Curlie

Curlie

School of Data

School of Data

Spreadsheet Transformations to Metadata

Spreadsheet Transformations to Metadata

Industry Classification

Industry Classification

Text Packages in R

Text2Vec
TopicModels
Quanteda
CoreNLP
OpenNLP
Spacyr
LDAvis

Tensorflow with R

Tensorflow with R

Common Ways to Avoid Overfitting in NN

• Get lots of sample data
• Reduce the network capacity
• Early stopping
• Batch normalization
• Add weight regularization
• Add dropout

Cleanipedia

Cleanipedia

WebAssembly

WebAssembly

auto-sklearn

auto-sklearn

NVIDIA Deep Learning GPUs

NVIDIA GTX 1060 - (lightweight users - beginners)
NVIDIA GTX 1070 - (versatile for startups)
NVIDIA GTX 1080 - (versatile for startups)
NVIDIA GTX 1080Ti - (all-round high-end)
NVIDIA Titan Xp - (heavy users e.g. computer vision/video)
AWS GPU

Relational to Semantic Mappings

R2RML
RML

Metaphacts

Metaphacts

AWS Neptune

AWS Neptune

AWS vs Azure

Azure	AWS
Storage (Blobs, Tables, Queues, Files)	S3
Virtual Machines	EC2
Autoscale	AutoScaling
Docker Virtual Machine Ext	EC2 Container Service
Blob Storage	Elastic Block Storage
HDInsight	EMR
Cloud Services Websites & Apps	Elastic Beanstalk
Backup	Glacier
Storsimple	Storage Gateway
Import export	Import / export
CDN	Cloudfront
SQL	RDS
DocumentDB	DynamoDB
Managed Cache (Redis)	Elastic Cache
Data Factory	Data Pipeline / Glue
Virtual Network	VPC
ExpressRoute	Direct Connect
Traffic Manager	Route53
Load Balancing	Elastic Load Balancing
Active Directory	IAM
Multi-Factor	Multi-Factor
Operational Insights	CloudTrail
Application Insights	CloudWatch
Event Hubs	Kinesis
Notification Hubs	SNS
Key Vault	Key Management Service
Resource Manager	Cloud Formation
API Management	API Gateway
Automation	Opsworks
Batch	SQS
Search	CloudSearch
Service Bus	SWF
Stream Analytics	Kinesis
Biztalk	SES
Machine Learning (preview)	Machine Learning
Functions	Lambda
Gets more expensive as you use more on the 'you only pay for what you use model' (especially Linux instances)	Less expensive as you use more on the 'you only pay for what you use model'   (has plenty of linux options and windows options are priced the same as azure)
Generally, slower in services	Generally, faster in services
Very confusing naming for services some of which are not even services but just reused applications in the cloud	Service names are distinguishable and clear
Azure charges by the minute	AWS charges by the hour
ML and data analytics options are not that great.  Azure only has cosmodb, hdinsight, event hub, storage, its AI services are not great either, tight coupling to azure windows services, not ideal for data science work - with azure for non-windows type work is not seamless. CosmoDB is a strong option –it supports documentdb, mongodb, property graph, cassandra, table – not perfect but ok.	Data Pipeline  Kinesis  S3  Deep Learning library support + AMIs,  Support for GPU instances  More flexible DBs, But no multi-model option, Redshift, EMR (supports spark, flink, presto, hbase, hive, pig), glue, athena, elasticsearch, cloudsearch, clustering is easier for linux machines (can you see running hadoop on windows?), AWS has neptune that supports property graphs, rdf, sparql,  tinkerpop.

azure vs aws service mappings