heartcored98 / standalone-deeplearning-chemistry

화학 분자를 딥러닝 아키텍쳐로 다루는 저장소입니다.

Standalone-DeepLearning-Chemistry

2019 딥러닝 홀로서기 화학편 세미나용 저장소입니다.

본 세미나는 머신러닝의 기초, 파이토치 기본 튜토리얼을 시작으로 분자 구조를 딥러닝 모델들로 다루는 기초부터 중급 과정까지를 다룹니다.

만약 저희 세미나가 마음에 드셨다면 우측 상단에 있는 🌟Star를 박아주세요! 미리 감사드리겠습니다!

[10:00-10:40] 딥러닝 기본기 (hypothesis, loss function, back-propagation update)
[10:50-12:00] pytorch 기본 사용법 (data loader, nn.Module, loss function, back propagation, gpu training)
[13:00-13:50] fingerprint - MLP (fingerprint input, MLP, toxicity binary prediction)
[14:00-14:50] smiles - CNN (encode smile with one-hot encoding, CNN, toxicity binary prediction)
[15:00-15:50] molecular graph - GCN (encode molecule with graph, GCN, toxicty binary prediction)
[16:00-16:50] experiment management + hyperparameter tuning with Tensorboard
[17:30-18:00] Wrap Up and Q&A

Contents

[Session 01] ML Basic

• Category of ML Problems
• Hypothesis / Loss(Cost) Function / Optimization
• Model Capacity / Over(Under)-fitting / Regularization

[Session 02] Pytorch Basic

• Handling Tensor
• Data Loader
• MNIST Tutorial

[Session 03] MLP with Fingerprint Representation

• Task Description
• Prepare Dataset (Fingerprint)
• Custom Data Loader
• MLP Model
• GPU Training

[Session 04] CNN with SMILES Representation

• Prepare Dataset
• SMILES 2 2D-matrix
• CNN Model
• Wrapping Up Code Block

[Session 05] GCN with Graph Representation

• Prepare Graph Dataset
• SMILES 2 Graph
• GCN Model

[Session 06] Experiment Management and Hyperparameter Tuning with Tensorboard

• Cifar 10 dataset
• Summarywriter with Tensorboard
• Hparams plugins

[Session 07] Few Practical Tips