Playground
Besides formal and comprehensive projects, I am also interested in solving some interesting problems. Because of the 'casual' traits of these mini projects, these all could be published on GitHub.
1. Mini flow
It uses python to implement a mini tensorflow framework, which supports tensor and model difinition, backpropagation, autograd, and optimizers. This is a good source to learn fundamental deep learning knowledge.
2. Movie Comment Spider
In this repo, I implement a web spider framework that could get the movie comment content and rate stars from the biggest movie comment website -- Douban.com. It is a good source for Chinese NLP learners and researchers to get the corpus.
3. Chinese News Person’s Opinion Mining
In this repo, I implement an unsupervised learning method to mine the persons their statements in Chinese news. It uses word embedding, heuristic search, and dependency parsing methods, could extract the exact person and his / her statements in complicated and complex sentences.
4. CIKM Text Challenge
In this repo, I implement a text classification challenge for CIKM. Using textCNN and embedding methods could get the top 10% in this competition.
5. An experiment of Recurrent Lab
In this repo, I implement an RNN network from scratch. Because the RNNs are difficult to be understood and analyzed. Therefore, this repo tests the different occasions and monitors the weight and bias change.
6. Machine Translation using Sequence to Sequence
In this repo, I implement a Neural Machine Translation model. Based on the Seq2Seq and Attention.
7. Chinse PM2.5 Data Crawler
I implement a web spider in this repo to get PM2.5 information about different Chinese cities. It is a good resource for learners and researchers.
8. Gambling Transaction Mining
It uses semi-supervised learning and unsupervised learning to find illegal transactions.
9. AI Drawing Cat
By a very simple intimation method, the program could draw an input picture as an animated drawing process.
10. TextCNN
In this project, I implement a text CNN classifier, which is a general tool for classifying text.
11. Paper Key Entities Extractor
In this project, using semi-supervised and embedding ways could get the important key entities and words from a paper. If you input some initial keywords, the model will give you more about the related entities in this paper.
12. Toxic Text Classification
In this project, using word-embedding, text CNN and GRU, I implement a model to solve the Kaggle toxic text classification problem, which became the top 10% in this challenge after some simple finetuning.
13. Majority Voting Nets
In this project, I want to try a method to solve the weak-supervised learning problem. By building different minor networks, the network could give the final result by voting. Although this is useful, I find it is just a type of ensemble learning.
14. Music Embedding
In this project, I use the word embedding way to embed the music elements into vectors. After getting these embeddings, we could generate music just using the Seq2Seq model or similar Language Model in NLP.