Google Cloud Platform Certificate; Professional Machine Learning Engineer

 

Architecture for MLOps using TFX, Kubeflow Pipelines, and Cloud Build

Link: https://cloud.google.com/architecture/architecture-for-mlops-using-tfx-kubeflow-pipelines-and-cloud-build

TFX, Kubeflow 파이프라인, Cloud Build를 사용하는 MLOps 아키텍처  |  클라우드 아키텍처 센터  |  Googl

- new data: only CT pipeline needed

- new pipeline or new implementation: CI/CD pipeline + CT pipeline needed

 

 

 

 

Data preprocessing for machine learning: options and recommendations

https://cloud.google.com/architecture/data-preprocessing-for-ml-with-tf-transform-pt1#where_to_do_preprocessing

머신러닝을 위한 데이터 사전 처리: 옵션 및 권장사항  |  클라우드 아키텍처 센터  |  Google Clou

* Full-pass transformation: use only the training data to compute statistics

 

 

 

 

 

De-identification and re-identification of PII in large-scale datasets using Cloud DLP

https://cloud.google.com/architecture/de-identification-re-identification-pii-using-cloud-dlp

Cloud DLP를 사용하여 대규모 데이터 세트에서 PII 익명화 및 재식별  |  클라우드 아키텍처 센터  

 

 

 

Machine Learning Crash Course

https://developers.google.com/machine-learning/crash-course/ml-intro

Introduction to Machine Learning  |  Machine Learning Crash Course  |  Google Developers

* Precision & Recall: Trade-off

- if decrease threshold, # TP & # FP increase -> Recall increase = Precision decrease

 

 

 

GCP Products

- best data pipeline, for stream and batch processing, necessary for stream

- SQL syntax, support ML models and importing TF models to the platform

- storage + analysis (fully managed)

- implement functions at this platform (new data -> pub/sub + cloud function)

- handles workflow for kubeflow pipeline

- use it to migrate from on-premises to GCP

- IaaS; Virtual machines

- modeling, training, analyzing, evaluation ... solution for managing ML models

- PaaS; platform that executes your code

- scheduler for task automation

- fully managed database for MYSQL

- fully managed database for large scale & low-latence workloads

- workflow orchestration service built on Apace Airflow

- data ingestion

- integrate data from multiple cloud sources

- Apache Spark & Hadoop clusters

 

 

 

 

Dumps

https://www.examtopics.com/exams/google/professional-machine-learning-engineer/

Google Professional Machine Learning Engineer Free Certification Exam Material | ExamTopics

 

tf.distribute.MirroredStrategy supports synchronous distributed training on multiple GPUs on one machine.

tf.distribute.TPUStrategy lets you run your TensorFlow training on Tensor Processing Units (TPUs).

tf.distribute.MultiWorkerMirroredStrategy is very similar to MirroredStrategy. It implements synchronous distributed training across multiple workers, each with potentially multiple GPUs.

tf.distribute.OneDeviceStrategy is a strategy to place all variables and computation on a single specified device.