PyData Barcelona Keynote

© 2016 Continuum Analytics - Confidential & Proprietary© 2016 Continuum Analytics - Confidential & Proprietary
Data Science and Machine
Learning with Anaconda
PyData Barcelona
May 20, 2017
Travis E. Oliphant
Co-founder, President, Chief Data Scientist

© 2016 Continuum Analytics - Confidential & Proprietary
A bit about me
• PhD 2001 from Mayo Clinic in Biomedical
Engineering (Ultrasound and MRI)
• MS/BS degrees in Elec. Comp. Engineering
• Creator and Developer of SciPy (1999-2009)
• Professor at BYU (2001-2007)
• Author of NumPy (2005-2012)
• Started Numba (2012)
• Founding Chair of NumFOCUS / PyData
• Previous Python Software Foundation Director
• Co-founder of Continuum Analytics
• CEO => President, Chief Data Scientist
SciPy

Company
2012 - Created Two Orgs for Sustainable Open Source
Community
Enterprise software, support and
services to empower people who
change the world to get rapid insight
from all of their data — built on open-
source that we contribute to and
sustain.

NumFOCUS is a 501(c)(3) nonproﬁt that supports and promotes world-
class, innovative, open source scientiﬁc computing.
Community-led non-profit organization
Independent Board of Directors:
• Andy Terrel (President)
• Ralf Gommers (Secretary)
• Didrik Pinte (Treasurer)
• Lorena Barba
• Matthew Turk
• Jennifer Klay
https://www.numfocus.org/community/donate/

Data Science is growing with Python leading the way
probably the biggest reason is
Machine Learning!

© 2016 Continuum Analytics - Confidential & Proprietary 6
Neural network with
several layers trained
with ~130,000 images.
Matched trained
dermatologists with 91%
area under sensitivity-
specificity curve.
Keys:
• Access to Data
• Access to Software
• Access to Compute

https://blogs.dropbox.com/tech/2017/04/creating-a-modern-ocr-pipeline-
using-computer-vision-and-deep-learning/
Automatically annotating and indexing your images in
Dropbox

Augmented Reality: Translating Service
Google Translate via Image
Not always perfect…

Anaconda makes this machine learning
“magic” accessible to mere mortals
An explosion of innovation is happening in software
Managing this innovation reproducibly is what Anaconda excels at
Anaconda’s coverage of the Python for Data ecosystem helps you with both
the fun part:
• modeling, predicting, classifying, visualizing
…and the not so fun part:
• feature labeling, data-cleaning, data-extraction, scaling, deploying

Bringing
Technology
Together

Data Science Workflow
New Data
Notebooks
Understand Data
Getting Data
Understand World
Reports
Microservices
Dashboards
Applications
Decisions
and
Actions
Models
Exploratory Data Analysis and Viz
Data Products

…DATA
Open Data
Science
Insight
Decisions
Actions
Results
Creations
…
Applications
Dashboards
Notebooks
MicroServices
HDFS SQL Server Teradata Cassandra S3 Flat Files
Virtual Data Lake / Data Catalogue

Empowering the Data Science Team

• Setup your environment
locally (single node) on any
platform or a cluster
HDFSDatabases
• Use same expression to
query data no matter where
it lives
Blaze
• Scale your data processing,
without chaning frameworks
or paradigms
dask
• Present and tell your data
story to decision makers
Bokeh
+
datashader
• Build large scale meaningful
interactive data visualizations
anaconda project
• Deploy your
interactive analytical/
predictive application
Sharing insights
Big Data
Collaboration
Deployment
numba

A
Python v2.7
Conda Sandboxing Technology
B
Python
v3.4
Pandas
v0.18
Jupyter
C
R
R
Essentials
conda
NumPy
v1.11
NumPy
v1.10
Pandas
v0.16
• Language independent
• Platform independent
• No special privileges required
• No VMs or containers
• Enables:
• Reproducibility
• Collaboration
• Scaling
“conda – package everything”

221 W. 6th Street
Suite #1550
Austin, TX 78701
+1 512.222.5440
info@continuum.io
@ContinuumIO
$ anaconda-project run plot —show
conda install tensorflow

Some conda features
• Excellent support for “system-level” environments — like having mini VMs but
much lighter weight than docker (micro containers)
• Minimizes code-copies (uses hard/soft links if possible)
• Simple format: binary tar-ball + metadata
• Metadata allows static analysis of dependencies
• Easy to create multiple “channels” which are repositories for packages
• User installable (no root privileges needed)
• Integrates very well with pip and other language-specific package managers.
• Cross Platform

Basic Conda Usage
Install a package conda install sympy
List all installed packages conda list
Search for packages conda search llvm
Create a new environment conda create -n py3k python=3
Remove a package conda remove nose
Get help conda install --help

Advanced Conda Usage
Install a package in an environment conda install -n py3k sympy
Update all packages conda update --all
Export list of packages conda list --export packages.txt
Install packages from an export conda install --file packages.txt
See package history conda list --revisions
Revert to a revision conda install --revision 23
Remove unused packages and cached tarballs conda clean -pt

• Environments are simple: just link the package to a different directory
• Hard-links are very cheap, and very fast — even on Windows.
• Conda environments are completely independent installations of everything
• No fiddling with PYTHONPATH or sym-linking site-packages — “Activating”
an environment sets everything up so it *just works*
• Unix: 
• Windows:
source activate py3k
Environments
conda create -n py3k python=3.5
activate py3k

lightweight isolated sandbox to manage your dependencies in
a single file and allow reproducibility of your project
environment.yml
$ conda env create
$ source activate ENV_NAME
$
More Sophisticated Environments

Numba
Bokeh
Keras

Machine Learning Explosion
Scikit-Learn
Tensorflow
Keras
XGBoost
theano
lasagne
caffe/caffe2
torch
mxnet / minpy
neon
CNTK
DAAL
Chainer
Dynet
Apache Singa
Shogun
https://github.com/josephmisiti/awesome-machine-learning#python-general-purpose
http://deeplearning.net/software_links/
http://scikit-learn.org/stable/related_projects.html

Without NumPyfrom math import sin, pi
def sinc(x):
if x == 0:
return 1.0
else:
pix = pi*x
return sin(pix)/pix
def step(x):
if x > 0:
return 1.0
elif x < 0:
return 0.0
else:
return 0.5
functions.py
>>> import functions as f
>>> xval = [x/3.0 for x in
range(-10,10)]
>>> yval1 = [f.sinc(x) for x
in xval]
>>> yval2 = [f.step(x) for x
in xval]
Python is a great language but
needed a way to operate quickly
and cleanly over multi-
dimensional arrays.

With NumPyfrom numpy import sin, pi
from numpy import vectorize
import functions as f
vsinc = vectorize(f.sinc)
def sinc(x):
pix = pi*x
val = sin(pix)/pix
val[x==0] = 1.0
return val
vstep = vectorize(f.step)
def step(x):
y = x*0.0
y[x>0] = 1
y[x==0] = 0.5
return y
>>> import functions2 as f
>>> from numpy import *
>>> x = r_[-10:10]/3.0
>>> y1 = f.sinc(x)
>>> y2 = f.step(x)
functions2.py
Offers N-D array, element-by-element
functions, and basic random numbers,
linear algebra, and FFT capability for
Python
http://numpy.org
Fiscally sponsored by NumFOCUS

NumPy: an Array Extension of Python
• Data: the array object
– slicing and shaping
– data-type map to Bytes
• Fast Math (ufuncs):
– vectorization
– broadcasting
– aggregations

shape
NumPy Array
Key Attributes
• dtype
• shape
• ndim
• strides
• data

NumPy Slicing (Selection)
>>> a[0,3:5]
array([3, 4])
>>> a[4:,4:]
array([[44, 45],
[54, 55]])
>>> a[:,2]
array([2,12,22,32,42,52])
50 51 52 53 54 55
40 41 42 43 44 45
30 31 32 33 34 35
20 21 22 23 24 25
10 11 12 13 14 15
0 1 2 3 4 5
>>> a[2::2,::2]
array([[20, 22, 24],
[40, 42, 44]])

Summary
• Provides foundational N-dimensional array composed
of homogeneous elements of a particular “dtype”
• The dtype of the elements is extensive (but not very
extensible)
• Arrays can be sliced and diced with simple syntax to
provide easy manipulation and selection.
• Provides fast and powerful math, statistics, and linear
algebra functions that operate over arrays.
• Utilities for sorting, reading and writing data also
provided.

SciPy “Distribution of Python Numerical Tools masquerading as a Library”
Name Description
cluster KMeans and Vector Quantization
fftpack Discrete Fourier Transform
integrate Numerical Integration
interpolate Interpolation routines
io Data Input and Output
linalg Fast Linear algebra
misc Utilities
ndimage N-dimensional Image processing
Name Description
odr Orthogonal Distance Regression
optimize
Constrained and Unconstrained
Optimization
signal Signal Processing Tools
sparse Sparse Matrices and Algebra
spatial Spatial Data Structures and Algorithms
special Special functions (e.g. Bessel)
stats Statistical Functions and Distributions

a powerful plotting engine
import numpy as np
import matplotlib.mlab as mlab
import matplotlib.pyplot as plt
np.random.seed(0)
# example data
mu = 100 # mean of distribution
sigma = 15 # standard deviation of distribution
x = mu + sigma * np.random.randn(437)
num_bins = 50
fig, ax = plt.subplots()
# the histogram of the data
n, bins, patches = ax.hist(x, num_bins, normed=1)
# add a 'best fit' line
y = mlab.normpdf(bins, mu, sigma)
ax.plot(bins, y, '--')
ax.set_xlabel('Smarts')
ax.set_ylabel('Probability density')
ax.set_title(r'Histogram of IQ: $mu=100$, $
sigma=15$')
# Tweak spacing to prevent clipping of ylabel
fig.tight_layout()
plt.show()

import scipy.misc as misc
im = misc.face()
ax = plt.imshow(im)
plt.title('Racoon Face of size
%d x %d' % im.shape[:2])
plt.savefig(‘face.png’)

import numpy as np
with plt.xkcd():
fig = plt.figure()
ax = fig.add_axes((0.1, 0.2, 0.8, 0.7))
ax.spines['right'].set_color('none')
ax.spines['top'].set_color('none')
plt.xticks([])
plt.yticks([])
ax.set_ylim([-30, 10])
data = np.ones(100)
data[70:] -= np.arange(30)
plt.annotate('THE DAY I REALIZEDnI COULD
COOK BACONnWHENEVER I WANTED’,
xy=(70, 1),
arrowprops=dict(arrowstyle=‘->'),
xytext=(15, -10))
plt.plot(data)
plt.xlabel('time')
plt.ylabel('my overall health')
fig.text(0.5, 0.05, '"Stove Ownership"
from xkcd by Randall Monroe',
ha='center')

from mpl_toolkits.mplot3d import Axes3D
from matplotlib import cm
from matplotlib.ticker import LinearLocator,
FormatStrFormatter
import numpy as np
fig = plt.figure()
ax = fig.gca(projection='3d')
# Make data.
X = np.arange(-5, 5, 0.25)
Y = np.arange(-5, 5, 0.25)
X, Y = np.meshgrid(X, Y)
R = np.sqrt(X**2 + Y**2)
Z = np.sin(R)
# Plot the surface.
surf = ax.plot_surface(X, Y, Z,
cmap=cm.coolwarm, linewidth=0,
antialiased=False)
# Customize the z axis.
ax.set_zlim(-1.01, 1.01)
ax.zaxis.set_major_locator(LinearLocator(10))
ax.zaxis.set_major_formatter(FormatStrFormatter(
'%.02f'))
# Add a color bar which maps values to colors.
fig.colorbar(surf, shrink=0.5, aspect=5)
plt.show()

Easy Data Wrangling
• Adds indexes and labels to 1-d and 2-d NumPy arrays
(Series and DataFrame)
• Many convenience functions and methods to manipulate
messy data-sets including time-series.
• Powerful indexing with automatic data alignment.
• Easy handling of missing data.
• Allows easy joining and merging Data Sets
• Pivots and reshaping (split-apply-combine)
• Powerful group-by operations with summarization
• Builtin visualization using labels and indexes

Easy Data Wrangling
• Series Data Structure
• built for 1-dimensional series data
• homogeneous data
• Two arrays. One of data and another which is the index that can be a
homogeneous array of any type like integers, objects, or date-times.
• DataFrame
• built for 2-dimensional collections of tabular data (think Excel sheet)
• heterogeneous data comprised of multiple Series
• includes an index column allowing sophisticated selection and alignment

Easy Data Wrangling
medals = pd.read_csv('data/medals.csv', index_col='name')
medals.head()
gold = medals['medal'] == 'gold'
won = medals['count'] > 0
medals.loc[gold & won, 'count'].sort_values().plot(kind='bar', figsize=(12,8))

Easy Data Wrangling
google = pd.read_csv('data/goog.csv', index_col='Date', parse_dates=True)
google.info()
google.head()
google.describe()

Easy Data Wrangling
df = pd.read_excel("data/pbpython/salesfunnel.xlsx")
df.head()
table = pd.pivot_table(df,
index=["Manager","Rep","Product"],
values=["Price","Quantity"],
aggfunc=[np.sum,np.mean])

Machine Learning made easy
• Supervised Learning — uses “labeled” data to train a model
• Regression — predicted variable is continuous
• Classification — predicted variable is discrete (but c.f. Logistic “Regression”)
• Unsupervised Learning
• Clustering — discover categories in the data
• Density Estimation — determine representation of data
• Dimensionality Reduction — represent data with fewer variables or feature
vectors
• Reinforcement Learning — “goal-oriented” learning (e.g. drive a car)
• Deep Learning — neural networks with many layers
• Semi-supervised Learning (use some labeled data for training)

Mathematical representation
y = f (x, ✓)Supervised Learning
x
y
✓
Input Data or “feature vectors”
Parameters that determine the model
Training is the process of estimating
these parameters
Labels for training
ˆy Predicted outputs ˆy = f (x, ✓)
f (·, ·) Learning model.
May be part of a family of
models with hyper-
parameters selecting the
specific model

Unsupervised Learning
y = f (x, ✓)Auto-encoding
x
y
✓
Parameters of the model
Labels for training — set equal to Input
Auto-encoded model now represents the data in a lower-dimensional space
Model parameters, or estimated data can be used as feature-vectors
Network can “de-noise” future inputs (project new inputs onto data-space)
f (·, ·)
ˆx = g(ˆ✓)
g(·) = f(x, ·)
Underlying model
Data-specific model
Estimated data.

Supervised Deep Learning
y = f (x, ✓)Same structure
x
y
✓
All the weights between layers
Labels for training
ˆy Predicted outputs ˆy = f (x, ✓)
wijk
zij = g
X
k
wijkzi 1k
!
g (u) =
1
1 + e u

Unsupervised Deep Learning
y = f (x, ✓)Auto-encoding
x
y
✓
All the weights between layers
Labels for training — set equal to Input
wijk
zij = g
X
k
wijkzi 1k
!
g (u) =
1
1 + e u
Auto-encoded network now represents the data in a lower-dimensional space
Outputs of hidden networks (or weights) can be used as feature-vectors
Network can “de-noise” future inputs (project new inputs onto data-space)

Basic scikit-learn experience
1) Create or Load Data
>>> from sklearn import datasets
>>> iris = datasets.load_iris()
>>> digits = datasets.load_digits()
A Scikit-learn dataset is a “dictionary-like” object with input data
stored as the .data attribute and labels stored as the .target attribute
.data attribute is always a 2D array (n_samples, n_features)
May need to extract features from raw data to produce data scikit-learn can use

2) Choose Model (or Build Pipeline of Models)
>>> from sklearn import svm
>>> clf = svm.SVC(gamma=0.001, C=100.)
Most models are model-families and have “hyper-parameters” that specify
the specific model function. Good values for these can be found via grid-search
and cross-validation (easy target for parallelization).
Here “gamma” and “C” are hyper-parameters.
Many choices of models: http://scikit-learn.org/stable/supervised_learning.html

3) Train the Model
>>> clf.fit(data[:-1], labels[:-1])
Models have a “fit” method which updates the parameters-to-be-estimated in
the model in-place so that after fitting the model is “trained”
For validation and scoring you need to leave out some of the data to use
later. cross-validation (e.g. k-fold) techniques can also be parallelized easily.
Here we “leave-one-out” (or n-fold)

4) Predict new values
>>> clf.predict(data[-1:])
Prediction of new data uses the trained parameters of the model. Cross-
validation can be used to understand how sensitive the model is to different
partitions of the data.
>>> from sklearn.model_selection import cross_val_score
>>> scores = cross_val_score(clf, data, target, cv=10)
array([ 0.96…, 1. …, 0.96… , 1. ])

The Jupyter Notebook is an open-source
web application that allows you to create
and share documents that contain live
code, equations, visualizations and
explanatory text. Uses include: data
cleaning and transformation, numerical
simulation, statistical modeling, machine
learning and much more.

Data lineage
Interactive
Visualizations
Advanced notebook extensions
Collaborative Executable Notebooks

64
Scaling Up and Out with
Numba and Dask

Scale Up vs Scale Out
Big Memory &
Many Cores
/ GPU Box
Best of Both
(e.g. GPU Cluster)
Many commodity
nodes in a cluster
ScaleUp
(BiggerNodes)
Scale Out
(More Nodes)
Numba
Dask
Blaze

66
Scaling Up! Optimized
Python with JIT compilation
from Numba

Numba
• Started in 2012
• Release 33 (0.33) in May
• Version 1.0 coming in 2017
• Particularly suited to Numeric computing
• Lots of features!!!
• Ahead of Time Compilation
• Wide community adoption and use
conda install numba

Array Allocation
Looping over ndarray x as an iterator
Using numpy math functions
Returning a slice of the array
Numba decorator 
(nopython=True not required)
2.7x Speedup
over NumPy!
Example: Filter an array

Image Processing
@jit('void(f8[:,:],f8[:,:],f8[:,:])')
def filter(image, filt, output):
M, N = image.shape
m, n = filt.shape
for i in range(m//2, M-m//2):
for j in range(n//2, N-n//2):
result = 0.0
for k in range(m):
for l in range(n):
result += image[i+k-m//2,j+l-n//2]*filt[k, l]
output[i,j] = result
~1500x speed-up

Does not replace the standard Python interpreter 
(all of your existing Python libraries are still available)
Numba Compatibility

New GPU Data Frame project (pyGDF) — GOAI

72
Scaling Out with Dask
(integrates with but doesn’t
depend on Hadoop)

Dask
• Started as part of Blaze in early 2014.
• General parallel programming engine
• Flexible and therefore highly suited for
• Commodity Clusters
• Advanced Algorithms
• Wide community adoption and use
conda install dask
pip install dask[complete] distributed --upgrade

Big DataSmall Data
Numba
Moving from small data to big data

Dask: From User Interaction to Execution
delayed

Dask: Parallel Data Processing
Synthetic views of
Numpy ndarrays
Synthetic views of
Pandas DataFrames
with HDFS support
DAG construction and
workflow manager

Dask is a Python parallel computing library that is:
• Familiar: Implements parallel NumPy and Pandas objects
• Fast: Optimized for demanding for numerical applications
• Flexible: for sophisticated and messy algorithms
• Scales up: Runs resiliently on clusters of 100s of machines
• Scales down: Pragmatic in a single process on a laptop
• Interactive: Responsive and fast for interactive data science
Dask complements the rest of Anaconda. It was developed with 
NumPy, Pandas, and scikit-learn developers.
Overview of Dask

x.T - x.mean(axis=0)
df.groupby(df.index).value.mean()
def load(filename):
def clean(data):
def analyze(result):
Dask array (mimics NumPy)
Dask dataframe (mimics Pandas) Dask delayed (wraps custom code)
b.map(json.loads).foldby(...)
Dask bag (collection of data)
Dask Collections: Familiar Expressions and API

>>> import pandas as pd
>>> df = pd.read_csv('iris.csv')
>>> df.head()
sepal_length sepal_width petal_length petal_width species
0 5.1 3.5 1.4 0.2 Iris-setosa
1 4.9 3.0 1.4 0.2 Iris-setosa
2 4.7 3.2 1.3 0.2 Iris-setosa
3 4.6 3.1 1.5 0.2 Iris-setosa
4 5.0 3.6 1.4 0.2 Iris-setosa
>>> max_sepal_length_setosa = df[df.species
== 'setosa'].sepal_length.max()
5.7999999999999998
>>> import dask.dataframe as dd
>>> ddf = dd.read_csv('*.csv')
>>> ddf.head()
sepal_length sepal_width petal_length petal_width species
0 5.1 3.5 1.4 0.2 Iris-setosa
1 4.9 3.0 1.4 0.2 Iris-setosa
2 4.7 3.2 1.3 0.2 Iris-setosa
3 4.6 3.1 1.5 0.2 Iris-setosa
4 5.0 3.6 1.4 0.2 Iris-setosa
…
>>> d_max_sepal_length_setosa = ddf[ddf.species
== 'setosa'].sepal_length.max()
>>> d_max_sepal_length_setosa.compute()
5.7999999999999998
Dask Dataframes

Dask Graphs: Example Machine Learning
Pipeline

Example 1: Using Dask DataFrames on a cluster with
CSV data
• Built from Pandas DataFrames
• Match Pandas interface
• Access data from HDFS, S3, local, etc.
• Fast, low latency
• Responsive user interface
January, 2016
Febrary, 2016
March, 2016
April, 2016
May, 2016
Pandas
DataFrame}
Dask
DataFrame
}

>>> import numpy as np
>>> np_ones = np.ones((5000, 1000))
>>> np_ones
array([[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
...,
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.]])
>>> np_y = np.log(np_ones + 1)[:5].sum(axis=1)
>>> np_y
array([ 693.14718056, 693.14718056,
693.14718056, 693.14718056, 693.14718056])
>>> import dask.array as da
>>> da_ones = da.ones((5000000, 1000000),
chunks=(1000, 1000))
>>> da_ones.compute()
array([[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
...,
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.]])
>>> da_y = da.log(da_ones + 1)[:5].sum(axis=1)
>>> np_da_y = np.array(da_y) #fits in memory
array([ 693.14718056, 693.14718056,
693.14718056, 693.14718056, …, 693.14718056])
# If result doesn’t fit in memory
>>> da_y.to_hdf5('myfile.hdf5', 'result')
Dask Arrays

NumPy
Array
}
}Dask
Array
Example 2: Using Dask Arrays with global temperature data
• Built from NumPy 
n-dimensional arrays
• Matches NumPy interface
(subset)
• Solve medium-large
problems
• Complex algorithms

Scheduler
Worker
Worker
Worker
Worker
Client
Same network
User Machine (laptop)Client
Worker
Dask Schedulers: Distributed Scheduler

Cluster Architecture Diagram
Client Machine Compute
Node
Compute
Node
Compute
Node
Head Node

• Single machine with multiple threads or processes
• On a cluster with SSH (dcluster)
• Resource management: YARN (knit), SGE, Slurm
• On the cloud with Amazon EC2 (dec2)
• On a cluster with Anaconda for cluster management
• Manage multiple conda environments and packages  
on bare-metal or cloud-based clusters
Using Anaconda and Dask on your Cluster

YARN
JVM
Bottom Line 
2X-100X faster overall performance
• Interact with data in HDFS and
Amazon S3 natively from Python
• Distributed computations without the
JVM & Python/Java serialization
• Framework for easy, flexible
parallelism using directed acyclic
graphs (DAGs)
• Interactive, distributed computing
with in-memory persistence/caching
Bottom Line
• Leverage Python &
R with Spark
Batch
Processing
Interactive
Processing
HDFS
Ibis
Impala
PySpark & SparkR
Python & R
ecosystem
MPI
High Performance,
Interactive,
Batch
Processing
Native
read & write
NumPy, Pandas, …
720+ packages
High Performance
Hadoop

Scheduler Visualization with Bokeh

Look at all of the data with Bokeh’s datashader.
Decouple the data-processing from the visualization. Visualize arbitrarily
large data.
Numba + Dask
• E.g. Open Street Map data:
• About 3 billion GPS coordinates
• https://blog.openstreetmap.org/
2012/04/01/bulk-gps-point-data/.
• This image was rendered in <5
seconds on a standard MacBook
with 16 GB RAM
• Renders in less than a second on
several 128GB Amazon EC2
instances

Categorical data: 2010 US Census
• One point per
person
• 300 million total
• Categorized by
race
• Interactive
rendering with
Numba+Dask
• No pre-tiling

91
Interactive Data Vizualization
Apps with Bokeh

Interactive Data Visualization
• Interactive viz, widgets, and tools
• Versatile high level graphics
• Streaming, dynamic, large data
• Optimized for the browser
• No Javascript
• With or without a server

Rapid Prototyping Visual Apps
• Python interface
• R interface
• Smart plotting

94
Plotting Billions of Points
and Map Integration with
Datashader

Datashader: Rendering a Billion Points of Data
•datashader provides a fast,
configurable visualization pipeline
for faithfully revealing even very
large datasets
• Each of these visualizations
requires just a few lines of code
and no magic numbers to adjust
by trial and error.

Datashader

97
Data Visualization and
Applications made easy with
Holoviews

HoloViews: Stop plotting your data
•Exploring data can be tedious if you use a plotting library directly, because you will need to
specify details about your data (units, dimensions, names, etc.) every time you construct a new
type of plot.
•With HoloViews, you instead annotate your data once, and then ﬂexible plotting comes for free
— HoloViews objects just display themselves, alone or in any combination.
•It’s now easy to lay out subﬁgures, overlay traces, facet or animate a multidimensional dataset,
sample or aggregate to reduce dimensionality, preserving the metadata each time so the results
visualize themselves.
HoloViews makes it simple to create beautiful interactive Bokeh or Matplotlib
visualizations of complex data.

tiles = gv.WMTS(WMTSTileSource(url='https://server.arcgisonline.com/ArcGIS/rest/services/'
'World_Imagery/MapServer/tile/{Z}/{Y}/{X}.jpg'))
tile_options = dict(width=800,height=475,xaxis=None,yaxis=None,bgcolor='black',show_grid=False)
passenger_counts = sorted(df.passenger_count.unique().tolist())
class Options(hv.streams.Stream):
alpha = param.Magnitude(default=0.75, doc="Alpha value for the map opacity")
colormap = param.ObjectSelector(default=cm["fire"], objects=cm.values())
plot = param.ObjectSelector(default="pickup", objects=["pickup","dropoff"])
passengers = param.ObjectSelector(default=1, objects=passenger_counts)
def make_plot(self, x_range=None, y_range=None, **kwargs):
map_tiles = tiles(style=dict(alpha=self.alpha), plot=tile_options)
df_filt = df[df.passenger_count==self.passengers]
points = hv.Points(gv.Dataset(df_filt, kdims=[self.plot+'_x', self.plot+'_y'], vdims=[]))
taxi_trips = datashade(points, width=800, height=475, x_sampling=1, y_sampling=1,
cmap=self.colormap, element_type=gv.Image,
dynamic=False, x_range=x_range, y_range=y_range)
return map_tiles * taxi_trips
selector = Options(name="")
paramnb.Widgets(selector, callback=selector.update)
hv.DynamicMap(selector.make_plot, kdims=[], streams=[selector, RangeXY()])
Data Widgets and
Applications from
Jupyter Notebooks!
https://anaconda.org/jbednar/nyc_taxi-paramnb/notebook

100
Jupyterlab (Where Jupyter is
heading)

• IDE
• Extensible
• Notebook ->
Applications
JupyterLab

More Than Just Notebooks

Building Blocks
File Browser Notebooks Text Editor
TerminalOutputWidgets

A completely
modular
architecture

PyData Barcelona Keynote

More Related Content

What's hot (16)

Similar to PyData Barcelona Keynote (14)

More from Travis Oliphant (12)

Recently uploaded (20)

PyData Barcelona Keynote