colname").cumcount() + 1 # +1 because otherwise counting starts at 0

colname1 | cumcount
a | 1
a | 2
b | 1
a | 3
c | 1

Make percentages from absolute values in a DataFrame

perc = pd.DataFrame(df.apply(lambda x: 100 * (x / float(x.sum())))).unstack()
# perc.columns = perc.columns.droplevel()

Handling jsons

more

import json
import requests

r = requests.get("https://api.tenders.exposed/networks/58d77f85-bbc6-447d-a292-c3f17b7936b0/").text
data = json.loads(r)

# go down the tree as far as you need
data["network"]["nodes"]

# normalize that baby
pd.json_normalize(data["network"]["nodes"])

Replace strings (from dict)

map_min = {"Ministerie van Economische Zaken, Landbouw en Innovatie" : "Ministerie van Economische Zaken",
           "Provincie Overrijssel" : "Provincie Overijssel"}
nl = nl.replace({'Overheid': map_min}) # "Overheid" is the column name

Fuzzy merging

from fuzzymatcher import link_table, fuzzy_left_join

fuzzy_left_join(df1, df2, "df1-matchingcol", "df2-matchingcol")

Split cell and explode into multiple rows

exploded = df[["id","colname"]].set_index("id")["colname"].str.split(";").explode().reset_index().drop_duplicates()
pd.merge(left=df, right=exploded, on="id")

Stack and unstack

img src

Styles

def highlight_max(s):
    '''
    highlight the maximum in a Series yellow.
    '''
    is_max = s == s.max()
    return ['background-color: yellow' if v else '' for v in is_max]

df.style.apply(highlight_max)

Styles

df.style.bar()
df.style.background_gradient(cmap='Blues', axis=None)

Sheets styling

https://github.com/robin900/gspread-formatting

Google Sheets >> DataFrame

import pandas as pd
import gspread
from oauth2client.service_account import ServiceAccountCredentials

scope = ['https://spreadsheets.google.com/feeds', 'https://www.googleapis.com/auth/drive']
creds = ServiceAccountCredentials.from_json_keyfile_name('the_credentials.json', scope)
client = gspread.authorize(creds)

url = "sheeturl"

sheet = client.open_by_url(sheet_url)

# entities = gd.get_as_dataframe(sheet.worksheet("sheetnamen"),  skiprows=1) # get a sheet
# entities = entities.dropna(axis=0, how="all").dropna(axis=1, how="all") # remove empty rows and columns
 #gd.set_with_dataframe(sheet.worksheet("sheetnamen"), entities, row=2) # push to a sheet

multiple csv >> DataFrame & filter

all_csvs = glob.glob("/data/*.csv")

cols = ["col1", "col2"]

df = pd.concat((pd.read_csv(table,
                 usecols=cols,
                 delimiter=';', low_memory=False) for table in all_csvs), ignore_index=True)

csvs in zip >> DataFrame

zf = zipfile.ZipFile("zippedcsvs.zip")
#zf.infolist()
df = pd.concat((pd.read_csv(zf.open(text_file.filename), encoding = "ISO-8859-1") for text_file in zf.infolist() if text_file.filename.endswith('.csv')))

ODS >> DataFrame

from pyexcel_ods import get_data

sheet = get_data("data.ods")["sheetname"]
df = pd.DataFrame(sheet)
df.columns = df.iloc[0]
df = df[1:]

DataFrame >> list of lists

lol = df.values.tolist()

dataframe >> dict

use one column as key and another as value

df.set_index('colname').T.to_dict('records')

Transpose

df.T

Selenium

get screenshot of the whole page

url = "https://www.web.com/"
path = "naeme.png"
    
driver = webdriver.Firefox()
driver.get(url)

time.sleep(5)

el = driver.find_element_by_tag_name('body')
el.screenshot(path)
driver.quit()

Numpy

Set value based on a value of multiple columns

np.where(((df["col1"] == "pink") &
          (data["col2"] == "pink")) |
         ((data["col1"] == "blue") &
          (data["col2"] == "blue")), "same", "not")

Split the dataframe into multiple equal parts

Use when data is next to each other rather than below

dfs = np.split(df, 12, axis=1)

Altair

Disable max rows

alt.data_transformers.enable('default', max_rows=None)

sorted barchart

column = is the nominal column we count on.
if sum, ...

alt.Chart(df).mark_bar().encode(
    x = "count(column):Q",
    y = alt.Y(
        'column:N',
        sort=alt.EncodingSortField(
            field="column",  # The field to use for the sort
            op="count",  # The operation to run on the field prior to sorting
            order="descending"  # The order to sort in
        ))
)

Descriptions of a grouped barchart

row = alt.Row("parameter", header=alt.Header(labelAngle = 0.1))

Custom colors

color = alt.Color('variable',
                   scale=alt.Scale(
                   domain=['good', 'medium', 'bad'],
                   range=['blue', 'yellow' ,'red']))

link from chart

alt.Chart(df).transform_calculate(
    url=alt.datum.colname
).mark_circle().encode(
    x = "col1",
    y = "col2",
    href='url:N',

Neo4j

Load network from a csv file

LOAD CSV WITH HEADERS FROM "file:///data_clean.csv" AS row
MERGE (b:buyer {buyer: coalesce(row.Buyer, "Unknown")})
MERGE (v:vendor {vendor: row.Vendor})
MERGE (p:property {property: coalesce(row.Asset, "Unknown")})
MERGE (v)-[r:SOLD]->(p)
MERGE (b)-[q:BOUGHT]->(p)
RETURN *;

Command line

Animated gifs

ffmpeg -i video.avi -t 10 out%02d.gif
gifsicle --delay=10 --loop *.gif > anim.gif

Maps

mapshaper

make a chloroplet

git

remove the last commit not pushed from the history while keeping local files intact

git reset --soft HEAD~1

Inspired by the cheatsheet of my dear colleague Hay Kranen

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment