Machine Learning Notebook 0806

• the first attempt of the network

• fix the problem to processing too slow(with embedding the sequence first)

  • do not use DataFrame in building the DataLoader, translate DataFrame into tensor before put into DataLoader

  1 2 3 4 5 6 7

  # train train_seq = tensor(np.array(train_data.iloc[:,305:])).unsqueeze(dim=1).to(torch.float32) train_mol = tensor(np.array(train_data.iloc[:,5:305])).unsqueeze(dim=1).to(torch.float32) train_Ki = tensor(np.array(train_data.iloc[:,4])) trainDataset = TensorDataset(train_mol,train_seq,train_Ki) trainDataLoader = DataLoader(trainDataset, batch_size=128)

• add normalization layer after ReLU

need to do tomorrow

• check the embedding results
• improve the network accuracy

notice

• when building the CNN part of the network, should add normalization layer every time after Conv
• don’t put DataFrame into DataSet

figures and reference for paper

• about the data:

  • the statistic review of protein length(exclude a single data larger than 7k)

  

  • x label: protein sequence length
  • y label: number of sequence