ogrisel

23 ms·

181.
by ogrisel 12y ago
What is the average and standard deviation of the performance level on this dataset?
182.
Deep learning tutorial: using convolutional nets to detect facial keypoints(danielnouri.org)
5 points by ogrisel 12y ago|0 comments
183.
by ogrisel 12y ago
There might still be ways to speed up at play time by evaluating batches of candidate points in the same evaluation function call in the GPU.
184.
by ogrisel 12y ago
Actually: > Faster evaluation function: I didn’t use the GPU for playing, only for training. So indeed it could probably be optimized. But first I would profile as it can be the case that the bottleneck is Matrix Matrix multiplication in
185.
by ogrisel 12y ago
It's using theano that generates CUDA code. Before reimplementing I would start with profiling the generated code to spot the main computational bottlenecks.
186.
by ogrisel 12y ago
Yes, there is a recent trend to use Recurrent Neural Networks to model the structure and semantics of sentences. This used in particular to do research for Machine Translation by people at Google and the University of Montreal in particular
187.
by ogrisel 12y ago
> does not need labeled data It needed training on 1TB of labeled images in the first place. Arguably it can be used to transfer that knowledge to other tasks with a much smaller amount of labeled samples but still requires supervision.
188.
by ogrisel 12y ago
Yes it's hosted on SF with a CNAME dns entry. We want to move to github but we need to update the website build infra to do so.
189.
by ogrisel 12y ago
Thanks for the clarifications.
190.
by ogrisel 12y ago
> That would not help. It's the wires attached to the transformer that get the current surge, not the transformer itself. Wouldn't it possible to add automated circuit breakers in case of high current surge then? > Plus a Fa
191.
by ogrisel 12y ago
I am wondering: if the large power transformers are the most critical asset, wouldn't it make sense to try to protect them in Faraday cages capable of sustaining high levels of current without melting down? The cost of such cages are l
192.
by ogrisel 12y ago
I agree that the tone of the first chapters is rather neutral and objective or at least not as controversial as expected. I think the author's recommendations which by definition are subjective are to be found in the last chapter which
193.
by ogrisel 12y ago
You can use `all_model_filenames = joblib.dump(model, filename)` after fit on your dev enviroment. joblib will store each numpy array in the model datastructure as an independent file and `all_model_filenames[0] == filename` refers to the f
194.
by ogrisel 13y ago
Agreed but that alone does not explain all the plot [1]: the slope of all the curves was zero or even slightly negative until 1980. Then the share of the top 0.01% started to climb very significantly. I don't know what could explain su
195.
by ogrisel 13y ago
But it's highly non-linear. They have an expected Pearson correlation of zero. Note: there are simple counter example of variables that have a Pearson correlation of 0 yet some non-linear dependency structure as in the 3rd row of the p
196.
by ogrisel 13y ago
The integer seeds of the PRNG on hand and the first (or second or nth) generated pseudo random number on the other hand.
197.
by ogrisel 13y ago
Also some less-known providers such as Brightbox decided to contribute directly to Apache Libcloud rather than writing there own custom Python client SDK.
198.
by ogrisel 13y ago
The Azure API is much more complex that any other providers: implementing it without introducing a dependency on the Azure Python SDK [1] is going to be complex. Just have a look at its source code (e.g. [2] and [3]) to get an idea. All dri
199.
by ogrisel 13y ago
To me the engineering audience group is a big deal. Regular engineers that do no research per-se might be interested in following research related to their application field to ensure that their developments follow the state of the art. As
200.
by ogrisel 13y ago
You should contact the NIPS organizers to have this kind of tools built-in into the conference site. Would be great to have this similarity-based navigation on http://openreview.net as well.
201.
by ogrisel 13y ago
What would be the impact of such a coroutine emulation if the threading is used to leverage multi-core hardware for high performance computing such as done by Atlas [1], OpenBLAS [2] or MKL [3]? These libraries are tuned to maximize CPU cac
202.
by ogrisel 13y ago
No but with n binary dimensions (with value 0 or 1) you can encode 2^n unique identifiers. So with 1000 continuous dimensions (typically values between -1 and 1 coded on 32 bit floats) you can encode quite a bunch of concepts and their nuan
203.
by ogrisel 13y ago
AFAIK not by default. Furthermore the MVCC mode is still considered experimental: http://www.h2database.com/html/advanced.html#mvcc
204.
by ogrisel 13y ago
docker is more like a frontend to pack, ship and deploy specific applications and their configuration as "containers" to be run in lightweight isolated environments (such as LXC currently). Currently if you want to use docker unde
205.
by ogrisel 13y ago
Yes I understood that. But right now if you want to use docker.io on your OSX dev box you have to run linux inside a VM to be able to leverage the docker.io features (building and testing apps in lightweight containers) for later deployment
206.
by ogrisel 13y ago
Will this make it possible to use docker.io directly under OSX without having to use vagrant / virtualbox to run a linux host OS for LXC containers? Right now, only Xen, KVM and EC2 HVM are supported hypervisors. Hopefully an OSX might
207.
by ogrisel 13y ago
Assuming host OS is Linux, I wonder what are the benefit of "KVM + OSv + my_app" vs "LXC + my_app" (e.g. via docker.io).
208.
by ogrisel 13y ago
PyMC follows the library approach to probabilistic programming rather than inventing yet another application specific language that only a niche of developers will be willing to spend time to learn. Despite not introducing a new language sy
209.
by ogrisel 13y ago
memory_profiler [0] mentioned in the post actual can detect memory leaks in compiled C extensions as it uses the OS through the psutil module to measure the aggregate memory usage of the Python process rather than asking the Python runtime
210.
by ogrisel 13y ago
Yes but the current implementation of DBSCAN in sklearn is not very scalable. I think some contributors want to fix this but I am not sure what the current status is.

More ›