In this tutorial you will learn how to perform a tumbling windowTumbling windowA strategy for processing (stream) data by specific limited frames, usually time ... Read more operation to detect suspicious activity.
Your task is to detect suspicious user login attempts during some period of time.
You have a record of login data. Your goal is to detect suspicious users who have logged in more than 5 times in a single minute.
To do this, you will be using the windowby syntax with a pw.temporal.tumbling() object. Let's jump in!
Your input data table has the following columns:
Let's start by ingesting the data:
First ingest the data.
# Uncomment to download the required files.
# %%capture --no-display
# !wget https://public-pathway-releases.s3.eu-central-1.amazonaws.com/data/suspicious_users_tutorial_logins.csv -O logins.csv
import pathway as pw
logins = pw.io.csv.read(
The CSV data has the string values "True" and "False" in the successful column.
Let's convert this to a Boolean column:
logins = logins.with_columns(successful=(pw.this.successful == "True"))
Then, let's filter attempts and keep only the unsuccessful ones.
failed = logins.filter(~pw.this.successful)
Now, perform a tumbling window operation with a duration of 60 (i.e. 1 minute).
Use the instance keyword to separate rows by the ip_address value.
result = failed.windowby(
failed.time, window=pw.temporal.tumbling(duration=60), instance=pw.this.ip_address
...and finally, let's keep only the IP addresses where the number of failed logins exceeded the threshold (5):
suspicious_logins = result.filter(pw.this.count >= 5)
[2023-12-08T16:31:25]:INFO:Preparing Pathway computation
[2023-12-08T16:31:25]:INFO:CsvFilesystemReader-0: 0 entries (1 minibatch(es)) have been sent to the engine
[2023-12-08T16:31:25]:INFO:CsvFilesystemReader-0: 30 entries (2 minibatch(es)) have been sent to the engine
[2023-12-08T16:31:25]:WARNING:CsvFilesystemReader-0: Closing the data source
| ip_address | count
^23S3BS8... | 188.8.131.52 | 7
And that's it! You have used a tumbling window operation to identify suspicious user activity and can now act on this information to increase the security of your platform.
Reach out to us on Discord if you'd like to discuss real time anomaly detectionReal Time Anomaly detectionTraining models with high volume of data allows the identification of unusual dat... Read more use cases like this one in more detail!
Data Structures and Optimization Expert
Realtime Server Log Monitoring
Time between events in a multi-topic event stream