How I cracked the GCP Professional ML Engineer certification in 8 days!
After having successfully aced Google’s most recent certification — Google Certified professional Machine Learning Engineer, many people reached out to me on my LinkedIn recently for suggestions on how to prepare for the exam. So I thought why not put a guide of everything I did to ace this exam.
Talking about my professional experience, I have been working as a Machine Learning Architect / Engineer for more than 5 years and have created numerous ML powered enterprise applications on various platforms. So more or less working with Tensorflow / GCP stack is a part of my weekly curriculum.
Now talking about the exam specifically, GCP Professional ML Engineer exam focuses mostly on three core areas:
- Knowledge of ML concepts and related tools — Tensorflow, Keras, XGB
- Knowledge of GCP ML services — AI Platform, ML APIs, BQML
- Knowledge of MLOps and related tools — TFX, Kubeflow, Best practices
The exam has 60 questions:
- 35 pretty straight forward and span only on 1 area
- 15 questions span on 2 areas
- 10 difficult which span across all 3 or even beyond
It took me around 8 days, 4 years of GCP ML experience and 5+ years of ML experience. Hence below is a complete summary of how one should prepare for the exam. Preparation duration can span all the way from 8 days to 2+ months subjected to your expertise in the 3 core areas.
One should plan to study these topics in the order of mention:
- ML Crash course by Google — https://developers.google.com/machine-learning/crash-course/
- GCP AutoML Training — https://cloud.google.com/automl/docs
- GCP ML APIs — Natural Language API, Vision API, Audio API
- AI Platform Training — https://cloud.google.com/ai-platform/training/docs
- AI Platform Built-in algos — https://cloud.google.com/ai-platform/training/docs/algorithms
- AI Platform Prediction — https://cloud.google.com/ai-platform/prediction/docs
- AI Platform DL containers — https://cloud.google.com/ai-platform/deep-learning-containers/docs
- AI Platform explanation — https://cloud.google.com/ai-platform/prediction/docs/ai-explanations/overview
- Continuous evaluation — https://cloud.google.com/ai-platform/prediction/docs/continuous-evaluation
- TF Profiler — https://www.tensorflow.org/guide/profiler
- TF Distributed — https://www.tensorflow.org/guide/distributed_training
- TFX pipelines and components — https://www.tensorflow.org/tfx/guide/understanding_tfx_pipelines
- AI Platform pipelines — https://cloud.google.com/ai-platform/pipelines/docs
- BQML Syntaxes and types of Algos — https://cloud.google.com/bigquery-ml/docs/tutorials
- Basics of Non-ML services: Dataflow, Dataproc, PubSub, DataFusion
- Different type of ML Accuracy metrics
Some tips for the exam and special topics:
- Always remember, every ML solutioning has to go in this order: Start with GCP ML APIs to check any existing API that can be leveraged -> Else AutoML Training -> Else AI Platform Inbuilt algo -> Else AI Platform Custom Training on TF -> Else AI Platform Custom Training on Containers
- Should have knowledge of SKLearn Pipelines, Keras sequential models
- JSON Syntax for AI Platform prediction API
- Comparison of Edge model Vs general model Vs High accuracy
- Attribution techniques for image model explanation
- Containerized Training / Containerized Prediction
- Custom prediction routines on AI platform
- TF Dataset optimization techniques
- Read about different parameters in Tensorflow model serving
- Study about streaming data systems design with DataFlow & PubSub
- Should know the difference between CNNs and RNNs
- Difference between Precision and Recall for binary classification
This information is more than sufficient to crack the certification examination, I would suggest at-least 2 revisions of the whole content. First with an intent to rememorize and second with the intent to speedup.
All the best for the certification !