360 questions
2
votes
1
answer
37
views
How do I update pixelClassificationLayer() to a custom loss function?
I have seen in the Mathworks official website for the pixelClassificationLayer() function that I should update it to a custom loss function using the following code:
function loss = modelLoss(Y,T)
...
- 23
0
votes
1
answer
201
views
Cross-entropy loss with varying number of classes
Is there a standard/efficient way in Pytorch to handle cross-entropy loss for a classification problem where the number of classes depends on the sample?
Example:
In a batch of size 3, I have:
logits1 ...
- 105
0
votes
1
answer
105
views
Backward pass of Softmax with CrossEntropy
I'm trying to calculate the Softmax backward pass;
fn softmax_backward(&self, logits: &Vec<f32>) -> Vec<f32> {
let probability = self.forward(logits);
let mut derivative ...
- 11
-1
votes
1
answer
54
views
How to track loss of multi-label MLP?
I am given binary data points of dimension 5,000. I am asked to perform machine learning predicting a binary vector of length 1k, where each position of the output is a class. The classes are not ...
- 53
0
votes
1
answer
76
views
Analytical gradient of Softmax entropy loss does not match the numerical gradient
I'm trying to implement the gradient of the softmax entropy loss in Python. However, I can see that the analytical gradient does not match the numeric gradient. Here is my Python code:
import numpy as ...
0
votes
1
answer
173
views
Why does my CNN for a Binary Classification problem have a constant 50% accuracy with BCELoss vs 80%+ with Cross Entropy Loss?
I am creating a CNN from scratch with Pytorch. I have a balanced dataset of images, split in half for both classes. I am trying to use the BCEwithLogitsLoss function from torch.nn as I have read that ...
- 1
0
votes
1
answer
92
views
cross entropy loss and torch weights mismatch
My targets are primarily class 0, less frequently class 1 or 2
Trying to do cross entropy loss with class weights
The following code
weights = torch.tensor([1., 10, 10.]).to(device)
lossfn = nn....
- 929
0
votes
1
answer
98
views
Medical Binary Classification probability : BCE vs CrossEntropy
I am a medical resident that really enjoys learning about ML, I spent a lot of time reading here and want to make use of it in my field, medical imaging. We have an exam called "Datscan ...
1
vote
1
answer
440
views
Accessing the N values of PyTorch's Cross-Entropy loss function
I am trying to access specific values of PyTorch's Cross-Entropy loss function (torch.nn.functional.cross_entropy) that I believe are being calculated when the input is a vector of length N. I would ...
- 11
0
votes
1
answer
60
views
CrossEntropyLoss loss function type problem
I am trying to solve a classification problem but something is not working for me in the code.
I have a network that receives 30 inputs and outputs 4 actions.
Each action can receive an integer value ...
- 1
0
votes
0
answers
24
views
Why categorical_crossentropy gives 16?
I'm trying to calculate the math behind the code:
import tensorflow as tf
import numpy as np
y_pred_one_hot = np.array([0, 0, 0, 1], dtype='float32')
y_true_one_hot = np.array([0, 1, 0, 0], dtype='...
- 4,860
0
votes
1
answer
112
views
Pytorch's `binary_cross_entropy` seems to implement ln(0) = -100. Why?
I'm curious as to why Pytorch's binary_cross_entropy function seems to be implemented in such a way to calculate ln(0) = -100.
The binary cross entropy function from a math point of view calculates:
H ...
0
votes
1
answer
59
views
What causes this model to not improve?
I'm very new to pytorch and have a system that I believe can unpack and run through data, but when it does so the accuracy it returns with, even after hundreds of epochs, is still worse than random ...
3
votes
1
answer
269
views
Pytorch CrossEntropy Loss, getting error: "RuntimeError: Boolean value of Tensor with more than one value is ambiguous"
I have a classification model, producing predictions for 4 classes in a tensor of shape (256, 1, 4)...256 is the batch size, while the "1" for the second dimension is due to some model ...
- 1,590
0
votes
1
answer
35
views
Tensorflow binary classifier with weighted loss function - Why does train history accuracy doesn't match train accuracy?
I'm trainig a NN with the following code:
model = tf.keras.Sequential([
tf.keras.layers.InputLayer(input_shape=(input_length,)),
tf.keras.layers.Dropout(0.8, seed=42),
tf.keras.layers....
- 204