Predicting with GCP - Speaker Deck

Slide 1

Slide 1 text

@Giuliabianchl Predicting with Google Cloud Platform April 24, 2019 - Giulia Bianchi

Slide 2

Slide 2 text

@Giuliabianchl @Giuliabianchl giulbia Data scientist > DXD-WiMLDS <

Slide 3

Slide 3 text

@Giuliabianchl End of 2017 Google AutoML announcement GCP podcast 117

Slide 4

Slide 4 text

@Giuliabianchl End of 2018 Google AI Hub announcement

Slide 5

Slide 5 text

@Giuliabianchl @Giuliabianchl To access any of them you need to have access to GCP anyway...

Slide 6

Slide 6 text

@Giuliabianchl ● Set up a GCP account (gmail ID needed) ● Sign in Google Cloud Console and set up a project ($$$) ● Install Cloud SDK Google Cloud Platform

Slide 7

Slide 7 text

@Giuliabianchl Did you say Cloud?

Slide 8

Slide 8 text

@Giuliabianchl ● Cloud → managed services ● From "on-premises" to *aaS ○ IaaS ○ PaaS ○ SaaS ● Access to theoretically unlimited resources ● Rapidity of provisioning ● Reliability Cloud what ?! And why should I even care?

Slide 9

Slide 9 text

@Giuliabianchl GCP AI services - non exhaustive! SaaS SaaS SaaS SaaS PaaS SaaS

Slide 10

Slide 10 text

@Giuliabianchl @Giuliabianchl Cloud ML Engine… now known as AI Platform

Slide 11

Slide 11 text

@Giuliabianchl AI Platform Doc

Slide 12

Slide 12 text

@Giuliabianchl Parenting 2.0 giulbia/baby_cry_detection https://www.youtube.com/watch?v=N-LXrheCIKM

Slide 13

Slide 13 text

@Giuliabianchl Training pipeline Input training data Feature engineering Train model SVM Model saved in laptop

Slide 14

Slide 14 text

@Giuliabianchl Prediction pipeline Preliminary step Model deployed in RPi

Slide 15

Slide 15 text

@Giuliabianchl Prediction pipeline Input new data Feature engineering Predict! Model deployed in RPi Prediction saved in RPi

Slide 16

Slide 16 text

@Giuliabianchl Prediction pipeline Input new data Feature engineering Predict! Model deployed in RPi Prediction saved in RPi Done in 45 sec.

Slide 17

Slide 17 text

@Giuliabianchl Parenting 3.0 giulbia/gcp-rpi giulbia/baby_cry_rpi

Slide 18

Slide 18 text

@Giuliabianchl Prediction pipeline Preliminary step Model deployed in GCP

Slide 19

Slide 19 text

@Giuliabianchl Prediction pipeline Input new data Feature engineering Predict! Model deployed in GCP Prediction saved in GCP

Slide 20

Slide 20 text

@Giuliabianchl Prediction pipeline Input new data Feature engineering Predict! Model deployed in GCP Prediction saved in GCP Done in 5 sec.

Slide 21

Slide 21 text

@Giuliabianchl @Giuliabianchl Some code

Slide 22

Slide 22 text

@Giuliabianchl Cloud Function - main.py

Slide 23

Slide 23 text

@Giuliabianchl Cloud Function - requirements.txt

Slide 24

Slide 24 text

@Giuliabianchl Cloud Function - function deployment

Slide 25

Slide 25 text

@Giuliabianchl Cloud Function - call to ML engine for prediction

Slide 26

Slide 26 text

@Giuliabianchl Cloud Function - logs

Slide 27

Slide 27 text

@Giuliabianchl @Giuliabianchl Take away

Slide 28

Slide 28 text

@Giuliabianchl ... has a huge potential for data scientists ... is not fully data scientist friendly yet ... it needs more documentation ... evolves super fast Google Cloud Platform...