Deploying a Machine Learning Model on Heroku with FastAPI
This project aims for predicting whether income exceeds $50K/yr based on census data >50K, <=50K.
Set up Environment
pip install -r /path/to/requirements.txt
ML Pipeline
The ML pipeline is defined by app.predict.py
Model Card
Model Details
The model is built by random forest, defined by max_depth=2, random_state=0 The API is at https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html#sklearn.ensemble.RandomForestClassifier
Intended use for the model and the intended users
The model is used for predicting income based on census or person data for users who expect to know how to earn >50K.
Metrics
Overall Performance: accuracy: 0.7599; fbeta: 0.00627; recall: 0.00128; precision: 0.227272
Data
Census Income Data Set/ https://archive.ics.uci.edu/ml/datasets/census+income
Training Data: app/data/adult.data
Testing Data: app/data/adult.test
Attribute Information
age: continuous.
workclass: Private, Self-emp-not-inc, Self-emp-inc, Federal-gov, Local-gov, State-gov, Without-pay, Never-worked.
fnlwgt: continuous.
education: Bachelors, Some-college, 11th, HS-grad, Prof-school, Assoc-acdm, Assoc-voc, 9th, 7th-8th, 12th, Masters, 1st-4th, 10th, Doctorate, 5th-6th, Preschool.
education-num: continuous.
marital-status: Married-civ-spouse, Divorced, Never-married, Separated, Widowed, Married-spouse-absent, Married-AF-spouse.
occupation: Tech-support, Craft-repair, Other-service, Sales, Exec-managerial, Prof-specialty, Handlers-cleaners, Machine-op-inspct, Adm-clerical, Farming-fishing, Transport-moving, Priv-house-serv, Protective-serv, Armed-Forces.
relationship: Wife, Own-child, Husband, Not-in-family, Other-relative, Unmarried.
race: White, Asian-Pac-Islander, Amer-Indian-Eskimo, Other, Black.
sex: Female, Male.
capital-gain: continuous.
capital-loss: continuous.
hours-per-week: continuous.
native-country: United-States, Cambodia, England, Puerto-Rico, Canada, Germany, Outlying-US(Guam-USVI-etc), India, Japan, Greece, South, China, Cuba, Iran, Honduras, Philippines, Italy, Poland, Jamaica, Vietnam, Mexico, Portugal, Ireland, France, Dominican-Republic, Laos, Ecuador, Taiwan, Haiti, Columbia, Hungary, Guatemala, Nicaragua, Scotland, Thailand, Yugoslavia, El-Salvador, Trinadad&Tobago, Peru, Hong, Holand-Netherlands.
Bias inherent either in data or model
Potentially, the model may be biased by sex.
Accuracy of Female Slice: 0.8905
Accuracy of Male Slice: 0.7151
API Instruction using Local Host
API is defined by main.py
Step1: Type in uvicorn app.main:app --reload in the terminal
Step2: Go to http://127.0.0.1:8000/docs
Step3: Default/POST/income_prediction
Step4: Click Try it out
Step5: Enter the following dictionary in Request body
{ "age": 30, "workclass": "State-gov", "fnlwgt": 77516, "education": "Masters", "education-num": 15, "marital-status": "Never-married", "occupation": "Prof-specialty", "relationship": "Not-in-family", "race": "White", "sex": "Female", "capital-gain": 2174, "capital-loss": 0, "hours-per-week": 40, "native-country": "United-States", "income": 1 }
Step6: Click Execute
Step7: Got to Server response for results
API Continuous Integration using GitHub Actions
https://github.com/features/actions
The CI is defined by config.yaml
Data Versioning using Data Version Control
The pipeline is defined by dvc.yaml, and the versioning is stored by dvc.lock
Deployment if new data or model is available by typing in dvc repro
Directed Acyclic Graph by typing in dvc dag, defined by dvc.yaml
API Continuous Deployment using Heroku
https://dashboard.heroku.com/apps
The app is started by Procfile
The API is deployed at https://census-income-pred.herokuapp.com/
Live API Testing
import requests
import json
response = requests.get('https://census-income-pred.herokuapp.com/')
print(response.status_code)
print(response.json())
data = { "age": 30, "workclass": "State-gov", "fnlwgt": 77516, "education": "Masters", "education_num": 15, "marital_status": "Never-married", "occupation": "Prof-specialty", "relationship": "Not-in-family", "race": "White", "sex": "Female", "capital_gain": 2174, "capital_loss": 0, "hours_per_week": 40, "native_country": "United-States", "income": 1 }
r = requests.post('https://census-income-pred.herokuapp.com/income_prediction', data=json.dumps(data))
print(r.status_code)
print(r.json())