Lab1 description:
Link to Iris prediction interface: https://huggingface.co/spaces/Yuyang2022/Iris_Prediction
Link to Iris prediction monitor: https://huggingface.co/spaces/Yuyang2022/Iris_Monitor
Part 1: Feature Engineering Refer to Machine_Learning_Model.ipynb for the full implementation
- Drop columns with low predictive power (For example, drop "Name", "Ticket", "PassengerId")
- Convert Cabin column to Deck column
- Impute Nan values (Age with median age, Deck,Embarked with "N" to indicate missing)
- One hot encode categorical variables
Part 2: Model training
- Selected Ensemble method Bagging for model
- Use GridSearch to search for best parameter from specified parameter space
- Use best parameter found to train model in titanic-training-pipeline.py
Part 3: Scripts to upload data to hopsworks
- Uploaded transformed titanic dataset to hopsworks feature group (titanic-feature-pipeline.py)
- Uploaded trained model to hopsworks model registry (titanic-training-pipeline.py)
- Generate random features and upload to hopsworks (titanic-feature-pipeline-daily.py)
- Predict and upload the results in the form of images to hopsworks (titanic-batch-inference-pipeline.py)
-
Image of today's actual label and prediction
-
Dataframe of latest 5 predictions
-
Confusion matrix of all previous predictions
-
Part 4: Hugging face interface
Link to titanic prediction interface https://huggingface.co/spaces/WayneLinn/ID2223-Lab1-Titanic-Survivor-Prediction
All entries must be provided to avoid errors. Output interpretation (In case it is not clear) "Dead 110 years ago rip" -> dead "Dead but not on the titanic" -> survived
Link to titanic prediction monitor https://huggingface.co/spaces/WayneLinn/ID2223-Lab1-Titanic-Prediction-Monitor
Bonus: Added 2 buttons to allow retrivial of the latest result and update the images.
- Press Load
- Wait 2-5 seconds
- Press refresh to see latest result