Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Learn how to design and implement effective Machine Learning systems from start to finish.
| Date | Stars |
|---|---|
| 2026-07-31 | 277 |
| 2026-08-06 | 277 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Machine learning design primer
Some helpful notes for Machine Learning System Design Interview preparation, which I gathered from various resources to prepare for machine learning systems design interview.
## Motivation
> Learn how to design Machine Learning systems and prepare for an interview.
## The main resources
[Facebook Field Guide to Machine Learning](https://research.facebook.com/blog/2018/05/the-facebook-field-guide-to-machine-learning-video-series/)
[CS 329S: Machine Learning Systems Design, Stanford, Winter 2022](https://stanford-cs329s.github.io/)
[ML Systems Design Interview Guide](http://patrickhalina.com/posts/ml-systems-design-interview-guide/)
[ML System Design interview example](https://youtu.be/VPg2Uu1MYgI)
[Yandex MLSD interview guide](https://youtu.be/krDfPEYVVvk)
["Machine Learning System Design" book](https://github.com/arsenyinfo/ml_system_design)
["Designing Machine Learning Systems" book](https://github.com/chiphuyen/dmls-book)
## Overview:
1. [Framework for solving MLSD cases](#framework-for-solving-mlsd-cases)
2. [Detailed notes on some concepts](#detailed-notes-on-some-concepts)
3. [Cases](#cases.md)
## Contributing
Feel free to submit pull requests to help:
+ Fix errors
+ Improve sections
+ Add new sections and cases
# Framework for solving MLSD cases
1. [Problem definition](#problem-definition)
2. [Data](#data)
3. [Evaluation](#evaluation)
4. [Features](#features)
5. [Model](#model)
6. [Error analysis](#error-analysis)
7. [Further actions](#further-actions)
## Overall tips
+ Ask questions
+ Tell pros and cons of different solutions
+ Start with a simple solution as a baseline
After listening to the conditions of the case, ask the interviewer clarifying questions. Repeat the main points to make sure you understand everything correctly. During the interview, ask the questions and state your assumptions.
**Understand requirements:**
+ Users/samples number
+ Peak numbers of requests
+ Batch or online predictions
+ Edge or server computations
## Problem definition.
Define proxy machine learning metric for the business goal.
1. Define the business goal.
2. Define ML task type. Classification/regression/other
3. Split the task into subtasks. Example: maximize users engagement while minimizing the spread of extreme views and misinformation.
## Data
1. [ Data source and data type](#data-source)
+ Where the data comes from? Is it in the same format or should we transform and join it?
+ One sample of data. What is X(features) and what is Y(labels)?
1. [Labeling](#Labeling)
+ Are the labels known? Is there Natural labelling? Should we label some data?
2. [Sampling](#sampling)
3. Data recency and Distribution drift.
## Evaluation
1. [Offline evaluation](#offline-evaluation)
1. **Data split**
+ Random split or should split by date, users, products to prevent data leakage?
2. **Metric**
+ Choose a metric, that is interpretable and sensitive to the task. Think what errors will be most harmful, FP or FN for classification, over or underpredicting for regression.
1. **Baseline**
+ Mention baseline Non-ml-solution. You will compare your machine learning models with this baseline.
1. [Online evaluation](#online-evaluation)
1. **Online-offline gap**
2. **Online comparing**
1. A/B randomised test
2. A/A test.
## Features
1. What type of data do we have? Can we encode it?
2. [Feature representation, data preprocessing.](#feature-representation)
3. Data augmentation.
## Model
1. Pick the model.
2. Pros and cons of the model.
3. Architecture overview at a glance.
+ Linear model
+ GBDT
+ Embeddings + KNN
+ Neural networks
1. What loss function you will use?
## Error analysis
1. Sample-wise learning curves (to estimate how model can be improved by adding new data);
2. Residual analysis (detecting bias and common errors);
3. Finding patterns in errors and residuals.
4. Addressing theExcerpt of 10,342 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:351ba2e1c6398541, topic:interview-questions