Entity Resolution In A Few Lines Of Python
pip install zingg and resolve customers, suppliers and products from any notebook or Spark job. Describe your fields, point at your data, label a few pairs, train and match, all in Python, on Databricks, Microsoft Fabric, AWS or your own Spark.
Four Steps, One Script
Install
pip install zingg, then import the client and the pipes.
Describe Your Fields
One FieldDefinition per column, with a match type that says how to compare it.
Point At Your Data
Pipes read the input and write the output: Delta, Parquet, Snowflake, BigQuery, Unity Catalog, JDBC or a DataFrame you already have.
Run A Phase
Pick the phase with ClientOptions and call initAndExecute(). The same script labels, trains and matches.
from zingg.client import *
from zingg.pipes import *
args = Arguments()
fname = FieldDefinition("fname", "string", MatchType.FUZZY)
email = FieldDefinition("email", "string", MatchType.EMAIL)
dob = FieldDefinition("dob", "string", MatchType.EXACT)
args.setFieldDefinition([fname, email, dob])
args.setModelId("customers")
args.setZinggDir("models")
customers = Pipe("customers", "delta")
customers.addProperty("location", "/lake/crm/customers")
args.setData(customers)
resolved = Pipe("resolved", "delta")
resolved.addProperty("location", "/lake/crm/resolved")
args.setOutput(resolved)
options = ClientOptions([ClientOptions.PHASE, "match"])
zingg = Zingg(args, options)
zingg.initAndExecute()pip install zinggResolve Snowflake Data Your Way
Read and write Snowflake tables from Python, or run Zingg Enterprise natively inside your Snowflake account.
Snowflake Tables From Python
SnowflakePipe reads your Snowflake tables and writes resolved records back, through the Snowflake Spark connector. Run it from the Spark you already use, such as Databricks, Fabric or EMR.
Native Inside Snowflake
Zingg Enterprise runs inside your Snowflake account on Snowpark Container Services. You start each phase from SQL as a Snowflake job, with no servers outside Snowflake, so records never leave it.
import os
from zingg.pipes import *
customers = SnowflakePipe("customers")
customers.setURL("myorg-myaccount.snowflakecomputing.com")
customers.setUser(os.environ["SF_USER"])
customers.setPassword(os.environ["SF_PASSWORD"])
customers.setDatabase("CRM")
customers.setSFSchema("PUBLIC")
customers.setWarehouse("COMPUTE_WH")
customers.setDbTable("CUSTOMERS")
args.setData(customers)The Whole Of Zingg, From Python
Fields And Match Types
Fuzzy names and addresses, exact identifiers, emails, postcodes, numbers with units, long text. Mark a column DONT_USE to carry it through to the output without matching on it.
Read And Write Anywhere
Delta and Parquet through Pipe, SnowflakePipe, BigQueryPipe, UCPipe for Unity Catalog tables, and JDBC or any Spark source. Or pass a DataFrame from your own code.
Every Phase From One Script
findTrainingData, label, train, match and link, plus findAndLabel and trainMatch to do two steps in one run.
Notebooks On Your Spark
ZinggWithSpark reuses the session your notebook already has, on Databricks, Microsoft Fabric, AWS EMR or Glue.
Part Of Your Pipelines
It is a Python program, so it runs wherever your jobs run: Airflow, Databricks Workflows, Fabric pipelines, keeping the configuration in Git.
Enterprise In The Same Shape
The Zingg Enterprise Python API adds deterministic matching, incremental runs, mapping match types for nicknames, explanations of every match and output statistics.
Tell Zingg How Each Field Compares
Each FieldDefinition takes a MatchType. Zingg learns how much each field matters from the pairs you label.
| MatchType | What it compares |
|---|---|
| FUZZY | Typo-tolerant matching for names, addresses and free text |
| EXACT | A strong signal when two values are identical |
| Compares the part of an email before the @ | |
| PINCODE | Postal and ZIP codes across common formats |
| NUMERIC | Numbers inside strings, ignoring the letters around them |
| NUMERIC_WITH_UNITS | Numbers with their units, such as 500 ml or 16 GB |
| TEXT | Overlapping words in longer descriptions |
| ONLY_ALPHABETS_EXACT / FUZZY | Letters only, numbers stripped, compared exactly or fuzzily |
| NULL_OR_BLANK | Lets the model learn what a missing value means |
| DONT_USE | Kept in the output, not used for matching |
Same API, More For Production
| Zingg Community | Zingg Enterprise | |
|---|---|---|
| Package | pip install zingg | pip install zinggEC |
| Arguments | Arguments, FieldDefinition | EArguments, EFieldDefinition |
| Matching | AI matching trained on your labels | AI matching plus deterministic rules in the same run |
| Updates | Rerun match on the full data | runIncremental for new, changed and deleted records |
| Nicknames and abbreviations | Not included | MAPPING match types from a lookup file |
| Sources | Delta, Parquet, Snowflake, BigQuery, Unity Catalog, JDBC, DataFrames | Adds in-memory pipes |
| Explanations | Not included | Why records matched: the pairs, their scores and how each cluster formed |
| Output | Clusters with a Zingg ID per run | Persistent ZINGG_ID, pass-through records and run statistics |
Python API Questions
How do I install the Zingg Python API?
Run pip install zingg for Zingg Community. Zingg Enterprise uses pip install zinggEC and needs an Enterprise licence. The Python API runs on Spark.
Where can I run Zingg Python programs?
On any Spark you have: Databricks, Microsoft Fabric, AWS EMR, AWS Glue or a local Spark. In a notebook, ZinggWithSpark uses the notebook's Spark session. As a script, run it with zingg.sh --run.
Do I still need a JSON configuration file?
No. The Python API sets the same fields, data and model options in code, so the configuration lives with the rest of your pipeline.
Can I label training pairs from Python?
Yes. Run findTrainingData and label, or findAndLabel to do both, then train. In notebooks you can label pairs in the notebook itself.
Can I use the Zingg Python API with Snowflake?
Yes. SnowflakePipe reads Snowflake tables as input and writes the resolved output back, through the Snowflake Spark connector. Zingg Enterprise can also run natively inside Snowflake on Snowpark Container Services, started from SQL.
Can Zingg read a DataFrame I already have?
Yes. Pass the DataFrame as the input instead of a file or table pipe, and Zingg matches it on the same Spark session.
Start Resolving From Your Notebook
Install the open source package today, or talk to us about Zingg Enterprise.