Hi Folks, First of all, thanks for this fantastic piece of software :) I am trying to run the lda vw example using the wiki1K.dat file inside test/train-sets ...
4
John Langford
john_c_langford
Mar 29, 2011 3:26 am
Matt's the expert here, but I believe this is a change in the codebase as of 5.1. Instead of printing everything out, we wanted to use -f <filename> and -i...
5
Carlos Castillo
castillopdq
Mar 29, 2011 3:31 am
Hi all, I've been trying to use VW using the daemon mode on my Mac Book Pro. I have not been able to, it seems like I'm missing something. I'm trying this...
6
Matt Hoffman
flashrefuse
Mar 29, 2011 3:34 am
That should be right, I'll change the tutorial slides to reflect this. Matt...
7
John Langford
john_c_langford
Mar 29, 2011 3:47 am
At present, you need VW to send messages to VW with the --sendto command. This is nothing like an ideal---we should probably support normal text with an option...
8
John Langford
john_c_langford
Mar 29, 2011 6:50 am
There were a few feature requests at today's tutorial in LA. They are: (1) Better help understanding weights. Two things should help here---text output of...
9
Christopher Smith
cbsmith
Mar 29, 2011 7:40 am
I was looking at the Wikipedia article on the fast inverse square root, and noticed that using the SSE instruction rsqrtss is both faster and more precise than...
10
cqbkrishnan
Mar 29, 2011 1:40 pm
Is there a way to convert the binary regressor file to text? Thanks, -bharath...
11
Nikos
nikos.karamp...
Mar 29, 2011 1:47 pm
... I have actually experimented with the SSE instruction on vowpal_wabbit. It is indeed 7% faster, but there's no difference in accuracy. Basically we apply...
12
John Langford
john_c_langford
Mar 29, 2011 3:13 pm
Matt says he's going to make it happen shortly. -John...
13
Christopher Smith
cbsmith
Mar 29, 2011 6:44 pm
... I wonder why Elan sees a far greater performance win and improved accuracy. Are your numbers isolated to the inverse square root calculation itself or VW...
14
Nikos
nikos.karamp...
Mar 29, 2011 7:46 pm
... root, ... and more ... The ... http://assemblyrequired.crashworks.org/2009/10/16/timing-square-root/ ... yield ... vowpal_wabbit. It ... accuracy. ... ...
15
John Langford
john_c_langford
Apr 1, 2011 6:27 pm
I posted v5.1 tutorial slides on the wiki. https://github.com/JohnLangford/vowpal_wabbit/wiki/v5.1_tutorial.pdf after Daniel's bugfix below. The LDA tutorial...
16
warezio
Apr 1, 2011 6:55 pm
Hi, I blogged about a (slow!) filter tree reduction which is done in perl using vw as a black box. ...
17
dsrosenb
Apr 1, 2011 7:17 pm
Hi, I'd like to use VW in a prediction server setting, where I send in some input, and it sends back the prediction. Any suggestions on how to do this? One...
18
John Langford
john_c_langford
Apr 1, 2011 7:35 pm
In linux, this works with "-p /dev/stdout". Does that work on your system? -John...
19
dsrosenb
Apr 1, 2011 7:47 pm
Funny - I tried that and it segfaults for me. (I'm on centos linux). I do use that technique in other contexts though, so our /dev/stdout isn't totally...
20
John Langford
john_c_langford
Apr 1, 2011 7:50 pm
... VW currently operates in two asynchronous threads: the parser and the learner. In essence, we need to insert an interface between these threads which...
21
John Langford
john_c_langford
Apr 1, 2011 7:58 pm
Version? Compile problems? make clean; make? It works for me with v5.1 on ubuntu 10.04: 3:58PM humpty-52: ./vw -d ~/housing/edgemont_sale_prices -p /dev/stdout...
22
John Langford
john_c_langford
Apr 1, 2011 8:07 pm
Nikos, I get a "document not found" when looking for the patch. -John...
23
John Langford
john_c_langford
Apr 1, 2011 8:58 pm
This is incorporated. Use it with --readable_model <f> Matt, I made the test part a little bit more verbose. You might want to tweak the LDA portion of the...
24
Vaclav Petricek
vaclavpetricek
Apr 1, 2011 9:08 pm
To send predictions to stdout you can try: cat input | vw --quiet -i model -p /dev/stdout Vaclav...
25
Bharath Krishnan
cqbkrishnan
Apr 1, 2011 9:17 pm
I am trying to understand how this works... I was expecting the output to be similar to the lambda (per-topic word probabilities), but there seems to be some...
26
Matt Hoffman
flashrefuse
Apr 1, 2011 9:55 pm
It's transposed from the way it reads in the slides, I'm in the process of changing that. Rows are words, and columns are topics. The reason there are 8192...
27
Bharath Krishnan
cqbkrishnan
Apr 2, 2011 1:53 am
Thanks, that makes sense. I could have sworn I was looking at word counts in the readable model before, but it looks like that was just operator error. If your...
The problem seems to have been with the permissions I had for vw. I was originally using: [drosen@beast ]$ ls -al /usr/local/bin/vw -rwxr-x--x 1 root root...
30
Ryan Gabbard
rg_bbn
Apr 4, 2011 7:44 pm
Hello everyone, I'm trying to use VW learn topics using LDA on a large number of documents and then classify (much smaller batches of) new documents using that...
31
John Langford
john_c_langford
Apr 6, 2011 1:24 am
Why do you expect the predictions to be identical? My intuition is that they would differ. -John...
