In this project, we use Pyspark and Pytorch to build a stream processing pipeline to get images from Twitter API and do real time images detection and hashtags generation for tweets.

We used GCP platform to train all models with NVIDIA Tesla V100.
In order to combine Pyspark and Pytorch, we implmented our system on AWS with platform of Databricks. It provides a enviroment where we can run both Pyspark and Pytorch.

• collect_images.ipynb Collect images from Twitter API for semi-supervised learning.
• train_model.ipynb Train ResNet50

• sender.ipynb Create TCP connection, request images with specific hashtags from Twitter API.
• spark_receiver.ipynb Receive DStream from sender and process data to predict the label. At last, it will restore 4 images per window as one npy file for displaying.
• display.ipynb Display the result.
  You should run sender and spark_receiver first. Once they get images and start processing, you should run display. It can display images by reading npy files in the folder. WARNING: It will delete all npy files to do real time display. So if you wan to store all results, please do not run this code.

Please check all implement details and analysis is our report