Apple Data Scientist Interview Questions

In an effort to streamline the deployment of machine learning models in production, Apple requires a robust model versioning system to keep track of different versions of models, their metadata, and deployment statuses.
Problem statement: Define a class ModelRegistry that can manage the versions of models efficiently. The class should support the following methods:
- add_model(name: str, version: str, metadata: dict) -> None: Add a new model with its metadata.
- get_model(name: str, version: str) -> Optional[dict]: Retrieve the metadata of a specific model version.
- delete_model(name: str, version: str) -> bool: Delete a specific model version from the registry.Example 1:
Input: model_registry.add_model('ImageClassifier', 'v1.0', {'accuracy': 0.95})
No output.
Input: model_registry.get_model('ImageClassifier', 'v1.0')
Output: {'accuracy': 0.95}.
Input: model_registry.delete_model('ImageClassifier', 'v1.0')
Output: True.Constraints:
- 1 <= len(name) <= 100
- 1 <= len(version) <= 20.
- Metadata is ideally a dictionary with string keys and values.

To enhance WatchOS features, Apple is interested in understanding how frequently users are opening apps on their Watches. You are tasked with creating a SQL query that will show the daily active users for each app.
Problem statement: Write a SQL query that returns the app_name, date, and the count of distinct active users for each app on each date. The data is in a table named app_usage with columns user_id, app_name, and timestamp.Example 1:
Input: SELECT ... FROM app_usage
Output: app_name | date | active_users
-----------|-------------|-------------
Music | 2023-10-01 | 150
Weather | 2023-10-01 | 120Example 2:
Input: SELECT ... FROM app_usage
Output: app_name | date | active_users
-----------|-------------|-------------
Fitness | 2023-10-02 | 200Constraints:
- 1 <= D <= 10^5 (number of rows in table)
- 1 <= user_id, app_name <= 1000.

To improve user engagement, Apple Music may want to forecast future downloads using historical data. Your task is to predict download trends based on a time series of previous downloads.
Problem statement: Write a function forecast_downloads(data: List[Tuple[str, int]], months: int) -> List[float] that takes the historical download counts and forecasts future counts for the specified months.
- data: a list of tuples containing the date in YYYY-MM-DD format and download count.
- months: an integer representing the number of months to forecast.Example 1:
Input: data = [("2023-01-01", 200), ("2023-02-01", 240), ("2023-03-01", 300)], months = 3
Output: [340.0, 380.0, 420.0]
Explanation: The pattern suggests an increase of 40 downloads per month.Example 2:
Input: data = [("2023-01-01", 150), ("2023-02-01", 100), ("2023-03-01", 50)], months = 2
Output: [0.0, 0.0]Constraints:
- 1 <= len(data) <= 1000
- 0 <= downloads <= 10^5.

Apple is looking to enhance customer experience by personalizing recommendations based on their purchasing behavior. Analyzing user transactions to find similar purchasing patterns can help in creating targeted marketing strategies.
Problem statement: Create a function group_users(transactions: List[List[int]], threshold: int) -> List[List[int]] that takes a list of user transactions and groups users by shared purchasing patterns based on a threshold of similarity.
- transactions: a list of user transaction lists where each inner list contains integers representing item IDs.
- threshold: an integer representing the minimum number of shared item IDs for users to be grouped together.Example 1:
Input: transactions = [[1, 2, 3], [2, 3], [2, 3, 4], [5, 6]]
Output: [[1, 2, 3], [2, 3], [2, 3, 4]]
Explanation: User 1 has common items with Users 2 and 3; User 4 has no overlap.Example 2:
Input: transactions = [[1], [1, 2], [2, 3], [2], [5]]
Output: [[1, 1, 2], [2, 3, 2]]Constraints:
- 1 <= len(transactions) <= 100
- 0 <= threshold <= 10
- 0 <= item IDs <= 10^5.