Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
PyCon India - Commodity Machine Learning; past, present and future
Search
Andreas Mueller
September 25, 2016
0
2.6k
PyCon India - Commodity Machine Learning; past, present and future
PyCon India 2016 keynote
Andreas Mueller
September 25, 2016
Tweet
Share
More Decks by Andreas Mueller
See All by Andreas Mueller
Automating Machine Learning
amueller
4
1.1k
Engineering Scikit-Learn V2
amueller
0
180
Advanced Machine Learning with Scikit-Learn for Pycon Amsterdam
amueller
0
240
Scikit-learn: New project features in 0.17
amueller
0
63
Bootstrapping machine learning
amueller
0
96
PyData Berlin 2014 Keynote: Commodity machine learnin
amueller
0
80
Advanced Machine Learning with Scikit-Learn
amueller
1
420
Machine Learning With Scikit-Learn ODSC SF 2015
amueller
4
1.4k
Machine Learning With Scikit-Learn - Pydata Strata NYC 2015
amueller
1
2.9k
Featured
See All Featured
How to train your dragon (web standard)
notwaldorf
71
5.1k
Bootstrapping a Software Product
garrettdimon
PRO
302
110k
Designing Dashboards & Data Visualisations in Web Apps
destraynor
225
51k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
34
8.8k
Design by the Numbers
sachag
274
18k
Optimising Largest Contentful Paint
csswizardry
7
2.3k
I Don’t Have Time: Getting Over the Fear to Launch Your Podcast
jcasabona
19
1.6k
Git: the NoSQL Database
bkeepers
PRO
421
63k
How to Ace a Technical Interview
jacobian
272
22k
CoffeeScript is Beautiful & I Never Want to Write Plain JavaScript Again
sstephenson
153
14k
Building a Modern Day E-commerce SEO Strategy
aleyda
15
6.3k
Designing with Data
zakiwarfel
94
4.8k
Transcript
Commodity Machine Learning Past, present and future Andreas Mueller
What is machine learning?
Automatic Decision Making Spam? Yes No
Spam? Yes No
Programming Machine Learning
Machine learning is EVERYWHERE
None
None
None
Science Engineering Medicine ...
Commodity machine learning
past
+
None
dawn of open source tools...
The age of shell
Documentation? Testing?
Scikit-learn: User centric machine learning
.fit(X, y) .predict(X) .transform(X)
present
Choose your ecosystem.
Open! Documented! Tested!
Usability is key!
ML Frameworks PyMC, Edward, Stan theano, tensorflow, keras
None
from sklearn.model_selection import GridSearchCV from sklearn.pipeline import Pipeline
github.com/scikitlearncontrib/scikitlearncontrib
(near) Future
pip install scikitlearn==0.18rc2 0.18 for the release candidate:
sklearn.cross_validation sklearn.grid_search sklearn.learning_curve sklearn.model_selection
results = pd.DataFrame(grid_search.results_)
labels → groups n_folds → n_splits
from sklearn.cross_validation import KFold cv = KFold(n_samples, n_folds) for train,
test in cv: ... from sklearn.model_selection import KFold cv = KFold(n_folds) for train, test in cv.split(X, y): ...
from sklearn.mixture import GaussianMixture from sklearn.mixture import BayesianGaussianMixture
PCA() RandomizedPCA() PCA()
Gaussian Process Rewrite
Isolation Forests
Play from sklearn.neural_network import MLPClassifier Work import keras
pipe = Pipeline([('preprocessing', StandardScaler()), ('classifier', SVC())]) param_grid = {'preprocessing': [StandardScaler(),
None]} grid = GridSearchCV(pipe, param_grid)
40
(further) Future
Feature / Column names
from __future__ import sklearn.plotting
from __future__ import AutoClassifier
More Transparency
amueller.github.io @amuellerml @amueller
[email protected]