Streaming Live Events

This process is very simple, but it requires that each event is structured as a Python dictionary that maps field names (as they are onboarded with Fiddler) to values.

Example 1: A simple three-input fraud model.

my_event = {
    "age": 30,
    "gender": "Male",
    "salary": 80000.0,
    "predicted_fraud": 0.89,
    "is_fraud": 1
}

🚧

Note

If you have a pandas DataFrame, you can easily convert it into a list of event dictionaries in the above form by using its to_dict function.

my_events = my_df.to_dict(orient="records")

Then to upload the event to Fiddler, all you have to do is call the Fiddler client's fdl.Model.publish method.

# For a single event
model.publish([my_event])

# For multiple even
model.publish(my_events)

After calling the function, please allow 3-5 minutes for events to populate the Monitor page.

📘

Info

The event_timestamp field should contain the Unix timestamp in milliseconds for the time the event occurred. This timestamp will be used to plot the event on time series charts for monitoring.

If you do not specify an event timestamp, the current time will be used.

Example 2: Bank churn event

Here's an example using a bank churn model.

# Publish an event
model.publish([{
  "CreditScore": 650,      # data type: int
  "Geography": "France",   # data type: category
  "Gender": "Female",
  "Age": 45,
  "Tenure": 2,
  "Balance": 10000.0,      # data type: float
  "NumOfProducts": 1,
  "HasCrCard": "Yes",
  "isActiveMember": "Yes",
  "EstimatedSalary": 120000,
  "probability_churned": 0.105,
  "churn": 1
}]
)

The fdl.Model.publish API can be called in real-time right after your model inference.

📘

Info

You can also publish events as part of a batch call after the fact (click here for more information).