Coinbase Data Scientist Interview Questions

Coinbase deploys several models to predict and analyze risk based on user transactions. Create a design for a Model Versioning Registry that manages versions of machine learning models.Problem statement: Implement a ModelVersionRegistry class that provides functionalities to register new models, retrieve information about a specific model by version, and get the latest model version. The registry should support the following methods:
- register_model(model_name: str, version: str, metadata: dict): Registers a model with a specific version and metadata.
- get_model_by_version(model_name: str, version: str): Retrieves the model data for the given model name and version.
- get_latest_model(model_name: str): Gets the latest version of the registered model.Example 1:
Input:
registry = ModelVersionRegistry()
registry.register_model('risk_model', '1.0', {'accuracy': 0.95})
registry.register_model('risk_model', '1.1', {'accuracy': 0.97})Output: registry.get_latest_model('risk_model') returns {'version': '1.1', 'metadata': {'accuracy': 0.97}}Constraints:
- Model names and version strings are alphanumeric with a maximum length of 50 characters.
- Metadata can be a dictionary with string keys and any value type.

Coinbase needs to monitor user activity for improved insights and to manage features effectively. Your challenge is to create a User Activity Tracker that tracks user transactions over time and provides some analytical insights.Problem statement: Create a class UserActivityTracker capable of adding and retrieving user transaction counts over time. The class should provide the following functionalities:
- add_transaction(user_id: int): Adds a transaction for the specified user.
- get_transaction_count(user_id: int) -> int: Returns the total number of transactions for the specified user.
- get_total_transactions() -> int: Returns the total transaction count across all users.Example 1:
Input:
tracker = UserActivityTracker()
tracker.add_transaction(1)
tracker.add_transaction(2)
tracker.add_transaction(1)
Output: tracker.get_transaction_count(1) returns 2
Output: tracker.get_total_transactions() returns 3Constraints:
- User IDs range between 1 and 10^4.
- Number of transactions can go up to 10^5.

Coinbase often analyses transaction data to understand product adoption. You need to help find the number of distinct products a user has interacted with in a series of transaction records.Problem statement: Given a list of transaction records where each record is represented as a tuple containing (user_id, product_id), write a function to return the number of distinct products that each user has interacted with.- Method Signature: def count_distinct_products(transactions: List[Tuple[int, int]]) -> Dict[int, int]: Returns a dictionary with user_id as keys and the count of distinct product_ids as values.Example 1:
Input: transactions = [(1, 101), (1, 102), (2, 101), (1, 101), (2, 103)]
Output: {1: 2, 2: 2}
Explanation: User 1 interacted with products 101 and 102, while User 2 interacted with products 101 and 103.Constraints:
- 1 <= len(transactions) <= 10^5
- 1 <= user_id, product_id <= 10^4

Coinbase analytics processes vast amounts of transaction history data. You need to help the data team find the maximum sum of any contiguous subarray of transaction amounts over a specified transaction window.Problem statement: Write a function that returns the maximum sum of a contiguous subarray of n transaction amounts within the transaction history array. You need to capture the transaction amounts over a window_size.- Method Signature: def max_subarray_sum(transactions: List[int], window_size: int) -> int: Returns the maximum sum of a contiguous subarray of the specified size.Example 1:
Input: transactions = [1, -2, 3, 4, -1, 2, 1, -5, 4], window_size = 3
Output: 6
Explanation: The subarray [3, 4, -1] has the maximum sum of 6.Constraints:
- 1 <= len(transactions) <= 10^5
- -10^4 <= transactions[i] <= 10^4
- 1 <= window_size <= len(transactions)