Link Search Menu Expand Document

Connect Soda to Dask and Pandas (Experimental)

Last modified on 26-Jan-23

For use with programmatic Soda scans, only.

Define a programmatic scan for the data in the DataFrames. Refer to the following example.

import dask.datasets
import pandas as pd
from soda.scan import Scan

# Create a Soda scan object
scan = Scan()

# Load timeseries data from dask datasets
df_timeseries = dask.datasets.timeseries().reset_index()
df_timeseries["email"] = ""

# Create an artificial pandas dataframe
df_employee = pd.DataFrame({"email": ["", "", ""]})

# Add Dask dataframe to scan and assign a dataset name to refer from checks yaml
scan.add_dask_dataframe(dataset_name="timeseries", dask_df=df_timeseries)

# Add Pandas dataframe to scan and assign a dataset name to refer from checks yaml
scan.add_pandas_dataframe(dataset_name="employee", pandas_df=df_employee)

# Define checks in yaml format
# Alternatively, you can refer to a yaml file using scan.add_sodacl_yaml_file(<filepath>)
checks = """
for each dataset T:
    - include %
    - row_count > 0
profile columns:
    - employee.%
checks for employee:
    - values in (email) must exist in timeseries (email) # Error expected
    - row_count same as timeseries # Error expected
checks for timeseries:
  - avg_x_minus_y between -1 and 1:
      avg_x_minus_y expression: AVG(x - y)
  - failed rows:
      samples limit: 50
      fail condition: x >= 3
  - schema:
      name: Confirm that required columns are present
        when required column missing: [x]
        when forbidden column present: [email]
        when wrong column type:
          email: varchar
        when required column missing:
          - y
  - invalid_count(email) = 0:
      valid format: email
  - valid_count(email) > 0:
      valid format: email



Was this documentation helpful?

What could we do to improve this page?

Last modified on 26-Jan-23