DEV Community

hiroyone
hiroyone

Posted on • Updated on

How I cracked the Google Cloud Professional Machine Learning Engineer Certification Exam

Image description

TL;DR

  • Google Cloud launched the Professional Machine Learning(ML) Engineer Certification Exam on Oct 15, 2020🎉 (Not beta but GA version)
  • I passed the exam as one of the first test-takers on Oct. 17 in Tokyo, Japan.💪
  • Many exam questions were centered around practical use cases, not just about GCP service understanding.
  • Also, many test questions were beyond the scope of GCP services; Pure ML problems were asked.
  • To cover the exam, it is good to have a robust & comprehensive understanding of ML problems and solutions: ML foundations, major ML models, feature engineering, Tensorflow, model understanding tools, GCP services, Deep Learning technology.
  • Knowledge gained from the Professional Data Engineer Certification is partially but definitely helpful to this ML exam.
  • Compared with the Professional Cloud Architect and  the Professional Data Engineer Certifications I passed previously, I am certain that this exam is way more difficult, practical, and advanced, requiring both GCP service familiarity and ML domain expertise.

A list of recommended articles concerning this exam

Motivation

Google Cloud Professional Machine Learning Engineer Certification Exam was launched on Oct 15, 2020🎉 (Not beta but GA version)

I passed the exam as one of the first test-takers on Oct. 17 in Tokyo, Japan. I wanted to be the first Japanese who passed this unprecedentedly hard exam, requiring all GCP, ML, and programming skills (let alone English proficiency).

About me

I already passed two other GCP exams in the past: Professional Cloud Architect and Professional Data Engineer Certifications, whose expertise was definitely helpful to this ML exam.

Also, I've already had some preliminary knowledge and experience in machine learning and programming.

  • Worked as an ML engineer in the Japanese IT industry for the past few years.
  • Won second place in a mid-size public ML competition in 2018.
  • Completed Python fundamentals, Pandas, Numpy, NLP, Matplotlib, and others on-line in six months before starting my career.

Roadmap

  • Preliminary Step: Data Science and ML fundamentals in Python (100hrs)🤖
  • Tensorflow (20hrs)
  • Input Data Pipeline, Models Trainings, Evaluation Metrics, Model Analysis Tool📊, CPU/GPU/TPU
  • GCP Solutions (100hrs)
  • ML Models: pre-trained AI solutions → AutoML → Custom Model → Kubeflow Pipeline
  • ML Preps: AI Notebooks, Data Labeling Service, Data Loss Prevention
  • Big Data: BigQuery, Cloud Storage
  • ETL: Dataflow, Composer, Datafusion, Pubsub and so many others😊 Feature Engineering (20hrs)
  • DNN/CNN/RNN, Wide and Deep Model, Autoencoder
  • Overfitting/Underfitting: L1&L2 Regularizations, Dropouts
  • Fast Computation: Stochastic Gradient Descent, Mini-batch
  • Fast Gradient Descent: Adam (Momentum & RMS Props & learning rate decay)
  • Vanishing Gradient: ReLu
  • Exploding Gradient: Lower learning rate
  • Internal Covariate Shift: batch normalization
  • Dead Relu: Leaky Relu

Conclusion

As described above, there are a wide variety of ML subjects to cover in addition to GCP ML services, and I found that this exam was still hard to solve. That is why I felt that this exam is way harder than other GCP certification exams.
According to the Google Cloud Certified directory, there are 156 registered people who passed the ML exam in the world, three months after its launch.

Image description

I am happy and proud of myself for passing this tough exam, but I may have just lucked out.😂

Good luck to you too!👼

Discussion (0)