Python as part of a production machine learning stack by Michael Manapat PyData SV 2014
25 likes21,129 views
The document discusses the integration of machine learning at Stripe to enhance fraud detection and business violation identification through a structured pipeline using tools like scikit-learn and Luigi. It covers the classification process, including model training and feature extraction, along with the deployment of a scoring service for application submissions. The approach emphasizes efficiency, centralized monitoring, and streamlined code for training and scoring.
![Building
a
classifier:
e-‐cigareIes
data = pandas.from_pickle(‘ecigs’)
data.head()
text violator
0 " please verify your age i am 21 years or older ... True
1 coming soon toggle me drag me with your mouse ... False
2 drink moscow mules cart 0 log in or create an ... False
3 vapors electronic cigarette buy now insuper st... True
4 t-shirts shorts hawaii about us silver coll... False
[5 rows x 2 columns]](https://image.slidesharecdn.com/pythonaspartofaproductionmachinelearningstackbymichaelmanapatpydatasv2014-140609121426-phpapp01/85/Python-as-part-of-a-production-machine-learning-stack-by-Michael-Manapat-PyData-SV-2014-10-320.jpg)
![Features
for
text
classifica0on
cv = CountVectorizer
features =
cv.fit_transform(data['text'])
Sparse
matrix
of
word
counts
from
input
text
(omiSng
feature
selec>on)](https://image.slidesharecdn.com/pythonaspartofaproductionmachinelearningstackbymichaelmanapatpydatasv2014-140609121426-phpapp01/85/Python-as-part-of-a-production-machine-learning-stack-by-Michael-Manapat-PyData-SV-2014-11-320.jpg)
![Features
for
text
classifica0on
X_train, X_test, y_train, y_test =
train_test_split(
features, data['violator'],
test_size=0.2)
-‐Avoid
leakage
-Other
cross-‐valida>on
methods](https://image.slidesharecdn.com/pythonaspartofaproductionmachinelearningstackbymichaelmanapatpydatasv2014-140609121426-phpapp01/85/Python-as-part-of-a-production-machine-learning-stack-by-Michael-Manapat-PyData-SV-2014-12-320.jpg)
![Valida0on
probs = model.predict_proba(X_test)
fpr, tpr, thresholds =
roc_curve(y_test, probs[:, 1])
matplotlib.pyplot(fpr, tpr)](https://image.slidesharecdn.com/pythonaspartofaproductionmachinelearningstackbymichaelmanapatpydatasv2014-140609121426-phpapp01/85/Python-as-part-of-a-production-machine-learning-stack-by-Michael-Manapat-PyData-SV-2014-14-320.jpg)
Python as part of a production machine learning stack by Michael Manapat PyData SV 2014
- 1. Python as part of a produc0on machine learning stack Michael Manapat @mlmanapat Stripe
- 2. Outline -‐Why we need ML at Stripe -‐Simple models with sklearn -‐Pipelines with Luigi -‐Scoring as a service
- 3. Stripe is a technology company focusing on making payments easy -‐Short applica>on
- 4. Tokeniza0on Customer browser Stripe Stripe.js Token Merchant server Stripe API call Result
- 5. API Call import stripe stripe.Charge.create( amount=400, currency="usd", card="tok_103xnl2gR5VxTSB” email=customer@example.com" )"
- 6. Fraud / business viola0ons -‐Terms of service viola>ons (weapons) -‐Merchant fraud (card “cashers”) -‐Transac>on fraud -‐No machine learning a year ago
- 7. Fraud / business viola0ons -‐Terms of service viola>ons E-‐cigareMes, drugs, weapons, etc. How do we find these automa>cally?
- 8. Merchant sign up flow Applica>on submission Website scraped Text scored Applica>on reviewed
- 9. Merchant sign up flow Applica>on submission Website scraped Text scored Applica>on reviewed Machine learning pipeline and service
- 10. Building a classifier: e-‐cigareIes data = pandas.from_pickle(‘ecigs’) data.head() text violator 0 " please verify your age i am 21 years or older ... True 1 coming soon toggle me drag me with your mouse ... False 2 drink moscow mules cart 0 log in or create an ... False 3 vapors electronic cigarette buy now insuper st... True 4 t-shirts shorts hawaii about us silver coll... False [5 rows x 2 columns]
- 11. Features for text classifica0on cv = CountVectorizer features = cv.fit_transform(data['text']) Sparse matrix of word counts from input text (omiSng feature selec>on)
- 12. Features for text classifica0on X_train, X_test, y_train, y_test = train_test_split( features, data['violator'], test_size=0.2) -‐Avoid leakage -Other cross-‐valida>on methods
- 13. Training model = LogisticRegression() model.fit(X_train, y_train) Serializer reads from model.intercept_ model.coef_
- 14. Valida0on probs = model.predict_proba(X_test) fpr, tpr, thresholds = roc_curve(y_test, probs[:, 1]) matplotlib.pyplot(fpr, tpr)
- 15. ROC: Receiver opera0ng characteris0c
- 16. Pipeline -‐Fetch website snapshots from S3 -‐Fetch classifica>ons from SQL/Impala -‐Sani>ze text (strip HTML) -‐Run feature genera>on and selec>on -‐Train and serialize model -‐Export valida>on sta>s>cs
- 17. Luigi class GetSnapshots(luigi.Task): def run(self): " "... class GenFeatures(luigi.Task): def requires(self): return GetSnapshots()"
- 18. Luigi runs tasks on Hadoop cluster "
- 19. Scoring as a service "Applica>on submission Website scraped Text scored Applica>on reviewed ThriO RPC Scoring Service
- 20. Scoring as a service struct ScoringRequest { 1: string text 2: optional string model_name } struct ScoringResponse { 1: double score" " "// Experiments? 2: double request_duration }"
- 21. Why a service? -‐Same code base for training/scoring -‐Reduced duplica>on/easier deploys -‐Experimenta>on
- 22. -‐Log requests and responses (Parquet/Impala) -‐Centralized monitoring (Graphite)
- 23. Summary -‐Simple models with sklearn -‐Pipelines with Luigi -‐Scoring as a service Thanks! @mlmanapat