Importing/exporting data#

This section of the user guide covers importing/exporting data to the Virtual Data Lake (VDL) using crandas. Information on how to upload data via the web portal can be found in the knowledge base.

Users can upload existing pandas DataFrames, create new crandas CDataFrames, or upload CSV files. Access tables via handle or name and use for aggregated information.

Upload pandas Dataframe#

If you have an existing data in a pandas DataFrame that you want to upload to the VDL, you can use the crandas.upload_pandas_dataframe() function. This function takes a pandas DataFrame as its parameter and uploads it to the VDL. You can also optionally specify a name for the table.

For example, let’s say you have a pandas DataFrame called my_data that you want to upload to the VDL:

import crandas as cd
import pandas as pd

my_data = pd.DataFrame({"fruit": ["orange", "apple", "raspberry"]})
uploaded_data = cd.upload_pandas_dataframe(my_data)

This will upload my_data to the VDL and return a CDataFrame object that you can use to interact with the uploaded data. A CDataFrame behaves similarly to a pandas DataFrame, however it is stored in secret-shared form in the VDL.

The advantage of this approach is that it enables users to read any file type that is accepted by pandas by first utilizing pandas, followed by crandas.

Create new crandas CDataFrame#

Alternatively, if you want to create a new crandas CDataFrame from scratch, you can use the crandas.DataFrame() function. This function calls the pandas DataFrame constructor and uploads the resulting table using upload_pandas_dataframe(). If you specify a name for the table, it will be passed on to upload_pandas_dataframe().


When uploading data with missing values, it is important to specify certain additional data. For more information look here.

For example, let’s say you want to create a brand new CDataFrame called my_table with columns A, B, and C.

my_table = cd.DataFrame({
    "A": [1, 2, 3],
    "B": [4, 5, 6],
    "C": [7, 8, 9]
}, name="my_table")

This will create a new CDataFrame called my_table with the columns that we specified and upload it to the Virtual Data Lake. You can now use the my_table object to interact with the uploaded data.

Handles and names#

State objects, such as tables, now have two distinct identifiers: handles and names.

  • Handles are randomly generated 32-byte strings, usually encoded as 64 hexadecimal digits. Each table has a unique handle that remains fixed (e.g. 6E14C5275C5E90E31D84FCE0CE5F6D3D1BFE587C21D2278C52D4A092C4AB19F7).

  • Names are a user-friendly way to refer to handles and therefore tables. Handles are generated by the VDL and will always be unique but names are assigned by a user. If a table is given a name already assigned to a different table, the old table will lose its name but will still be accessible by its handle.

When uploading a new table with an existing name, the server reassigns the name to the new table, and the old name-handle mapping is lost. This is in contrast to handles, which always remain unique for each table.

To upload a table with a name, use the following syntax:

cdf = cd.upload_pandas_dataframe(df, name="input")

The name will then be assigned.

Upload CSV file to the VDL#

To upload a CSV file to the VDL, you can utilize the crandas.read_csv() function. This function accepts the name of the CSV file as its parameter and facilitates its upload to the server. In addition, users may opt to specify a name for the resulting table.

To upload a file called my_data.csv to the VDL, we simply need to do this:

uploaded_data = cd.read_csv("my_data.csv")

This will upload my_data.csv to the VDL and return a CDataFrame object that you can use to interact with the uploaded data. Note that for this to work, the file must be in the current directory, otherwise we must specify the path to it.

Access an uploaded table#

Any table that has been uploaded to the VDL can be referenced by its handle: a hexadecimal string of characters. To access a table, we use the crandas.get_table() function. This function takes the table’s handle/name as its parameter and returns a CDataFrame.

# Using the handle
my_table = cd.get_table("63FE905BB6DF9AD2E7D32DD092C75B1FC2CEB52BDBC4AEAB7AAEF14DBFCB6224")

This will return the CDataFrame object for my_table, which you can then perform operations on.

Instead of referencing a table by its handle, it can also be accessed by its name. If you (or someone you are collaborating with) have previously uploaded a table to the VDL with the name my_table, you can access it by that name:

# Using the name
my_table = cd.get_table("my_table")

However, this is not without risk: If a script is allowed to process any dataset having the given name, the analyst could bypass the approved script by renaming and inserting their own input tables. Therefore, we recommend to use table handles instead.

How to open CDataFrames#

After performing a a computation, you will want to access the resulting data. The method allows you to retrieve a CDataFrame and open it. This downloads the open data, exposing it and not making it private. In general, opening CDataFrames is not allowed. In non-demo environments, there will be strict controls over which CDataFrames can be opened.


An attempt to use .open() in a production environment will be met with an error.

Given a CDataFrame we can retrieve it using which will output a pandas DataFrame.

# create the CDataFrame
df = cd.DataFrame({'A': [1, 2], 'B': [3, 4]})
# open the CDataFrame
opened_df =

The opened table opened_df is now a normal pandas DataFrame with data in the clear.

>>> print(opened_df)
   A  B
0  1  3
1  2  4

These are the ways we can import and export data in the VDL.

By using these functions, you can easily upload data to VDL for further privacy-preserving analysis and processing.

Listing uploaded data#

It is possible to obtain a list of all tables that have been uploaded to the VDL using the function crandas.stateobject.list_uploads(). The result of this function is a pandas dataframe with handles and metadata:

>>> cd.stateobject.list_uploads()
                                              handle                   created                             type
0  91E4033337F1ED4D13ED23CA4DCBFB279FBA4C58C7E249... 2023-12-04 12:49:00+00:00  CDataFrame (3 rows x 2 columns)
1  DF20C85FB51E7F114822B6FCF865A260D42872214795F7... 2023-12-04 12:48:16+00:00  CDataFrame (4 rows x 2 columns)

Note that this function only list uploads: this does not include computation results (e.g., the result of joining two tables) or demo tables created using crandas.crandas.demo_table(). See the documentation of crandas.stateobject.list_uploads() for more details.

Removing objects from the VDL#

After working with data, you might want to delete it from the VDL. This is as simple as calling the StateObject.remove() method.

# You simply remove a CDataFrame using the following command

This will not only get rid of the python CDataFrame used to interact with the table, but also the table in the server.

Now that we know how to add data to the VDL, we can learn how that data is structured so we can start working with it.