SlideShare a Scribd company logo
Clova
Keigo Hattori, Clova Center
@keigohtr
Keigo Hattori
keigohtr
Who I am
Keigo Hattori
Software Engineer
2009
2009~2017
2017 ~ LINE
x
• Clova
•
Agenda
1. Clova
2. Clova ML
( )
3. LINE
(We are hiring. Welcome ML engineer!)
Clova
Clova
Clovaを支える技術 機械学習配信基盤のご紹介
Unique Feature
• LINE
•
•
• (WAVE )
• Clova
•
Agenda
Clova
SVM
Deep LearningMachine Learning
Keras
Chainer
Caffe
scikit-learn
gensim
logistic regression
Random Forest
Neural Network
Perceptron
libsvm
liblinear
Theano
JupyterLab
TensorBoard ChainerUI
Clovaを支える技術 機械学習配信基盤のご紹介
•
→
•
→
•
→
•
→
•
→
Clova
Drucker ( )
Three features of Drucker
• Easy to serve
• Easy to management
• Easy to integrate
Drucker: Easy to serve
Write your algorithm on Drucker
template
Features
• Do not need to write proto
(grpc proto)
• Input/Output format is
determined but flexible
• Any algorithms are availble
Drucker: Easy to management
Use Drucker web interface
Features
• Upload the latest model
• Switch the models
• Evaluate the performance
Drucker: Easy to integrate
Integrate in the same manner
Features
• Do not need to write proto
(grpc proto)
• Input/Output format is
determined but flexible
Clovaを支える技術 機械学習配信基盤のご紹介
Drucker on Kubernetes
Kubernetes
• Rolling update
• Alive monitoring
Cloud Storage for ML model
• Auto replication
• Sync files on all servers
Cloud DB for model name
• Save model filename for
each environment (dev/prod)
@keigohtr
Keigo Hattori
keigohtr
Keigo Hattori
@line_clova #Clova
LINE Clova
http://clova-blog.line.me/ja/
THANK YOU

More Related Content

Clovaを支える技術 機械学習配信基盤のご紹介