Create safe data

The first and only APIs to enable you to balance, anonymize, and share your data. With privacy guarantees.

Discover. Transform. Share.

Create an anonymized or synthetic dataset to safely work with data while preserving privacy.

hello_gretel.py

from gretel_client import get_cloud_client

client = get_cloud_client("api", GRETEL_API_KEY)
my_data = load_from_csv("my_sensitive_data.csv")
labeled_data = client.detect_entities(my_data)

print(labeled_data)

$~ python hello_gretel.py
{
  "project_record_count": 11515,
  "total_entity_count": 8,
  "data": {
    "entities": [
      {
        "entity": "location",
        "count": 76151.0,
        "last_seen": "2020-07-06T20:55:54Z",
        "approx_card": 32748
      },
      {
        "entity": "latitude",
        "count": 23030.0,
        "last_seen": "2020-07-06T20:55:54Z",
        "approx_card": 10678
      },
      {
        "entity": "longitude",
        "count": 23030.0,
        "last_seen": "2020-07-06T20:55:54Z",
        "approx_card": 10905
      },
      {
        "entity": "md5",
        "count": 17353.0,
        "last_seen": "2020-07-06T20:55:54Z",
        "approx_card": 14377
      },
      {
        "entity": "url",
        "count": 17353.0,
        "last_seen": "2020-07-06T20:55:54Z",
        "approx_card": 109
      },
      {
        "entity": "uuid",
        "count": 15885.0,
        "last_seen": "2020-07-06T20:55:54Z",
        "approx_card": 13385
      },
      {
        "entity": "datetime",
        "count": 11515.0,
        "last_seen": "2020-07-06T20:55:54Z",
        "approx_card": 119
      },
      {
        "entity": "person_name",
        "count": 1013.0,
        "last_seen": "2020-07-06T20:55:42Z",
        "approx_card": 49
      }
    ]
  }
}

hello_gretel.py

import pandas as pd

from gretel_synthetics.batch import DataFrameBatch

config = {
"epochs": 15,
"vocab_size": 20000,
"dp": True, # enable differential privacy,
"field_delimiter": ",",
"checkpoint_dir": "my_models"
}

df = pd.read_csv("my_sensitive_data.csv")

gretel = DataFrameBatch(df=df, config=config)
gretel.create_training_data()
gretel.train_all_batches() # build AI models
gretel.generate_all_batch_lines()

synthetic_data_df = gretel.batches_to_df()
synthetic_data_df.head()

transform.py

from gretel_client.transformers import DataPath, DataTransformPipeline, FakeConstantConfig, RedactWithLabelConfig

SEED = 8675309
RECORD = {
"name": "Clark Kent",
"address": "344 Clinton Street, Apartment 3D, Metropolis",
"mobile": "800-579-7866",
"email": "[email protected]"
}

name = FakeConstantConfig(seed=SEED, fake_method="name")
address = FakeConstantConfig(seed=SEED, fake_method="address")
auto_redact = RedactWithLabelConfig(labels=["phone_number", "email_address"])

paths = [
DataPath(input="name", xforms=[name]),
DataPath(input="address", xforms=[address]),
DataPath(input="*", xforms=[auto_redact])
]

pipe = DataTransformPipeline(paths)
print(pipe.transform_record(RECORD))

$ python transform.py
{
    "name": "Dr. Julia Evans, MD",
    "address": "1061 Autumn Station Apt. 934 Lake Tiffany, FL 56915",
    "mobile": "PHONE_NUMBER",
    "email": "EMAIL_ADDRESS"
}

transform.py

import requests

from gretel_client import project_from_uri
from gretel_client.transformers import DataPath, DataTransformPipeline, FakeConstantConfig

# We don't want to write sensitive data into our analysis store
faker = FakeConstantConfig(
    seed=12345,
    labels=["email_address", "person_name", "phone_number", "date_of_birth", "address"])
path = [DataPath(input="*", xforms=[faker])]
transformer = DataTransformPipeline(path)

project = project_from_uri(GRETEL_URI)

# stream labeled data from the Gretel API, sanitize and
# send it to the analysis store
while True:
    for record in project.iter_records():
        safe_record = transformer.transform_record(record)
        requests.post("https://analysis.store:9200/customers/_doc", json=safe_record)