Your proven path to success in Machine Learning Interviews – developed by FAANG engineers. Unlock ML loops at top companies with a System Design approach.

System design is an important component of any ML interview. Being able to efficiently solve open-ended machine learning problems is a key skill that can set you apart from other engineers and increase the level of seniority at which you’re hired.

This course helps you build that skill, and goes over some of the most popularly asked interview problems at big tech companies. You’ll walk step-by-step through solving these problems, focusing in particular on how to design machine learning systems rather than just answering trivia-style questions. 

Once you’re done with the course, you’ll be able to not just ace the machine learning interview at any tech company, but impress them with your ability to think about systems at a high level. If you have a machine learning or system design interview coming up, you’ll find the course tremendously valuable.

Grokking the Machine Learning Interview

Features are the backbone of any learning system. Let's think about the main actors or dimensions that will play a key role in our feature engineering process. 
1. Ad
2. Advertiser
3. User
4. Context



> 📝Context refers to the engagement history, user interests, current location, time and date, as discussed in the previous lessons.

# Features for the model
Now it’s time to generate features based on these actors. The features would fall into the following categories:

1. Ad specific features


2. Advertiser specific features

3. User specific features


4. Context specific features

5. User-ad cross features


6. User-advertiser cross features

A subset of the features is shown below:


## Ad specific features

- **ad_id**
   
   A unique id is assigned to each ad and can be used as a sparse feature. Utilizing ad_id as a sparse feature allows the model to memorize historical engagement for each ad, and it can also be used in interesting cross features for memorization (such as ad_id * user interests). Additionally, we can also generate embeddings during training time for the ad using its id, as we will discuss in the ad prediction section.

- **ad_content_raw_terms**
    
   Ad terms can also be very useful sparse features. They can tell us a lot about the ad, e.g., a good model can learn from the text’s content to identify what the ad is about, such as politics or sports. Raw terms allow the models (especially NN models) to learn such behavior from given raw terms.
   
- **historical_engagement_rate**

  This feature specifies the rate of user engagement with the ad. Here we will measure engagement in different windows such as different times of the day or days of the week.
For instance, we can have the following features:
  - **ad_engagement_history_last_24_hrs**
    
    Since ads are short-lived, recent engagement is important. This feature captures the most recent engagement with the ad.

  - **ad_engagement_history_last_7_days**
    
    This captures the activity on the ad on each day of the week. For example, an ad can get more engagement on weekends rather than on weekdays. 
 
  > 📝 This feature can tell the model that a particular ad is performing well. This prior engagement data can help  predict future engagement behavior.
   
- **ad_impression**
  
   This feature records the number of times an ad is viewed. This is helpful since we can train the model on the engagement rate as a feature when we have a reasonable number of ad impressions. We can select the cut-off point until which we want to consider the impressions. For example, we can say that if a particular ad has twenty impressions, we can measure the historical engagement rate. 

- **ad_negative_engagement_rate**

  This feature keeps a record of negative engagement (hide, report) with the ad.
   
- **ad_embedding**

  We can generate embedding for an ad given all the data that we know about it e.g. ad terms and engagement data. This embedding is then a dense representation of the ad that can be used in modeling. Please refer to our [embedding lesson](https://www.educative.io/courses/grokking-the-machine-learning-interview/embeddings) to see how we can generate this embedding.

  Ad embedding can also be used to detect the ad's category, e.g., it can tell if the ad belongs to sports, etc.

Let's engineer some features for the prediction model.

Introduction

Practical ML Techniques/Concepts

Search Ranking

Feed Based System

Recommendation System

Self-Driving Car: Image Segmentation

Entity Linking System

Ad Prediction System

Feature Engineering

Features for the model

Ad specific features