LoC Data Package Tutorial: Stereograph Card Collection

This notebook will demonstrate basic usage of using Python for interacting with data packages from the Library of Congress via the Stereograph Card Images Data Package which is derived from the Library’s Stereograph Cards collection. We will:

1. Output a summary of the contents of this data package

2. Read and query metadata from a data package

3. Download and display images from a data package

Prerequisites

In order to run this notebook, please follow the instructions listed in this directory’s README.

Output data package summary

First, we will select Stereograph Card Images Data Package and output a summary of it’s contents

import io

import pandas as pd                     # for reading, manipulating, and displaying data
import requests

from helpers import get_file_stats

DATA_URL = 'https://data.labs.loc.gov/stereographs/' # Base URL of this data package

# Download the file manifest
file_manifest_url = f'{DATA_URL}manifest.json'
response = requests.get(file_manifest_url, timeout=120)
response_json = response.json()
files = [dict(zip(response_json["cols"], row)) for row in response_json["rows"]] # zip columns and rows

# Convert to Pandas DataFrame and show stats table
stats = get_file_stats(files)
pd.DataFrame(stats)

	FileType	Count	Size
0	.jpg	39,597	4.03GB

Query the metadata in a data package

Next we will download a data package’s metadata, print a summary of the items’ subject values, then filter by a particular subject.

All data packages have a metadata file in .json and .csv formats. Let’s load the data package’s metadata.json file:

metadata_url = f'{DATA_URL}metadata.json'
response = requests.get(metadata_url, timeout=300)
data = response.json()
print(f'Loaded metadata file with {len(data):,} entries.')

Loaded metadata file with 39,532 entries.

Next let’s convert to pandas DataFrame and print the available properties

df = pd.DataFrame(data)
print(', '.join(df.columns.to_list()))

access_restricted, aka, campaigns, contributor, coordinates, date, description, digitized, extract_timestamp, group, hassegments, id, image_url, index, language, latlong, location, location_str, locations, lonlat, mime_type, online_format, original_format, other_title, partof, reproductions, resources, shelf_id, site, subject, timestamp, title, unrestricted, url, item, related, dates, number, number_former_id, number_lccn, number_oclc, number_carrier_type, number_source_modified, type, location_city, location_country, location_state, location_county, manifest_id

Next print the top 20 most frequent Subjects in this dataset

# Since "subject" are a list, we must "explode" it so there's just one subject per row
# We convert to DataFrame so it displays as a table
df['subject'].explode().value_counts().iloc[:20].to_frame()

	subject
stereographs	37427
photographic prints	33217
albumen prints	3596
new york (state)	2367
	2030
new york	1906
united states	1618
history	1404
italy	1381
civil war	1281
japan	1216
washington (d.c.)	1077
(	1040
norway	943
louisiana purchase exposition	856
missouri	849
saint louis	844
saint louis, mo.)	841
switzerland	819
ireland	722

Now we filter the results to only those items with subject “washington (d.c.)”

df_by_subject = df.explode('subject')
dc_set = df_by_subject[df_by_subject.subject == 'washington (d.c.)']
print(f'Found {dc_set.shape[0]:,} items with subject "washington (d.c.)"')

Found 1,077 items with subject "washington (d.c.)"

Download and display images

First we will merge the metadata with the file manifest to link the file URL to the respective item.

df_files = pd.DataFrame(files)
dc_set_with_images = pd.merge(dc_set, df_files, left_on='id', right_on='item_id', how='inner')
print(f'Found {dc_set_with_images.shape[0]:,} dc items with image files')

Found 1,071 dc items with image files

Finally we will download and display the first 4 images that have subject “washington (d.c.)”

import math
from IPython.display import display     # for displaying images
from PIL import Image                   # for creating, reading, and manipulating images

count = 4
dc_set_with_images = dc_set_with_images.head(count).reset_index()

# Define image dimensions
image_w = 600
image_h = 600
cols = math.ceil(count / 2.0)
rows = math.ceil(count / 2.0)
cell_w = image_w / cols
cell_h = image_h / rows

# Create base image
base_image = Image.new("RGB", (image_w, image_h))

# Loop through image URLs
i = 0
for i, row in dc_set_with_images.iterrows():
    file_url = f'https://{row["object_key"]}'

    # Downoad the image to memory
    response = requests.get(file_url, timeout=60)
    image_filestream = io.BytesIO(response.content)

    # And read the image data
    im = Image.open(image_filestream)

    # Resize it as a thumbnail
    im.thumbnail((cell_w, cell_h))
    tw, th = im.size

    # Position it
    col = i % cols
    row = int(i / cols)
    offset_x = int((cell_w - tw) * 0.5) if tw < cell_w else 0
    offset_y = int((cell_h - th) * 0.5) if th < cell_h else 0
    x = int(col * cell_w + offset_x)
    y = int(row * cell_h + offset_y)

    # Paste it
    base_image.paste(im, (x, y))
    i += 1

# Display the result
display(base_image)