2016/11/07

Devoxx 2016 - day 1: notes (2016/11/07)

Deep Learning: An Introduction

Breandan Considine


examples
reasons
  • big data
  • hardware (nvidia cuda)
  • algorithms
machine learning fundamentals
  • tensors: n-dimensional array
  • learning types:
    • supervised
    • unsupervised
    • reinforced
tensorflow examples
  • linear regression: single line in data
  • classification
    • "perceptron"
    • layers + weighs "gradient descent"
unsupervised learning
  • clustering, separation, association
  • clustering: random points, euclidean distance
data preprocessing
  • feature scalin,g normalization
  • decomposition & aggregation
  • dimensionality reduction
  • --> training set, validate & select best modelf
DeepLearning4J
  • builder pattern
  • components:
    • nd4j -- n-dimensional arrays reinforcement learning
  • agent has context + choices
  • rewards
  • goal: maximize cumulative reward
refs:

Make CSS Fun Again with Flexbox!

Hubert Sablonnière


refs:
flex:
  • float styles are only for text flow
  • display: flex: // flex: 1 ( for 1 row)
    • parent / flexcontainer
    • children: elements
  • flex-grow: weight
  • flex-shrink: minimum
  • flex-basis: default
others
  • justify-content: center / flex-end
  • align-items: center

Easily secure your Front and back applications with KeyCloak

Sébastien Blanc


http://www.keycloak.org/
Open Source Identity and Access Management
  • jwt (rfc 7519)
  • openid, kerberos etc
  • adapters (wildfly, spring boot, node.js ...)
  • native clients
  • login brokers
  • otp
demo
  • web.xml: roles/ security config
  • keycloak.json config
  • atom editor https://atom.io/
  • node.js: add keycloak.protect()

Sentiment analysis of social media posts using Apache Spark

Niels Dommerholt


http://spark.apache.org/ dataprocessing engine

sentiment analysis
  • positive / negative
  • java 8 / streams api
  • coursera
demo
  • JavaSparkContext: config

Apache Spark? If only it worked

Marcin Szymaniuk


origin: http://blog.explainmydata.com/2014/05/spark-should-be-better-than-mapreduce.html*

details

  • RDD: Resilient Distributed Dataset :
    • cache
    • no priority
  • sizing executors: configure memory (should autobalance in recent versions)
  • known pitfalse: 2g block limit, gc's -- check level of parallelism (groupByKey, repartition)
  • check locality: NODE_LOCAL -- increase exectors if needed
  • broadcast variable
  • avoid groupbykey -- use reducebykey
  • debugging:
    • log aggregation
    • hdfs monitoring logging
    • gclogs  

No comments: