Tools for

Human-Centric Machine Learning

@

Ville Tuulos

North Star AI, March 2019

a business

problem

predict

churn

a model

to predict

churn

data

a model

to predict

churn

data

model

data

transforms

data

model

data

transforms

results

data

model

data

transforms

results

compute

data

model

data

transforms

results

compute

schedule

action

data

data

transforms

results

compute

schedule

action

data

audits

model

model

audits

data

data

transforms

results

compute

schedule

action

data

audits

model

model

audits

data

transforms

data

audits

model

model

audits

versioning

Screenplay Analysis Using NLP

Fraud Detection

Title Portfolio Optimization

Estimate Word-of-Mouth Effects

Incremental Impact of Marketing

Classify Support Tickets

Predict Quality of Network

Content Valuation

Cluster Tweets

Intelligent Infrastructure

Machine Translation

Optimal CDN Caching

Predict Churn

Content Tagging

Optimize Production Schedules

Human-Centric?

data

model

data

transforms

results

compute

schedule

action

business

owner

data

scientist

product

engineer

data

engineer

ML

engineer

data

model

data

transforms

results

compute

schedule

action

business

owner

data

scientist

product

engineer

data

data

transforms

results

compute

schedule

action

data

audits

model

model

audits

data

transforms

data

audits

model

model

audits

versioning

data

scientist

data

data

transforms

results

compute

schedule

action

data

audits

model

model

audits

data

transforms

data

audits

model

model

audits

versioning

machine

learning

infrastructure

data

scientist

Metaflow

Notebooks: Nteract

Job Scheduler: Meson

Compute Resources: Titus

Query Engine: Spark

Data Lake: S3

{

data

compute

prototyping

ML Libraries: R, XGBoost, TF etc.

models

ML Wrapping: Metaflow


from metaflow import FlowSpec, step

class MyFlow(FlowSpec):

  @step
  def start(self):
    self.next(self.a, self.b)

  @step
  def a(self):
    self.next(self.join)
  
  @step
  def b(self):
    self.next(self.join)
  
  @step
  def join(self, inputs):
    self.next(self.end)

MyFlow()

start

How to structure my code?

B

A

join

end

start

How to prototype and test

my code locally?

B

A

join

end

x=0

x+=2

x+=3

max(A.x, B.x)



@step
def start(self):
  self.x = 0
  self.next(self.a, self.b)

@step
def a(self):
  self.x += 2 
  self.next(self.join)
  
@step
def b(self):
  self.x += 3 
  self.next(self.join)
  
@step
def join(self, inputs):
  self.out = max(i.x for i in inputs)
  self.next(self.end)

start

How to distribute work over

many parallel jobs?

A

join

end






@step
def start(self):
  self.grid = [’x’,’y’,’z’] 
  self.next(self.a, foreach=’grid’)

@titus(memory=10000)
@step
def a(self):
  self.x = ord(self.input)
  self.next(self.join)
  
@step
def join(self, inputs):
  self.out = max(i.x for i in inputs)
  self.next(self.end)

start

How to get access to more CPUs,

GPUs, or memory?

B

A

join

end









@titus(cpu=16, gpu=1)
@step
def a(self):
  tensorflow.train()
  self.next(self.join)
  
@titus(memory=200000)
@step
def b(self):
  massive_dataframe_operation()
  self.next(self.join)

16 cores, 1GPU

200GB RAM





# Access Savin's runs
namespace('user:savin')
run = Flow('MyFlow').latest_run
print(run.id) # = 234 
print(run.tags) # = ['unsampled_model']

# Access David's runs
namespace('user:david')
run = Flow('MyFlow').latest_run
print(run.id) # = 184 
print(run.tags) # = ['sampled_model']

# Access everyone's runs
namespace(None)
run = Flow('MyFlow').latest_run
print(run.id) # = 184 

start

How to version my results and

access results by others?

B

A

join

end

david: sampled_model

savin: unsampled_model

start

How to deploy my workflow to production?

B

A

join

end

start

How to monitor models and

examine results?

B

A

join

end

x=0

x+=2

x+=3

max(A.x, B.x)

...and much more to improve productivity of data scientists

  • Deploy models as microservices
  • Support for R
  • High-throughput data access
  • Flexible parametrization of workflows
  • Isolated environments for 3rd party dependencies
  • Slack bot!

1. Models are a tiny part of an end-to-end ML system.

2. With proper tooling, data scientists can own the system, end-to-end.

3. Design the tooling with a human-centric mindset.

 

Summary

To improve results of an ML system,

improve the productivity of humans who operate it.

Tools for Human-Centric Machine Learning Infrastructure north star ai 2019

By Ville Tuulos

Tools for Human-Centric Machine Learning Infrastructure north star ai 2019

North Star AI 2019

  • 2,123