|
|
本帖最后由 WST 于 2021-1-1 14:28 编辑
.google и
Machine learning engineer 面试总结
- Leetcode 敲门砖。怎么连?就是干!200-300 + 练到看到题就能做
- Machine learning design
有些reference就是个论文,写的很复杂。但是重要的是理解问题的背景,然后搞一套自己的套路。就我自己来说就是:
1. Data collection. You have to understand the problem and figure out possible data you can collect.
2. Feature engineering: categorical feature, numerical feature, text feature (TF-IDF, word2vec), image (Use pre-trained CNN model like VGG as feature extractor)
3. Dimension reduction (optional): SVD based ALS (useful for TF-IDF), PCA
4. Modeling. What I do is to convert a recommendation system problem to one of the following models: ..
1. A simple distance metric (for example, number of shared friends in friend recommendation)
2. Logistic regression (Best and simplest one)
3. Matrix factorization (Netflix)
4. Neutral network using Triplet loss function (If you have a lot of text and/or image)
5. Model evaluation
1. F1 score. Waral dи,
2. AUC
3. A/B testing if interviewer ask for business evaluation ..
. 1point3acres.com
- System design
https://www.1point3acres.com/bbs ... =machine%2Blearning
. 1point3acres
机器学习`侠`练成记录 Becoming a Machine Learning Practitioner
- 比较全面,包括学习资料推荐
https://www.1point3acres.com/bbs ... =machine%2Blearning
|
|