Comments (11)
Dear Peghoty,
I'm so sorry to tell you that I'm not Yusuke Sugomori, you made a wrong E-mail address.
Stranger
At 2013-06-06 15:47:26,peghoty [email protected] wrote:
Dear Yusuke Sugomori,
In the C version of your deep learning program, I found a minor bug in DBN_predict (Line 193-200 of the DBN.c file). The initialization of " linear_output" should be moved into the k-loop as follows:
for(k=0; ksigmoid_layers[i].n_out; k++) {
linear_output = 0.0;
for(j=0; jsigmoid_layers[i].n_in; j++) {
linear_output += this->sigmoid_layers[i].W[k][j] * prev_layer_input[j];
}
linear_output += this->sigmoid_layers[i].b[k];
layer_input[k] = sigmoid(linear_output);
}
I guess the same problem may appear in the other versions.
Actually, I want to communicate with you for more details about the deep learning algorithm for specifc applications. May I know your email address?
peghoty
¡ª
Reply to this email directly or view it on GitHub.
from deeplearning.
Hi Peghoty,
I think you should go to the author's website (http://yusugomori.com/) to contact him. BTW, as a user I thank you for the bug report. It also occurs in c++ and java versions as I checked. Though I couldn't find the same bug in the python version, the performance of the python version is the same as the undebugged c++ and java version. So I suspect there's a bug in the python version as well, I just didn't find it. I don't know scala, so I didn't touch it.
best
from deeplearning.
Dear Peghoty,
Thank you for the report, and sorry for the delay in replying to you.
I have just fixed the bug in C, C++, and Java.
from deeplearning.
Dear Yusuke Sugomori,
Thanks for your reply. I'm a novice of deep learing and read your C code very carefully as a guidence for my studying.
May I ask you some more questions?
- In your demonstration example, the training samples are all binary, like (1,0,1,...), does the code deal with the real-valued input(like (0.2,0.5,1.4,...))? If not, how should I adjust the code?
- During the fine-tuning phase, you only update the parameters of the Logistic Regression classifier with parameters of the prevous levels
fixed, is this a popular way? or you just implement it for simplicity? If I want to update all the parameters together in the fine-tuning phase, any good suggestions?
Thanks again, hope to get your reply soon.
Peghoty
At 2013-06-23 23:57:48,yusugomori [email protected] wrote:
Dear Peghoty,
Thank you for the report, and sorry for the delay in replying to you.
I have just fixed the bug in C, C++, and Java.
¡ª
Reply to this email directly or view it on GitHub.
from deeplearning.
Same questions here. Thanks!
from deeplearning.
I have the same question like above and another question: did you guys test the code?It looks like that the result of predict function is not correct,all the test samples have the same predict value...
from deeplearning.
I also faced with such problem on my own testing data, say, all the test samples have the same predict value.
Here is a clue for you, though I'm not sure the above problem comes out because of this.
As we all know, during the pre-training phase, output of the previous RBM will be taken as input of the current RBM.
let x, y be the input and output of the previous layer, respectively, W be the weight matrix which reflects the connection from the current layer to the previous layer, and b be the bias vector of the current layer, then, y can be computed in two ways as follows:
- y = sigmoid(Wx + b)
- z = sigmoid(Wx + b), y = binomial(z)
In the code of yusugomori/DeepLearning, the second one is used. However, is that really reasonable? Why not use the first one?
Maybe you can have a try, wish you good luck.
peghoty
At 2013-07-12 10:15:21,samkuok [email protected] wrote:
I have the same question like above and another question: did you guys test the code?It looks like that the result of predict function is not correct,all the test samples have the same predict value...
¡ª
Reply to this email directly or view it on GitHub.
from deeplearning.
thanks,peghoty
It is really helpful,I will try to check that
from deeplearning.
I found that the result of code "ph_sample[i] * input[j] - nh_means[i] * nv_samples[j]" in the contrastive_divergence is always 0,so the weight won't be updated at all.
from deeplearning.
Thanks for the bug report. It should be
ph_mean[i] * input[j] - nh_means[i] * nv_samples[j]
I'll fix that.
from deeplearning.
Dear Peghoty,
Sorry for the delay.
- In your demonstration example, the training samples are all binary, like (1,0,1,...), does the code deal with the real-valued input(like (0.2,0.5,1.4,...))? If not, how should I adjust the code?
the sample_v_given_h function has to be changed when coping with real-valued data.
I wrote CRBM.py, but I'm not sure if it's correct...
- During the fine-tuning phase, you only update the parameters of the Logistic Regression classifier with parameters of the prevous levels fixed, is this a popular way? or you just implement it for simplicity? If I want to update all the parameters together in the fine-tuning phase, any good suggestions?
yes, it's for simpicity, and i'm also seeking the best solution...
from deeplearning.
Related Issues (20)
- About The SDA weights update problem : Parameter update formulas can I use Gauss-Newton method, specifically how to achieve with python,thabk you
- Over fitting in Stacked Denoising Autoencoders HOT 1
- About Finetune
- why not add makefile for c,cpp for ubuntu(or other linux)? HOT 2
- dbn.pretraining generates W with nan
- DBN_finetune: layer_input used before set
- A Problem with the Deep Belief Network C++ codes HOT 4
- ImportError: No module named logistic_sgd
- what about the license? HOT 1
- Quension about the updation of weight matrix. HOT 1
- Is there this type of deep learning model?
- Can I save the model from training?
- CDBN, Dataset with size of 8000 run into failure
- Debug mode of DBN C++ code leads to equal results HOT 1
- license? HOT 2
- How to evaluate the DBN (Scala) model? HOT 1
- maybe a error
- Hope CDBN is coming soon! HOT 2
- Could I use this in my Android project?
- RBM Weight Updates Issue HOT 5
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from deeplearning.