jaehong31 / den Goto Github PK
View Code? Open in Web Editor NEWLifelong Learning with Dynamically Expandable Networks, ICLR 2018
Lifelong Learning with Dynamically Expandable Networks, ICLR 2018
Hello,
I wanted to thank you for such interesting work.
When I try to add a layer to the model {dims = [784, 472, 312, 128, 10] }, I am getting an error.
File "DEN_run.py", line 51, in <module> perf, sparsity, expansion = model.add_task(t+1, data) File "/home/sumant/Project/DEN/DEN.py", line 508, in add_task self.expansion_layer[i-1] = self.ex_k - len(useless) IndexError: list assignment index out of range
I'm not sure why it cannot add these neurons to the layer that I added.
Is there a specific number of neurons that should be added to a new layer?
Thanks in advance
I found the experiment setting in the paper is quite different from the code. I am very confused.
I got this error:
from ._conv import register_converters as _register_converters
Traceback (most recent call last):
File "DEN_run.py", line 13, in
flags.DEFINE_integer("dims", [784, 312, 128, 10], "Dimensions about layers including output")
File "/Users/rupesh.karn/anaconda2/lib/python2.7/site-packages/tensorflow/python/platform/flags.py", line 58, in wrapper
return original_function(*args, **kwargs)
File "/Users/rupesh.karn/anaconda2/lib/python2.7/site-packages/absl/flags/_defines.py", line 315, in DEFINE_integer
DEFINE(parser, name, default, help, flag_values, serializer, **args)
File "/Users/rupesh.karn/anaconda2/lib/python2.7/site-packages/absl/flags/_defines.py", line 81, in DEFINE
DEFINE_flag(_flag.Flag(parser, serializer, name, default, help, **args),
File "/Users/rupesh.karn/anaconda2/lib/python2.7/site-packages/absl/flags/_flag.py", line 107, in init
self._set_default(default)
File "/Users/rupesh.karn/anaconda2/lib/python2.7/site-packages/absl/flags/_flag.py", line 196, in _set_default
self.default = self._parse(value)
File "/Users/rupesh.karn/anaconda2/lib/python2.7/site-packages/absl/flags/_flag.py", line 169, in _parse
'flag --%s=%s: %s' % (self.name, argument, e))
absl.flags._exceptions.IllegalFlagValueError: flag --dims=[784, 312, 128, 10]: Expect argument to be a string or int, found <type 'list'>
The issue is in the flag statement in DEN_run.py.
flags.DEFINE_integer("dims", [784, 312, 128, 10], "Dimensions about layers including output")
A list is passed instead of int.
Solution:
Divide the list into integer in DEN_run.py as :
flags.DEFINE_integer("dims0", 784, "Dimensions about input layer")
flags.DEFINE_integer("dims1", 312, "Dimensions about 1st layer")
flags.DEFINE_integer("dims2", 128, "Dimensions about 2nd layer")
flags.DEFINE_integer("dims3", 10, "Dimensions about output layer")
Then, modify in DEN.py for line "self.dims = config.dims" as :
self.dims0 = config.dims0
self.dims1 = config.dims1
self.dims2 = config.dims2
self.dims3 = config.dims3
nn_shape = [self.dims0, self.dims1, self.dims2, self.dims3]
self.dims = nn_shape
To run the code in Jupyter notebook, add these lines before creating the model from DEN class:
remaining_args = FLAGS([sys.argv[0]] + [flag for flag in sys.argv if flag.startswith("--")])
assert(remaining_args == [sys.argv[0]])
Hi,
I see you did not upload the CNN implementation of the code in this Repo. For my work I am taking many ideas from your paper and I would be working on 1D CNN for time series for which I could really use your implementation.
Thnx
Hello.
First would like to congratulate the team on this very interesting paper.
Would it be possible for the original authors to go through and comment the code in a little more detail. I'm an undergraduate taking an ML course and would like to work with DEN, but I'm having a hard time understanding it well enough to a level where I can modify it and make it my own.
Thank you!
Hi
Can you please also upload code for CNN's?
Thank you for your amazing work.
I have a request, I wonder if it is possible to release the code on CIFRA100, namely the CNN structure. I have trouble implementing them.
To achieve this, is it reasonable for me to directly replace Feedforward Networks in MNIST with CNN?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ 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. ๐๐๐
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 โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.