Python is for the birds and for the brains

PyData Atlanta presentation. August 17th, 2016.

Transcript

1. Python is for the birds and for the brains David Nicholson, graduate student
2. Outline • Neuroscience • Python • Bird(song)s • Machine Learning • More Neuroscience and Python
3. Neuroscience most complex object in the universe
4. Neuroscience most complex object in the universe 2nd most complex object in the universe
5. Neuroscience • multidisciplinary
6. Neuroscience • multidisciplinary – biology • anatomy • physiology • molecular biology • cell biology • evolution
7. Why Neuroscience Needs Python the two Rs of scientific python, and a couple of corollaries • Readability – ⇒ “writability”
8. Why Neuroscience Needs Python the two Rs of scientific python, and a couple of corollaries • Readability – ⇒ “writability” • Reproducibility (of analysis) – Jupyter notebook / JupyterLab – ⇒ A level playing field
9. Why Neuroscience Needs Python the two Rs of scientific python, and a couple of corollaries • Readability – ⇒ “writability” • Reproducibility – Jupyter notebook / JupyterLab – ⇒ A level playing field • Oh, and also: – Employability • One of top 5 languages
10. Songbirds: a model system … for neuroscience • learn song – socially, from a tutor picture: Jon Sakata 2016 spectrogram: http://doolinglab.umd.edu/zebra_finch.htm
11. Songbirds: a model system … for neuroscience • learn song – socially, from a tutor – during a critical period in development picture: Jon Sakata 2016 spectrogram: http://doolinglab.umd.edu/zebra_finch.htm
12. Songbirds: a model system … for neuroscience • learn song – socially, from a tutor – during a critical period in development – practice and improve using sensorimotor learning picture: Jon Sakata 2016 spectrogram: http://doolinglab.umd.edu/zebra_finch.htm
13. Birdsong consists of elements called syllables from: https://youtu.be/v4ulzgJB_Mg
14. Songbirds • maintain song in adulthood using sensory feedback Sober Brainard Nat. Neuro. 2009
15. Songbirds • maintain song in adulthood using sensory feedback Sober Brainard Nat. Neuro. 2009
16. Songbirds • have evolved a network of brain areas for learning and producing song, the song system Bolhuis et al. PNAS 2000;97:2282-2285
17. • but the song system sits within brain regions that birds share with humans, regions conserved across evolution – e.g. the basal ganglia
18. • but the song system sits within brain regions that birds share with humans, regions conserved across evolution – e.g. the basal ganglia
19. Songbirds • provide a model system where we can record electrical activity in the brain activity and relate that to muscle activity and behavior Samuel J. Sober et al. J. Neurosci. 2008;28:10370-10379
20. Songbirds • a “model system” of neuroscience • because multidisciplinary – what about the fields I left out? Let’s look at an example where Python helps: • bridge the gaps between fields • make it easier to produce readable, reproducible research
21. Netflix and label • a.k.a. the joys of data collection
22. Birdsong and machine learning • Birds typically sing hundreds of songs a day, many more than can be studied by hand
23. Birdsong and machine learning • Birds typically sing hundreds of songs a day, many more than can be studied by hand • Previous papers applied machine learning algorithms to classifying syllables
24. Birdsong and machine learning • Birds typically sing hundreds of songs a day, many more than can be studied by hand • Previous papers applied machine learning algorithms to classifying syllables • No study has compared different algorithms
25. machine learning (and birdsong) • Supervised learning – algorithms that produce classifiers • trained with an n-element training set of feature vectors – {yi , xi }, i = 1,2,3,…,n where » y is the correct class / label » x is a vector of m features – x = {x1 , x2 , x3 , … xm } – e.g., for syllables » {amplitude, pitch, duration,…}
26. Birdsong and machine learning • Two previously published papers applying machine learning to Bengalese Finch song – k nearest neighbors (k-NN)
27. Birdsong and machine learning • Two previously published papers applying machine learning to Bengalese Finch song – support vector machine (SVM)
28. Birdsong and machine learning • methods: – generate large hand-labeled datasets for four birds
29. Birdsong and machine learning • methods: – generate large hand-labeled datasets for four birds – extract features from syllables used to train algorithms
30. Birdsong and machine learning • methods: – generate large hand-labeled datasets for four birds – extract features from syllables used to train algorithms – use sci-kit learn to facilitate training models, measuring accuracy
31. Birdsong and machine learning • methods: – generate large hand-labeled datasets for four birds – extract features from syllables used to train algorithms – use sci-kit learn to facilitate training models, measuring accuracy • measure accuracy with standard 5-fold cross validation
32. Birdsong and machine learning • results: – to the Jupyter notebook  – https://github.com/NickleDave/ML-comparison- birdsong/
33. Python in Neuroscience, part II • Frontiers in Neuroinformatics (open access): http://dx.doi.org/10.3389/fninf.2015.00011 – bioinformatics • protein networks in Alzheimer’s – neuroimaging – stimulus generation for: • psychophysics • electrophysiology – modeling / simulation • NEURON,NEST,PCSIM,Brian,MOOSE,NENGO
34. Python in Neuroscience, part II • Binder ZOMG SO COOL – http://mybinder.org/ – http://mybinder.org/repo/sofroniewn/tactile- coding
35. Conclusion: Why (Neuro)science Needs Python, Again • Readability • Reproducibility • Employability