THE MISSING MANUAL FOR DATA SCIENCE: REMIX. RESUSE. REPRODUCE from Structure:Data 2013
1 like1,147 views
The document details a presentation by Dr. Matt Wood on the principles of reproducibility in data science, emphasizing the importance of data collection, sharing, and collaboration. It outlines key principles for improving reproducibility, such as the gravity of data, ease of use, the necessity of reuse, collaborative building, and the significance of provenance. Dr. Wood illustrates these concepts with examples and suggests practical steps for overcoming challenges associated with data science workflows.
THE MISSING MANUAL FOR DATA SCIENCE: REMIX. RESUSE. REPRODUCE from Structure:Data 2013
- 1. THE MISSING MANUAL FOR DATA SCIENCE: REMIX. RESUSE. REPRODUCE SPEAKER: Matt Wood Principal Data Scientist Amazon Web Services Monday, April 1, 13
- 2. The Missing Manual: Reproduce, Reuse, Remix Dr. Matt Wood matthew@amazon.com @mza Monday, April 1, 13
- 3. Monday, April 1, 13
- 4. Hello. Monday, April 1, 13
- 5. Monday, April 1, 13
- 6. Data. Monday, April 1, 13
- 7. Generation Collection & storage Analytics & computation Collaboration & sharing Monday, April 1, 13
- 8. Monday, April 1, 13
- 9. Generation challenge. Monday, April 1, 13
- 10. Amazing data generators: cell phones tracking cholera in Haiti Linus Bengtsson et al. PLoS Medicine, 2011 Monday, April 1, 13
- 11. Amazing data generators: social networks tracking influenza You Are What You Tweet: Analyzing Twitter for Public Health. M. J. Paul and M. Dredze, 2011 Monday, April 1, 13
- 12. Amazing data generators: web app logs targeting advertising 500% return on ad spend Monday, April 1, 13
- 13. Monday, April 1, 13
- 14. Monday, April 1, 13
- 15. Chromosome 11 : ACTN3 : rs1815739 Monday, April 1, 13
- 16. Chromosome X : rs6625163 Monday, April 1, 13
- 17. Chromosome 19 : FUT2 : rs601338 Monday, April 1, 13
- 18. Chromosome 2 : rs10427255 Monday, April 1, 13
- 19. Chromosome 10 : rs7903146 TYPE II Monday, April 1, 13
- 20. Chromosome 15 : rs2472297 +0.25 Monday, April 1, 13
- 21. Monday, April 1, 13
- 22. Generation challenge. Monday, April 1, 13
- 23. Generation challenge. X Monday, April 1, 13
- 24. Generation Collection & storage Analytics & computation Collaboration & sharing Monday, April 1, 13
- 25. Generation Collection & storage Analytics & computation Collaboration & sharing Monday, April 1, 13
- 26. Monday, April 1, 13
- 27. Utility computing. Monday, April 1, 13
- 28. Monday, April 1, 13
- 29. Monday, April 1, 13
- 30. Monday, April 1, 13
- 31. Remove constraints. Monday, April 1, 13
- 32. Monday, April 1, 13
- 33. Analytics challenge. Monday, April 1, 13
- 34. Analytics challenge. X Monday, April 1, 13
- 35. Generation Collection & storage Analytics & computation Collaboration & sharing Monday, April 1, 13
- 36. Monday, April 1, 13
- 37. Beautiful, unique. Monday, April 1, 13
- 38. Monday, April 1, 13
- 39. Impossible to recreate. Monday, April 1, 13
- 40. Monday, April 1, 13
- 41. Snowflake Data Science Monday, April 1, 13
- 42. Monday, April 1, 13
- 44. Monday, April 1, 13
- 45. Reproducibility scales data science. Monday, April 1, 13
- 46. Monday, April 1, 13
- 47. Reproduce. Reuse. Remix. Monday, April 1, 13
- 48. Monday, April 1, 13
- 49. Value++ Monday, April 1, 13
- 50. Monday, April 1, 13
- 51. Monday, April 1, 13
- 52. How do we get from here to there? IPLESF 5 PR INC O REPRO DUCIBILITY Monday, April 1, 13
- 53. PRINCIPLESF 5 O REPRODUCIBILITY Monday, April 1, 13
- 54. PRINCIPLESF 5 O REPRODUCIBILITY 1. Data has Gravity Monday, April 1, 13
- 55. Monday, April 1, 13
- 56. Increasingly large data collections. Monday, April 1, 13
- 57. Monday, April 1, 13
- 58. Challenging to obtain and manage. Monday, April 1, 13
- 59. Monday, April 1, 13
- 60. Expensive to experiment. Monday, April 1, 13
- 61. Monday, April 1, 13
- 62. Large barrier to reproducibility. Monday, April 1, 13
- 63. Monday, April 1, 13
- 64. Move data to the users. Monday, April 1, 13
- 65. Move data to the users. X Monday, April 1, 13
- 66. Monday, April 1, 13
- 67. Move tools to the data. Monday, April 1, 13
- 68. Monday, April 1, 13
- 69. Place data where it can be consumed by tools. Monday, April 1, 13
- 70. Monday, April 1, 13
- 71. Place tools where they can access data. Monday, April 1, 13
- 72. Monday, April 1, 13
- 73. Monday, April 1, 13
- 74. Monday, April 1, 13
- 75. Monday, April 1, 13
- 76. Monday, April 1, 13
- 77. More data, more users, more uses, more locations Monday, April 1, 13
- 78. Monday, April 1, 13
- 79. Cost Monday, April 1, 13
- 80. Monday, April 1, 13
- 81. Force multiplier. Monday, April 1, 13
- 82. Monday, April 1, 13
- 83. Cost and complexity kill reproducibility. Monday, April 1, 13
- 84. PRINCIPLESF 5 O REPRODUCIBILITY Monday, April 1, 13
- 85. PRINCIPLESF 5 O REPRODUCIBILITY 2. Ease of use is a prerequisite Monday, April 1, 13
- 87. Monday, April 1, 13
- 88. Help overcome the suck threshold. Monday, April 1, 13
- 89. Monday, April 1, 13
- 90. Easy to embrace and extend. Monday, April 1, 13
- 91. Monday, April 1, 13
- 92. Choose the right abstraction for the user. Monday, April 1, 13
- 93. Monday, April 1, 13
- 94. $ ec2-run-instances Monday, April 1, 13
- 95. Monday, April 1, 13
- 96. $ starcluster start Monday, April 1, 13
- 97. Monday, April 1, 13
- 98. Monday, April 1, 13
- 99. Package and automate. Monday, April 1, 13
- 100. Monday, April 1, 13
- 102. Monday, April 1, 13
- 103. Monday, April 1, 13
- 104. 1000 Genomes Project Cloud BioLinux Monday, April 1, 13
- 105. Monday, April 1, 13
- 106. 1000 Genomes Project + your genomic data Illumina Basespace Monday, April 1, 13
- 107. Cassandra Aegisthus Hadoop, Hive, Pig Amazon S3 Legacy data warehousing http://www.youtube.com/watch?v=oGcZ7WVx6EI Monday, April 1, 13
- 108. Sting Microstrategy R Cassandra Aegisthus Hadoop, Hive, Pig Amazon S3 Legacy data warehousing http://www.youtube.com/watch?v=oGcZ7WVx6EI Monday, April 1, 13
- 109. Monday, April 1, 13
- 110. PRINCIPLESF 5 O REPRODUCIBILITY Monday, April 1, 13
- 111. PRINCIPLESF 5 O REPRODUCIBILITY 3. Reuse is as important as reproduction Monday, April 1, 13
- 112. Seven Deadly sins of Bioinformatics: http://www.slideshare.net/dullhunk/the-seven-deadly-sins-of-bioinformatics Monday, April 1, 13
- 113. Seven Deadly sins of Bioinformatics: http://www.slideshare.net/dullhunk/the-seven-deadly-sins-of-bioinformatics Monday, April 1, 13
- 114. Monday, April 1, 13
- 115. Data scientists are hackers. Monday, April 1, 13
- 116. Monday, April 1, 13
- 117. They have their own way of working. Monday, April 1, 13
- 118. Monday, April 1, 13
- 119. Beware the Big Red Button. Monday, April 1, 13
- 120. Monday, April 1, 13
- 121. Fire and forget reproduction is a good first step, but limits longer term value. Monday, April 1, 13
- 122. Monday, April 1, 13
- 123. Monolithic, one-stop-shop. Monday, April 1, 13
- 124. Monday, April 1, 13
- 125. Work well for intended purpose. Monday, April 1, 13
- 126. Monday, April 1, 13
- 127. Challenging to install, dependency heavy. Monday, April 1, 13
- 128. Monday, April 1, 13
- 129. Difficult to grok. Monday, April 1, 13
- 130. Monday, April 1, 13
- 131. Data scientists are hackers: embrace it. Monday, April 1, 13
- 132. Monday, April 1, 13
- 133. Small things. Loosely coupled. Monday, April 1, 13
- 134. Monday, April 1, 13
- 135. Easier to grok, reuse and integrate. Monday, April 1, 13
- 136. Monday, April 1, 13
- 137. Lower barrier to entry. Monday, April 1, 13
- 138. PRINCIPLESF 5 O REPRODUCIBILITY Monday, April 1, 13
- 139. PRINCIPLESF 5 O REPRODUCIBILITY 4. Build for collaboration Monday, April 1, 13
- 140. Monday, April 1, 13
- 141. Workflows are memes. Monday, April 1, 13
- 142. Monday, April 1, 13
- 143. Reproduction is just the first step. Monday, April 1, 13
- 144. Monday, April 1, 13
- 145. Bill of materials: code, data, configuration, infrastructure. Monday, April 1, 13
- 146. Monday, April 1, 13
- 147. Full definition for reproduction. Monday, April 1, 13
- 148. Monday, April 1, 13
- 149. Utility computing provides a playground for data science. Monday, April 1, 13
- 150. Code + AMI + custom datasets + public datasets + databases + compute + result data Monday, April 1, 13
- 151. Code + AMI + custom datasets + public datasets + databases + compute + result data Monday, April 1, 13
- 152. Code + AMI + custom datasets + public datasets + databases + compute + result data Monday, April 1, 13
- 153. Code + AMI + custom datasets + public datasets + databases + compute + result data Monday, April 1, 13
- 154. PRINCIPLESF 5 O REPRODUCIBILITY Monday, April 1, 13
- 155. PRINCIPLESF 5 O REPRODUCIBILITY 5. Provenance is a first class object Monday, April 1, 13
- 156. Monday, April 1, 13
- 157. Versioning becomes really important. Monday, April 1, 13
- 158. Monday, April 1, 13
- 159. Especially in an active community. Monday, April 1, 13
- 160. Monday, April 1, 13
- 161. Doubly so with loosely coupled tools. Monday, April 1, 13
- 162. Monday, April 1, 13
- 163. Provenance metadata is a first class entity. Monday, April 1, 13
- 164. Monday, April 1, 13
- 165. Distributed provenance. Monday, April 1, 13
- 166. IPLESF 5 PRI NC O Y RODUCIBILIT REP Monday, April 1, 13
- 167. IPLESF 5 PRI NC O Y RODUCIBILIT REP 1. Data has gravity 2. Ease of use is a prerequisite 3. Reuse is as important as reproduction 4. Build for collaboration 5. Provenance is a first class object Monday, April 1, 13
- 168. Monday, April 1, 13
- 169. Thank you matthew@amazon.com aws.amazon.com @mza Monday, April 1, 13
- 170. Monday, April 1, 13