Machine Learning NoteBook 0809

8/9: to-do

finish the improved network

do some statistically evaluate on the first version of model and find the problems in data

current status

start to use the model with build 2 resnet blocks and a starting block for each of mol and seq

each block contain 2 Conv1d module and the starting block will have 1 more Conv1d module with 1*1 core

class Block(nn.Module):
  def __init__(self, in_channels, out_channels, use_conv1=False, strides=1):
    super().__init__()
    
    self.process = nn.Sequential (
      nn.Conv1d(in_channels, out_channels, 3, stride=strides, padding=1),
      nn.BatchNorm1d(out_channels),
      nn.ReLU(inplace=True),
      nn.Conv1d(out_channels, out_channels, 3, padding=1),
      nn.BatchNorm1d(out_channels)
    )
    
    if use_conv1:
      self.conv1 = nn.Conv1d(in_channels, out_channels, kernel_size=1, stride=strides)
    else:
      self.conv1 = None
      
  def forward(self, x):
      left = self.process(x)
      right = x if self.conv1 is None else self.conv1(x)
      
      return F.relu(left + right)

class cnnNet(nn.Module):
  def __init__(self):
    super().__init__()
    
    self.pre = nn.Sequential (
      nn.Conv1d(1, 32, 7, stride=2, padding=3, bias=False),
      nn.BatchNorm1d(32),
      nn.ReLU(inplace=True),
      nn.MaxPool1d(3, stride=1, padding=1)
    )
    
    self.layer1 = self._make_layer(32, 16, 2)
      
  def _make_layer(self, in_channels, out_channels, block_num, strides=1):

    layers = [Block(in_channels, out_channels, use_conv1=True, strides=strides)] # build the first layer with conv1
    
    for i in range(block_num):
        layers.append(Block(out_channels, out_channels))
      
    return nn.Sequential(*layers)
  
  def forward(self, x):
    x = self.pre(x)
    x = self.layer1(x)
    
    return x

the first version of network is finished
- the model is overfitting try to add the $p$ value in dropout
find a problem that the $Ki$ value in data have some extreme values try to remove those values and train the model again
download the PDB database to evaluate the model(may need to change another database)
already start to write the method’s data part, finished the Ki figure: (this is another version, the original version cannot be founded)

need to do tomorrow

make a to-do list for everything need to do tomorrow
check the performance of the new model
continue working on the Paper’s introduction and conclusion
start to make some figures include the statistic of protein sequence data and molecular data

notice

during the network, you should add nn.BatchNorm1d() after each layer if nn.Linear

figures and reference for paper

figure is shown above

# Code # Notebook # MachineLearning