Skip to main content

A library for recording and reading data in Jupyter and nteract Notebooks

Project description

scrapbook logo


Travis Build Status Azure Build Status image Documentation Status badge badge Code style: black

THE scrapbook library records a notebook’s data values and generated visual content as "scraps". Recorded scraps can be read at a future time.

See the scrapbook documentation for more information on how to use scrapbook.

Use Cases

Notebook users may wish to record data produced during a notebook's execution. This recorded data, scraps, can be used at a later time or passed in a workflow to another notebook as input.

Namely, scrapbook lets you:

  • persist data and visual content displays in a notebook as scraps
  • recall any persisted scrap of data
  • summarize collections of notebooks

Python Version Support

This library's long term support target is Python 3.5+. It currently also supports Python 2.7 until Python 2 reaches end-of-life in 2020. After this date, Python 2 support will halt, and only 3.x versions will be maintained.


Install using pip:

pip install nteract-scrapbook

For installing optional IO dependencies, you can specify individual store bundles, like s3 or azure:

pip install nteract-scrapbook[s3]

or use all:

pip install nteract-scrapbook[all]

Models and Terminology

Scrapbook defines the following items:

  • scraps: serializable data values and visualizations such as strings, lists of objects, pandas dataframes, charts, images, or data references.
  • notebook: a wrapped nbformat notebook object with extra methods for interacting with scraps.
  • scrapbook: a collection of notebooks with an interface for asking questions of the collection.
  • encoders: a registered translator of data to/from notebook storage formats.

scrap model

The scrap model houses a few key attributes in a tuple, including:

  • name: The name of the scrap
  • data: Any data captured by the scrapbook api call
  • encoder: The name of the encoder used to encode/decode data to/from the notebook
  • display: Any display data used by IPython to display visual content


Scrapbook adds a few basic api commands which enable saving and retrieving data including:

  • glue to persist scraps with or without display output
  • read_notebook reads one notebook
  • scraps provides a searchable dictionary of all scraps by name
  • reglue which copies a scrap from another notebook to the current notebook
  • read_notebooks reads many notebooks from a given path
  • scraps_report displays a report about collected scraps
  • papermill_dataframe and papermill_metrics for backward compatibility for two deprecated papermill features

The following sections provide more detail on these api commands.

glue to persist scraps

Records a scrap (data or display value) in the given notebook cell.

The scrap (recorded value) can be retrieved during later inspection of the output notebook.

"""glue example for recording data values"""
import scrapbook as sb

sb.glue("hello", "world")
sb.glue("number", 123)
sb.glue("some_list", [1, 3, 5])
sb.glue("some_dict", {"a": 1, "b": 2})
sb.glue("non_json", df, 'arrow')

The scrapbook library can be used later to recover scraps from the output notebook:

# read a notebook and get previously recorded scraps
nb = sb.read_notebook('notebook.ipynb')

scrapbook will imply the storage format by the value type of any registered data encoders. Alternatively, the implied encoding format can be overwritten by setting the encoder argument to the registered name (e.g. "json") of a particular encoder.

This data is persisted by generating a display output with a special media type identifying the content encoding format and data. These outputs are not always visible in notebook rendering but still exist in the document. Scrapbook can then rehydrate the data associated with the notebook in the future by reading these cell outputs.

With display output

To display a named scrap with visible display outputs, you need to indicate that the scrap is directly renderable.

This can be done by toggling the display argument.

# record a UI message along with the input string
sb.glue("hello", "Hello World", display=True)

The call will save the data and the display attributes of the Scrap object, making it visible as well as encoding the original data. This leans on the IPython.core.formatters.format_display_data function to translate the data object into a display and metadata dict for the notebook kernel to parse.

Another pattern that can be used is to specify that only the display data should be saved, and not the original object. This is achieved by setting the encoder to be display.

# record an image without the original input object

Finally the media types that are generated can be controlled by passing a list, tuple, or dict object as the display argument.

  display=('text/plain',) # This passes [text/plain] to format_display_data's include argument

  display={'exclude': 'text/plain'} # forward to format_display_data's kwargs

Like data scraps, these can be retrieved at a later time be accessing the scrap's display attribute. Though usually one will just use Notebook's reglue method (described below).

read_notebook reads one notebook

Reads a Notebook object loaded from the location specified at path. You've already seen how this function is used in the above api call examples, but essentially this provides a thin wrapper over an nbformat's NotebookNode with the ability to extract scrapbook scraps.

nb = sb.read_notebook('notebook.ipynb')

This Notebook object adheres to the nbformat's json schema, allowing for access to its required fields.

nb.cells # The cells from the notebook

There's a few additional methods provided, most of which are outlined in more detail below:


The abstraction also makes saved content available as a dataframe referencing each key and source. More of these methods will be made available in later versions.

# Produces a data frame with ["name", "data", "encoder", "display", "filename"] as columns
nb.scrap_dataframe # Warning: This might be a large object if data or display is large

The Notebook object also has a few legacy functions for backwards compatibility with papermill's Notebook object model. As a result, it can be used to read papermill execution statistics as well as scrapbook abstractions:

nb.cell_timing # List of cell execution timings in cell order
nb.execution_counts # List of cell execution counts in cell order
nb.papermill_metrics # Dataframe of cell execution counts and times
nb.papermill_record_dataframe # Dataframe of notebook records (scraps with only data)
nb.parameter_dataframe # Dataframe of notebook parameters
nb.papermill_dataframe # Dataframe of notebook parameters and cell scraps

The notebook reader relies on papermill's registered iorw to enable access to a variety of sources such as -- but not limited to -- S3, Azure, and Google Cloud.

scraps provides a name -> scrap lookup

The scraps method allows for access to all of the scraps in a particular notebook.

nb = sb.read_notebook('notebook.ipynb')
nb.scraps # Prints a dict of all scraps by name

This object has a few additional methods as well for convenient conversion and execution.

nb.scraps.data_scraps # Filters to only scraps with `data` associated
nb.scraps.data_dict # Maps `data_scraps` to a `name` -> `data` dict
nb.scraps.display_scraps # Filters to only scraps with `display` associated
nb.scraps.display_dict # Maps `display_scraps` to a `name` -> `display` dict
nb.scraps.dataframe # Generates a dataframe with ["name", "data", "encoder", "display"] as columns

These methods allow for simple use-cases to not require digging through model abstractions.

reglue copys a scrap into the current notebook

Using reglue one can take any scrap glue'd into one notebook and glue into the current one.

nb = sb.read_notebook('notebook.ipynb')
nb.reglue("table_scrap") # This copies both data and displays

Any data or display information will be copied verbatim into the currently executing notebook as though the user called glue again on the original source.

It's also possible to rename the scrap in the process.

nb.reglue("table_scrap", "old_table_scrap")

And finally if one wishes to try to reglue without checking for existence the raise_on_missing can be set to just display a message on failure.

nb.reglue("maybe_missing", raise_on_missing=False)
# => "No scrap found with name 'maybe_missing' in this notebook"

read_notebooks reads many notebooks

Reads all notebooks located in a given path into a Scrapbook object.

# create a scrapbook named `book`
book = sb.read_notebooks('path/to/notebook/collection/')
# get the underlying notebooks as a list
book.notebooks # Or `book.values`

The path reuses papermill's registered iorw to list and read files form various sources, such that non-local urls can load data.

# create a scrapbook named `book`
book = sb.read_notebooks('s3://bucket/key/prefix/to/notebook/collection/')

The Scrapbook (book in this example) can be used to recall all scraps across the collection of notebooks:

book.notebook_scraps # Dict of shape `notebook` -> (`name` -> `scrap`)
book.scraps # merged dict of shape `name` -> `scrap`

scraps_report displays a report about collected scraps

The Scrapbook collection can be used to generate a scraps_report on all the scraps from the collection as a markdown structured output.


This display can filter on scrap and notebook names, as well as enable or disable an overall header for the display.

  scrap_names=["scrap1", "scrap2"],
  notebook_names=["result1"], # matches `/notebook/collections/result1.ipynb` pathed notebooks

By default the report will only populate with visual elements. To also report on data elements set include_data.


papermill support

Finally the scrapbook provides two backwards compatible features for deprecated papermill capabilities:



Encoders are accessible by key names to Encoder objects registered against the encoders.registry object. To register new data encoders simply call:

from encoder import registry as encoder_registry
# add encoder to the registry
encoder_registry.register("custom_encoder_name", MyCustomEncoder())

The encode class must implement two methods, encode and decode:

class MyCustomEncoder(object):
    def encode(self, scrap):
        # is any type, usually specific to the encoder name
        pass  # Return a `Scrap` with `data` type one of [None, list, dict, *six.integer_types, *six.string_types]

    def decode(self, scrap):
        # is one of [None, list, dict, *six.integer_types, *six.string_types]
        pass  # Return a `Scrap` with `data` type as any type, usually specific to the encoder name

This can read transform scraps into a json object representing their contents or location and load those strings back into the original data objects.


A basic string storage format that saves data as python strings.

sb.glue("hello", "world", "text")


sb.glue("foo_json", {"foo": "bar", "baz": 1}, "json")


Implementation Pending!

papermill's deprecated record feature

scrapbook provides a robust and flexible recording schema. This library replaces papermill's existing record functionality.

Documentation for papermill record exists on ReadTheDocs. In brief, the deprecated record function:

pm.record(name, value): enables values to be saved with the notebook [API documentation]

pm.record("hello", "world")
pm.record("number", 123)
pm.record("some_list", [1, 3, 5])
pm.record("some_dict", {"a": 1, "b": 2})

pm.read_notebook(notebook): pandas could be used later to recover recorded values by reading the output notebook into a dataframe. For example:

nb = pm.read_notebook('notebook.ipynb')

Rationale for Papermill record deprecation

Papermill's record function was deprecated due to these limitations and challenges:

  • The record function didn't follow papermill's pattern of linear execution of a notebook. It was awkward to describe record as an additional feature of papermill, and really felt like describing a second less developed library.
  • Recording / Reading required data translation to JSON for everything. This is a tedious, painful process for dataframes.
  • Reading recorded values into a dataframe would result in unintuitive dataframe shapes.
  • Less modularity and flexiblity than other papermill components where custom operators can be registered.

To overcome these limitations in Papermill, a decision was made to create Scrapbook.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for nteract-scrapbook, version 0.4.1
Filename, size File type Python version Upload date Hashes
Filename, size nteract_scrapbook-0.4.1-py3-none-any.whl (265.5 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size nteract-scrapbook-0.4.1.tar.gz (363.9 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page