new-york-taxi-fare-prediction

Predicting Taxi Fares of New York City and deploying it on ServiceFoundry

Useful Links

Taxi Fare Prediction model Deployment Video

Blog with instructions on the run (Coming Soon)

Description of the problem

The aim is the predict the taxi fares of New York with input features like pickup & dropoff co-ordinates, passenger count & the timestamp.

Dataset Description

Here is link to the Kaggle Dataset: https://www.kaggle.com/competitions/new-york-city-taxi-fare-prediction

Sample Data:

Model Trained

We trained a XGBRegressor model. The model is trained using xgboost library and is deployed on the cloud using ServiceFoundry 🚀

Instructions to deploy on ServiceFoundry

Setting up servicefoundry

Install and setup servicefoundry on your computer.

pip install servicefoundry

servicefoundry use server https://app.develop.truefoundry.tech

servicefoundry login

Training model

To deploy using the python API run the following command in terminal

Change working directory to train folder


cd train

Create workspace and API key on the TrueFoundry platform
Replace the MLF_API_KEY value in the train.yaml file with the API Key found in secrets tab of your TrueFoundry account and it it in the train.yaml file (Instructions here)
Copy the workspace fqn of the workspace that you want to use from the workspace tab of TrueFoundry (Instructions here) and add it in the train.yaml file 
To deploy using python script:


python train_deploy.py

To deploy using CLI:


servicefoundry deploy --file train_deploy.yaml

Click on the dashboard link in the terminal
Click on the "Trigger Job" on the dashboard to run the training job

Deploying realtime inference

Note: It is necessary to train a model before being able to deploy it as a service 

Change working directory to infer_realtime folder


cd infer_realtime

Create workspace and API key on the TrueFoundry platform
Replace the MLF_API_KEY value in the infer_realtime_deploy.py file with the API Key found in secrets tab of your TrueFoundry account and add it in infer.yaml file (Instructions here)
Copy the workspace_fqn of the workspace that you want to use from the workspace tab of TrueFoundry (Instructions here) and add it in infer.yaml file 
Find the model_version_fqn of the model that you want to deploy from

Go to experiments tracking tab of TrueFoundry
Click on the project name that you trained (taxi-fare-train by default )
Click on models tab
Click on the model name to the model trained to open the tab showing different versions of the model
Copy the FQN of the latest version of the model
Add the latest version in the infer.yaml file

To deploy using python script:


python infer_deploy.py

To deploy using CLI:


servicefoundry deploy --file infer/infer_deploy.yaml

Click on the dashboard link in the terminal to open the service deployment page with FastAPI EndPoint

Querying the deployed model

This can either be done via the fastapi endpoint directly via browser.

You can also query with python script:

request_url = "https://taxi-fare-infer-vishank-betatest-ws.tfy-ctl-euwe1-develop.develop.truefoundry.tech"

features= {

"pickup_datetime":"2015-01-27 13:08:24 UTC",

"pickup_latitude":40.7638053894043,

"pickup_longitude":-73.973320007324219,

"dropoff_latitude":40.74383544921875,

"dropoff_longitude":-73.981430053710938,

"passenger_count":3

}

  

predictions_list = requests.post(

url=urljoin(request_url, "/predict"), json=features

).json()

Deploying Demo

Note: It is necessary to deploy live inference model before being able to deploy a demo 

Create workspace and API key on the TrueFoundry platform
Replace the MLF_API_KEY value in the infer_realtime_deploy.py file with the API Key found in secrets tab of your TrueFoundry account and add it in the demo.yaml file (Instructions here)
Copy the workspace_fqn of the workspace that you want to use from the workspace tab of TrueFoundry and add it in the train.yaml file (Instructions here)
Copy the inference_server_url from:

Go to deployment tab of TrueFoundry
Open the service that was deployment as live inference model ("taxi-fare-prediction" by default)
Copy the Endpoint link

To deploy using python script:


python demo/demo_deploy.py

To deploy using CLI:


servicefoundry deploy --file demo/demo_deploy.yaml

Click on the dashboard link in the terminal
Click on the "Endpoint" link on the dashboard to open the streamlit demo

vishank97 / new-york-taxi-fare-prediction Goto Github PK