Scikit-learn pipeline helper for pandas dataframe. This package created to simplify data preprocessing using scikit-learn pipeline. Instead return numpy array this package will return pandas dataframe.
Predict if a driver will file an insurance claim next year. This project use custom Scikit-learn pipeline and LightGBM for modeling. W&B was used for hyper parameters sweeps, experiments tracking, and artifacts logging.
Predict if a taxi driver will get high tips. This project use custom Scikit-learn pipeline and LightGBM for modeling. W&B was used for experiments tracking, and artifacts logging.
Predict house price with structured data and images. We use Keras Functional API and preprocessing data in neural network graph using preprocessing layer. For deployment we use TFServing and Streamlit for app demo.
Furniture detection using Tensorflow Object Detection API. Dataset was obtained from OpenImages and converted into TFRecord format. After training EfficientDet-D0 for 100k steps we got 17.1% mAP on test dataset. For deployment we use Docker and Streamlit for app demo.
Simple implementation of cassava disease identification for mobile device. We train efficientnet_lite3 model using tflite-model-maker and create simple android app.
Replicating 1st place solution of NDSC - Product Matching. ResNet50 and Fastext were used to extract image and title pairs, and LightGBM were used to predict if they are the same or different products.
Implement useful technique in images classification task to speed up training process and increase accuracy like image augmentation, mixed precision, bagging, and test time augmentation.
Finding the best place to open a business in Yogyakarta. Popular venue near urban and college area in Yogyakarta obtained from Foursquare API. K-means algorithm was used to identify similarity each area.