Jingyuan Han & Kaidi Xu
python==3.8.5
sklearn==1.1.1
hopsworks
modal
huggingface
The information of the dataset is:
There are 12 columns in the dataset, where 11 are features of the wine and one is the target ("quality").
Since it is a classification problem, we used a decision tree as the model. The training criterion is "gini".
We used both classification_report() and confusion matrix to evaluate the model. The performance of the model, however, was not so good. The classification accuracy was 0.438. The confusiton matrix of the test data was:
-
wine-eda-and-backfill This notebook was aimed to get the dataset, process the data, and upload the data to the feature group of hopworks (backfilling)
-
wine-training-pipeline This notebook was aimed to build a ML model using the data from a feature view, and evaluate the performance of the data. In the end, the model was uploaded to the model registry of hopsworks.
-
daily-wine-feature-pipeline This py file would generate a wine sample everyday and insert it to the feature group on hopsworks
-
wine_batch_inference_pipline This py file would make predictions of the data in the current feature group evey day. Moreover, it created a monitor feature group, which will record the latest 2 preditions.
A confusion matrix will be made based on the daliy predictions and be uploaded to the dataset api on hopsworks.