Example Transforms

This documentation is for qri CLI (command line) only. Composing Transform scripts on Desktop is a work in progress.

Transforms allow you to bind data with code in Qri, providing a quick means to automate the creation of new dataset versions from websites, APIs, and other qri datasets.

To get started with transforms, check out Transforms Qruickstart.

Below are several examples of transform scripts that demonstrate various approaches to automating qri datasets. If you have a transform script that you think might help others, consider opening a pull request to add it here.

NYC Open Data (Simple CSV Download)

In this common transform, we download a CSV from the internet and set it to the qri dataset's body. The script does not parse the CSV into memory, it simply takes the downloaded file and writes it directly to the new dataset version. This is useful for versioning and/or archiving a published CSV that changes regularly.

# load dependencies
load("http.star", "http")
load("encoding/csv.star", "csv")
# get the popular baby names dataset as a csv
def download(ctx):
csvDownloadUrl = "https://data.cityofnewyork.us/api/views/25th-nujf/rows.csv?accessType=DOWNLOAD"
return http.get(csvDownloadUrl).body()
# set the body
def transform(ds, ctx):
# ctx.download is whatever download() returned
csv = ctx.download
# set the dataset body
ds.set_body(csv, parse_as='csv')

MTA Elevator Outages (Web Scraping)

The New York MTA publishes current subway station elevator outages on the web. This transform scrapes the website response using the Starlark bsoup package, parsing values from an html table into a CSV.

Once a clean dataset version of the elevator statuses is in hand, the script appends the rows to the previous body with a timestamp for when the page was scraped. (each version of the dataset includes a full list of all observations since the dataset was created)

This transform is mean to be run on a schedule, perhaps once a day, for the purpose of providing a system-wide view of elevator outages over time.

# load dependencies
load("encoding/csv.star", "csv")
load("http.star", "http")
load("time.star", "time")
load("bsoup.star", "bsoup")
load("re.star", "re")
# helper function for parsing text from bsoup nodes
# for this example, each td in the site we are scraping contains an h4
# this gets the text in the h4 and cleans white space with strip()
def parseColumn(columns, index):
return columns[index].find('h4').get_text().strip()
# leftPad() pads input with zeroes, takes int or string as input
def leftPad(input, length):
# convert int to string
if type(input) == 'int':
input = str(input)
padded = input
for i in range(len(input), length):
padded = '0' + padded
return padded
# given a time struct, returns an ISO8601 date+time string
def getISOTimestamp(t):
year = leftPad(t.year(), 4)
month = leftPad(t.month(), 2)
day = leftPad(t.day(), 2)
hour = leftPad(t.hour(), 2)
minute = leftPad(t.minute(), 2)
second = leftPad(t.second(), 2)
offset = re.findall('\s([+-]\d{4})\s', str(t), flags=0)[0].strip()
return '{}-{}-{}T{}:{}:{}{}'.format(year, month, day, hour, minute, second, offset)
# download() is executed automatically by qri and stores its return value for use in transform()
def download(ctx):
# send an http get request to the elevator outages page
res = http.get('http://advisory.mtanyct.info/EEoutage/EEOutageReport.aspx?StationID=All')
# get the response body
body = res.body()
# parse the html with bsoup
soup = bsoup.parseHtml(body)
# use bsoup to navigate through the dom and find the table that has the outages
table = soup.find('table').find('table').find('table').find('tbody').find_all('tr')[20].find('tbody')
# rows is an array of bsoup nodes, each will become a row in our dataset
rows = table.find_all('tr')
# create an array of arrays to hold our CSV data
csvData = []
# get a timestamp for now to add to each row
timestamp = getISOTimestamp(time.now())
# loop over table rows (start at 1 to skip the first row), parse values into an array of strings
for i in range(1, len(rows)):
# use bsoup to select all the tds
columns = rows[i].find_all('td')
# parse column text using a custom parseColumn() function
station_name = parseColumn(columns, 0)
elevator_number = parseColumn(columns, 2)
location = parseColumn(columns, 3)
out_of_service = parseColumn(columns, 4)
reason = parseColumn(columns, 5)
estimated_return = parseColumn(columns, 6)
# append an array of our row values to csvData
csvData.append([ timestamp, station_name, elevator_number, location, out_of_service, reason, estimated_return])
return csvData
def transform(ds, ctx):
# the download() function was run by qri and stored its return value in ctx.download
# use the csv package to convert the array of arrays into a CSV string
# we want to append these rows to the dataset body
# if there is already data in the body, do not append the header row again
previousBody = ds.get_body()
newBody = ctx.download
# bodyArray will hold the new body to be written by the transform
# the first row is the header row
headerRow = ['timestamp', 'station_name', 'elevator_number', 'location', 'out_of_service', 'reason', 'estimated_return']
bodyArray = [headerRow]
# if there is a previous body, append its contents into bodyArray
if previousBody != None:
for row in previousBody:
# now append the new rows
for row in newBody:
# use ds.set_body to commit the new body to our qri dataset
body = csv.write_all(bodyArray)
ds.set_body(body, parse_as='csv')

World Bank Population (API Response + Existing Dataset)

This transform is an example of combining a CSV from the web with lookup values in an existing qri dataset. The World Bank API provides a CSV of total population, but it includes rows countries and aggregated entities combined. This transform uses an existing qri dataset of ISO 3166 alpha3 country codes to filter the World Bank API response, resulting in a dataset of total population for _countries.

You'll need the country codes dataset to run this transform. Run qri add b5/country_codes before saving.

load("http.star", "http")
load("encoding/csv.star", "csv")
load("zipfile.star", "ZipFile")
# world bank api only gives us access to country codes mixed with other (often useful!) country-like
# entities like "upper middle class" and "Sub-Saharan Africa (excluding high income)"
# we need a list of actual ISO 3166 alpha_3 country codes, which this dataset provides on column index 7
country_codes = load_dataset("b5/country_codes")
# download is a special function called automatically by Qri if defined
def download(ctx):
# perform a HTTP GET request to the world bank API
res = http.get("http://api.worldbank.org/v2/en/indicator/SP.POP.TOTL?downloadformat=csv")
# response is a zip file with names that change on each download. first, open the zip archive:
zf = ZipFile(res.body())
# grab the 2nd file of three files, which contains the data we're after
nl = zf.namelist()
if len(nl) != 3:
error("expected list of files to equal 3")
# read raw data into a string
rawCsvData = zf.open(nl[1]).read()
# pass raw CSV data to the transform step
return rawCsvData
# transform is a special function called automatically by Qri if defined
def transform(ds, ctx):
countryCodes = [cc[7] for cc in country_codes.get_body()]
# assign csv data from download to a variable
rawCsvData = ctx.download
# data comes with two citation-oriented header rows, let's lop them off by reading csv data
parsedCsv = csv.read_all(rawCsvData, lazy_quotes=True, fields_per_record=-1, skip=2)
# filter out "countries" that aren't valid ISO 3116 alpha 3 country codes
onlyContries = [x for x in parsedCsv if x[1] in countryCodes]
# construct dataset structure
st = structure(parsedCsv[0])
# convert back to csv data without header row
csvString = csv.write_all(onlyContries)
ds.set_body(csvString, parse_as='csv')
# structure is a custom function for extracting a dataset structure.
# we need this this because Qri doesn't guess the schema correctly for us
# so we build one by hand
def structure(header_row):
items = [{ 'title': title, 'type': 'integer' } for title in header_row]
for i in range(0,4):
items[i]['type'] = 'string'
return {
'format': 'csv',
'formatConfig': {
'lazyQuotes' : True,
'headerRow' : True,
'schema' : {
'type' : 'array',
'items' : {
'type' : 'array',
'items' : items

Count NYC 311 Complaints By Boro (Spatial Join and Aggregation)

This transform pulls two datasets from the NYC Open Data Portal. The 311 data are retrieved as json from the Open Data Portal's API, and a borough boundaries spatial dataset is retreived as geojson.

The transform function conducts a spatial join using the starlark geo package, and keeps a count of the complaints that fall within each of NYC's five boroughs.

load("http.star", "http")
load("geo.star", "geo")
def download(ctx):
# Download list of 311 complaints, currently capped to 1000 responses for testing purposes
complaints = http.get("https://data.cityofnewyork.us/resource/fhrw-4uyv.json?$offset=0&$limit=10000")
# Download the New York Borough Boundaries
boros = http.get("http://data.beta.nyc//dataset/68c0332f-c3bb-4a78-a0c1-32af515892d6/resource/7c164faa-4458-4ff2-9ef0-09db00b509ef/download/42c737fd496f4d6683bba25fb0e86e1dnycboroughboundaries.geojson")
return {
"complaints" : complaints.json(),
# send boros over as a JSON string, parseGeoJSON will do the decoding
"boros": boros.text(),
def transform(ds, ctx):
complaints = ctx.download["complaints"]
boundaries, properties = geo.parseGeoJSON(ctx.download["boros"])
# bouroughs data specifies a number of polygons for each borough,
# combine them all into MultiPolygons, one for each borough
boro_names = [ boro['borough'] for boro in properties]
geoms = reduce(append_polygon, zip(boro_names, boundaries), {})
geoms = [geo.MultiPolygon(geoms[x]) for x in geoms]
boro_names = list(set(boro_names))
# dict of complaint-counts, keyed by borough name
boro_counts = dict(zip(boro_names, [0]*len(boro_names)))
for complaint in complaints:
# only use complaints that have lat & lng values
if 'latitude' in complaint and 'longitude' in complaint:
point = geo.Point(float(complaint['longitude']), float(complaint['latitude']))
for boro_name, geom in zip(boro_names, geoms):
if geo.within(point, geom):
boro_counts[boro_name] += 1
def append_polygon(acc, prop):
polys = acc.get(prop[0], [])
acc[prop[0]] = polys
return acc
def reduce(fn, l, v):
for _, el in enumerate(l):
v = fn(v, el)
return v