Join us on July 16th as we explore real-world Reinforcement Learning Learn more

Extract metadata from a dataset

Level 2
Extract metadata from a dataset

I would like to extract some metadata (for example the column names) from a dataset.

 

Is there any existing recipe or preparation processor doing that, or have I to write it myself in a plugin ?

0 Kudos
6 Replies
Dataiker
Dataiker

Hi,

I would suggest leveraging the Python APIs to retrieve the information you need, such as the metadata or corresponding schema, from these datasets:
https://doc.dataiku.com/dss/latest/python-api/datasets.html
https://doc.dataiku.com/dss/latest/python-api/rest-api-client/datasets.html

More information about available functions can be found at the bottom under the reference doc section. 

Best,
Andrew

0 Kudos
Level 2
Author
So I understand the answer as "there isn't any existing solution" and I had to code it myself 😉

I think it could be a good thing to add as a standard recipe / processor
0 Kudos
Dataiker
Dataiker

Hi,

Sure we appreciate the feedback and I'll forward it on your behalf to our Product team for further review. 

Best,

Andrew

0 Kudos
Level 3

Here's some code that may be helpful. It illustrates reading the schema and some other dataset attributes. This example is geared toward SQL datasets (our primary use) so some of the details may differ for other types of datasets.

ds = dataiku.Dataset('FEATURES')

# location_info
loc_info = ds.get_location_info()
ds_info = loc_info['info']
if loc_info['locationInfoType'] == 'SQL' and 'table' not in ds_info:
    table = '<SQL Query>'
else:
    table = ds_info['table']
    
print('Dataset Type: {}'.format(loc_info['locationInfoType']))
print('Connection Name: {}'.format(ds_info['connectionName']))
print('Database Type: {}'.format(ds_info['databaseType']))
print('Table Name: {}'.format(table))

# schema
print('\nColumns in Dataset (Name - dss type / database type):')
for col in ds.read_schema():
    print('{0} - {1} / {2}'.format(col['name'], col['type'], col['originalType']))
    
# Note that ds.get_config() has all of above plus many other config items

 

Level 2
Hi @florianbriand 
What did you want to do with the metadata?  Is writing it into a file (e.g. a custom recipe) very convenient?
Herc
Level 2
Author

For information, the plugin I wrote is as simple as :

import dataiku
from dataiku.customrecipe import *
import pandas as pd
from pprint import pprint

main_input_names = get_input_names_for_role('main_input')
main_input_ds = dataiku.Dataset(main_input_names[0])
print("------------- INPUT -------------")
pprint(main_input_ds)

# For outputs, the process is the same:
main_output_names = get_output_names_for_role('main_output')
main_output_ds = dataiku.Dataset(main_output_names[0])

schema = main_input_ds.read_schema()  # {name, type} for each column
print("------------- SCHEMA -------------")
pprint(schema)

main_output_df = pd.DataFrame(schema)
print("------------- OUTPUT -------------")
pprint(main_output_df)

main_output_ds.write_with_schema(main_output_df)

 

Because in my case, I didn't need anything else than the schema.

But there are probably other metadata which could be helpful.

From my side, the need is just to get the columns list to make things like :

- check if every column required in subsequent flow are provided and give user information about missing columns
- automatically configure some other recipes, via variables, with the input columns
- ...

0 Kudos