Keras to Pytorch -> troubles with layers and shape

Admin 1
Viewed 103

I am in the process of converting a Keras model to PyTorch and would need your help.

Keras Code:

def model(input_shape):
    input_layer = keras.layers.Input(input_shape)
    conv1 = keras.layers.Conv1D(filters=16, kernel_size=3, padding="same")(input_layer)
    conv1 = keras.layers.BatchNormalization()(conv1)
    conv1 = keras.layers.ReLU()(conv1)
    global_average_pooling = keras.layers.GlobalAveragePooling1D()(conv1)
    output_layer = keras.layers.Dense(number_of_classes, activation="softmax")(global_average_pooling )
    return keras.models.Model(inputs=input_layer, outputs=output_layer)

Summary of Model:

enter image description here

My Code Is:

class model(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        #number_of_classes = data_config.number_of_classes
        self.conv1 = nn.Conv1d(256,128,1) # PyTorch does not support same padding, 
        self.bn1=nn.BatchNorm1d(128)
        #self.relu=nn.functional.relu_(16)
        self.avg = nn.AvgPool1d(1)
        
    def forward(self,x):
        x = self.conv1(x)
        x = self.bn1(x)
        #x = self.relu(x)
        x = self.avg(x)
        output = F.log_softmax(x)
        return output

Could someone please help me with this?

Greets!

conv-neural-network pytorch keras tensorflow python
2 Answers

My Current Code is:

class Net(nn.Module):
    def __init__(self):
        super(Net,self).__init__()
        self.conv1 = nn.Conv1d(256,128,1)
        self.batch1 = nn.BatchNorm1d(128)
        self.avgpl1 = nn.AvgPool1d(1, stride=1)
        self.fc1 = nn.Linear(128,3)
    
    def forward(self,x):
        x = self.conv1(x)
        x = self.batch1(x)
        x = F.relu(x)
        x = self.avgpl1(x)
        x = torch.flatten(x,1)
        x = F.log_softmax(self.fc1(x))
        return x

Keras Model Parameters: enter image description here

And my Parameters currently are: enter image description here

Admin1

I write PyTorch code like your model.summary from keras and use from torchsummary for showing your PyTorch summary like below:

import torch
import torch.nn as nn
import torch.nn.functional as F 

# pip install torchsummary
from torchsummary import summary


class Net(nn.Module):
  def __init__(self):
      super(Net,self).__init__()
      self.conv1 = nn.Conv1d(1,256,2)
      self.batch1 = nn.BatchNorm1d(256)
      self.avgpl1 = nn.AvgPool1d(1, stride=1)
      self.fc1 = nn.Linear(4096,3)
      
  def forward(self,x):
    x = self.conv1(x)
    x = self.batch1(x)
    x = F.relu(x)
    x = self.avgpl1(x)
    x = torch.flatten(x,1)
    x = F.relu(self.fc1(x))
    return x

net = Net()
print(net)
summary(net,(1, 17))

Output:

Net(
  (conv1): Conv1d(1, 256, kernel_size=(2,), stride=(1,))
  (batch1): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (avgpl1): AvgPool1d(kernel_size=(1,), stride=(1,), padding=(0,))
  (fc1): Linear(in_features=4096, out_features=3, bias=True)
)
Conv1d-1
BatchNorm1d-2
AvgPool1d-3
Linear-4
----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv1d-1              [-1, 256, 16]             768
       BatchNorm1d-2              [-1, 256, 16]             512
         AvgPool1d-3              [-1, 256, 16]               0
            Linear-4                    [-1, 3]          12,291
================================================================
Total params: 13,571
Trainable params: 13,571
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.00
Forward/backward pass size (MB): 0.09
Params size (MB): 0.05
Estimated Total Size (MB): 0.15
----------------------------------------------------------------
user_17405770
Related