ianwhale / nsga-net Goto Github PK
View Code? Open in Web Editor NEWNSGA-Net, a Neural Architecture Search Algorithm
NSGA-Net, a Neural Architecture Search Algorithm
The paper indicates that the crossover and mutation probabilities are 0.9 and 0.02 respectively. Pymoo, instead of using these probabilities, performs crossover and mutation on all offsprings. Moreover, the crossover and mutation that are performed in the code do not correspond to what is indicated in the article.
Thank you,
best regards.
Why do I create many experimental directories during a search?
Hello,
I'm sure this is just user error, but I can't seem to find how to install this misc package. I am using python 3.8.10, Torch 1.13.1+cu116, and pymoo 0.3.0.
PS C:\Users\tstevahn\To_Laptop\Server\Server_Code> & C:/Users/tstevahn/AppData/Local/Programs/Python/Python38/python.exe c:/Users/tstevahn/To_Laptop/Server/Server_Code/Test_NAS/validation/test.py --net_type micro --arch NSGANet --init_channels 26 --filter_increment 4 --SE --auxiliary --model_path weights.pt
Traceback (most recent call last):
File "c:/Users/tstevahn/To_Laptop/Server/Server_Code/Test_NAS/validation/test.py", line 20, in
from misc import utils
ModuleNotFoundError: No module named 'misc'
Hello, guys,I got an error trying to run CIFAR100: unrecognized arguments: --task cifar100. I checked the source code and found that there was no parameter task. I would like to know whether it needs to modify the source code to achieve this.thanks
The paper said that
NSGA-Net + macro search space takes 8 GPU-days
NSGA-Net (6 @ 560) + cutout takes 4 GPU-days
But I run this searching code on Tesla P100-PCIE-16GB, the results are different.
Macro takes 4 days
Micro takes 41 days(30 mins for one network, total 0.5hr * 40 * 50 / 24hr = 41.6days)
This results is far away from the paper. Can you explain?
Thanks!
Does this project support multi-gpu?
Hello,
From what I understand, the Bayesian optimization "exploitation" stage which is detailed in your paper doesn't exist in this repository. Am I correct? And if not can you point it out to me?
Thank you very much,
Elad
Hello,
I changed the macro search of your project in order to support 1-D data. This change mainly consisted of of changing the kernel sizes, and padding sizes of all operators in the file 'macro_decoder.py'.
For example, in all places where 'kernel_size=3' was written, I changed to 'kernel_size=(3,1)'.
Thus I was able to run the method on data of size (1125, 1) instead of (32,32) like CIFAR10.
After this transformation I got bad results (lower than my naive NAS method) by running the macro search with default command line arguments (as written in the README).
Does this approach make sense to you? Do you think any additional changes are needed in order to support 1D data?
In the paper
We set the number of phases np to three and the number of nodes in each phase no to six.
Duringarchitecturesearch, we limit the number of filters (channels) in any node to 16 for each
one of the generated network architecture. We then train them on our training set using standard stochastic gradient descent (SGD) back-propagation algorithm and a cosine annealing learning rate schedule. Our initial learning rate is 0.025 and we train for 25 epochs, which takes about 9 minutes on a NVIDIA 1080Ti GPU implementation in PyTorch .
But in the code, the default parameter --n_nodes(number of nodes per phases) is four.
I set the channel to 16 and n_nodes to six ,but the search process is slow. So, I want to know if you have the concrete configuration about the 9 minutes.
Also , I find that the search code run slower in multiple GPU than run in a GPU,can you explain the phenomenon?
Tnank you very much!
I have installed pymoo 0.3.1, but
ModuleNotFoundError: No module named 'pymoo.util.non_dominated_sorting'
There are BinaryRandomSampling and FloatRandomSampling, but there is no 'RandomSampling'
ModuleNotFoundError: No module named 'models.model_utils'
Hi guys, I can't find the code for checking duplicate architectures in the source code. Could you show me where it is? Thank you very much.
Hi, @ianwhale
Thanks for your great works on NAS and opening your codes.
Here is the error info 🐛 when I run nsga2_main.py :
ImportError: No module named 'config'
and I try to install config module with
pip install config
another problem occured.
AttributeError: module 'config' has no attribute 'parser'
Did I download wrong version of config or config is a missing python file?
when i run “pip3 install -r requirements.txt",it installed lib
but when it installed graphviz, it shows no match version
can anyone help me? thanks
Hi, @ianwhale @mikelzc1990,
Could you explain about the survival rate of the offspring?
Thank you so much
1.Runtime error: cuda is out of memory
2.perform macro and micro search simultaneously?
File "search/evolution_search.py", line 15, in
from search import nsganet as engine
File "/home/ai/Downloads/pratibha/nsga-net-master/search/nsganet.py", line 9, in
from pymoo.algorithms.genetic_algorithm import GeneticAlgorithm
ModuleNotFoundError: No module named 'pymoo.algorithms.genetic_algorithm'
Thank you
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.