Hi All,
I'm new to Mahout and I'm interested in experimenting with it's classifiers.
Right now, I'm just trying to get up and running with the demo's and
examples.
After checking out the mahout trunk, I've tried running the classification
example 20news, but after running the ./examples/bin/buil
When I started with Mahout I had the same errors. In my case, I just
didn't run PrepareTwentyNewsgroups. You may try to accurately repeat
all steps from https://cwiki.apache.org/MAHOUT/twenty-newsgroups.html
On 4 July 2011 12:52, Vijay Santhanam wrote:
> Hi All,
>
> I'm new to Mahout and I'm inte
Thanks Sergey,
I'm still receiving the same error after following those steps.
I've chosen not to use hadoop - does yours work WITH hadoop?
A few bits of info that might be relevant.
My examples/bin/work folder contains the expected folders from test data
preparation and training...
drwxr-xr-x@
Yes, I worked WITH hadoop, but there should be no difference.
Why do you use examples/bin/build/20news-bayes.sh instead of direct
running bin/mahout? Is it the same?
On 4 July 2011 13:12, Vijay Santhanam wrote:
> Thanks Sergey,
>
> I'm still receiving the same error after following those steps.
Paste somewhere your bayes-test-input file.
On 4 July 2011 13:20, Sergey Bartunov wrote:
> Yes, I worked WITH hadoop, but there should be no difference.
>
> Why do you use examples/bin/build/20news-bayes.sh instead of direct
> running bin/mahout? Is it the same?
>
> On 4 July 2011 13:12, Vijay S
Hi Sergey,
I've tried using both the sh script file and following the instructions at
https://cwiki.apache.org/MAHOUT/twenty-newsgroups.html - like you suggested.
Both return the same results.
I've uploaded my bayes-test-input folder to dropbox, the first file is
here...
http://dl.dropbox.com/u/7
Stop, did you _train_ the classifier successfully before running the _test_?
On 4 July 2011 13:30, Vijay Santhanam wrote:
> Hi Sergey,
>
> I've tried using both the sh script file and following the instructions at
> https://cwiki.apache.org/MAHOUT/twenty-newsgroups.html - like you suggested.
> Bo
Hi Sergey,
Yes, there were no errors.
And all the model data seems to have been populated into bayes-model folder.
Also, each main folder in bayes-model has a _SUCESS file.
See the tarball of my trained model here,
http://dl.dropbox.com/u/7881451/bayes-model.tar.gz
Please compare it to your trai
Well, that's strange. Sorry, I can't help you at the moment, maybe
someone else in the mailing list could.
On 4 July 2011 13:49, Vijay Santhanam wrote:
> Hi Sergey,
>
> Yes, there were no errors.
>
> And all the model data seems to have been populated into bayes-model folder.
> Also, each main fo
Thanks anyway Sergey. Could you perhaps upload your bayes-model folder so I
could try that out?
On Mon, Jul 4, 2011 at 7:57 PM, Sergey Bartunov wrote:
> Well, that's strange. Sorry, I can't help you at the moment, maybe
> someone else in the mailing list could.
>
> On 4 July 2011 13:49, Vijay
I tried deleting all the folders from the test and train data except for
alt.atheism, but I get the identical error.
I might try debugging the problem in eclipse rather than from commandline,
but Eclipse doesn't quite want to work either.
On Mon, Jul 4, 2011 at 8:02 PM, Vijay Santhanam
wrote:
>
Can you send me the console dump
Command line + Log written by the program and put it on say pastebin
Robin
On Mon, Jul 4, 2011 at 3:48 PM, Vijay Santhanam
wrote:
> I tried deleting all the folders from the test and train data except for
> alt.atheism, but I get the identical error.
>
> I might
Hi Em,
If I understood well what you're asking, you could implement a new
CandidateItemStrategy class. If you see that interface, there's this
method getCandidateItems(long userID, DataModel dataModel) that has all
parameters you need in order to filter out items that belong to the
unwanted
Hi Robin,
The console dump was a too large for pastebin, so I uploaded it here --
http://dl.dropbox.com/u/7881451/build-20news-bayes-console-output.txt
I performed a fresh checkout only hours ago, and I used script
examples/bin/build-20news-bayes.sh
I've opted to avoid hadoop, but from what I can
Hi Marco,
thank you for pointing me to this direction.
Again I have to ask: What would be more efficient? Rescoring or
CandidateItemStrategy? Where are the differences?
Thanks!
Am 04.07.2011 12:39, schrieb Marko Ciric:
>
> Hi Em,
>
> If I understood well what you're asking, you could impleme
Hi,
I got debugger running w/ eclipse so I could watch what was happening under
the hood.
Here's the exception again
Exception in thread "main" java.lang.IllegalArgumentException: Label not
found: alt.atheism from
at
com.google.common.base.Preconditions.checkArgument(Preconditions.java:88)
at
or
Hi,
Okay, I replaced all the tab characters with space characters for each file
in the bayes-test-input folder and now the classifier completes without
error.
Tomorrow I'll investigate why the trainer correctly parses the tab-separated
label correctly, but the classifier does not. Actually, I kno
Rescoring is done after an item is processed which would be "too late".
CandidateItemStrategy is the one that returns a set of all possible items
that could be recommended, inside an ItemBasedRecommender so it is done
before any rescoring (and even estimating) is processed. Therefore,
implementing
Are you using some non-standard Java character encoding?
On Mon, Jul 4, 2011 at 5:23 PM, Vijay Santhanam
wrote:
> Hi,
>
> Okay, I replaced all the tab characters with space characters for each file
> in the bayes-test-input folder and now the classifier completes without
> error.
>
> Tomorrow I'
No sir.
UTF-8 all the way.
When doing non-sequential training and classification, what class is used
for tokenization?
I get the feeling different tokenizer classes are used for sequential and
parallel training/classification.
On Mon, Jul 4, 2011 at 10:23 PM, Robin Anil wrote:
> Are you us
We are using the default lucene tokenizer. You can also pass in a tokenizer
via the command line.
On Mon, Jul 4, 2011 at 5:55 PM, Vijay Santhanam
wrote:
> No sir.
>
> UTF-8 all the way.
>
> When doing non-sequential training and classification, what class is used
> for tokenization?
>
> I get t
Sorry, I think I asked the wrong question.
I'm asking about training and classification (i.e. post-preparation -- after
steps 3/4 in https://cwiki.apache.org/MAHOUT/twenty-newsgroups.html) phases.
In parallel mode, what class is used for extracting the "label" for a
document?
In sequential mode,
This could be my doing. I noticed that various bits of code split
input files in different ways: StringTokenizer, Pattern, Splitter. And
using different delimiters: space, space/tab, or the weird collection
of delimiters from StringTokenizer. (BTW StringTokenizer is all but
deprecated for this reas
Hi,
I'm new to Mahout and many of the machine learning ideas, but from what I
understand of Naive Bayes classifier, it's possible to train a Naive Bayes
model with continuous, categorical and word-like features from my
understanding of the wikipedia entry
http://en.wikipedia.org/wiki/Naive_Bayes_c
Hi Sean,
Thanks for responding.
I would expect the sequential classifer tokenizer to be identical to what's
used in the parallel classifier tokenizer.
If that's not possible, then NGrams should perhaps be configurable with
where it finds it's first token (i.e. the label).
I'm very new to hadoop
I've read the source for FileDataModel and it suggested using a JDBC
backed implementation for larger datasets so I decided to upgrade our
recommendation system to use MySQLJDBCDataModel with
MySQLJDBCInMemoryItemSimilarity.
I've found that the JDBC backed versions performance is actually wors
Yes, this is trading memory for speed. If you can fit everything in memory,
then you should. FileDataModel is in memory.
MySQLJDBCDataModel is not in memory and queries the DB every time. This is
pretty slow, though by caching item-item similarity as you do, a lot of the
load is removed. However i
Ahh ok. So if I want everything in memory like the file backed solution
I should use ReloadFromJDBCDataModel? I'm going to give that a try right
now.
Typically which solution is recommended for production use?
Thanks
On 7/4/11 10:09 AM, Sean Owen wrote:
Yes, this is trading memory for speed.
Yes. Both are just fine to use in production. For speed and avoiding abuse
of the database, I'd load into memory and tell it to periodically reload.
But that too is a bit of a choice between how often you want to consume new
data and how much work you want to do to recompute new values.
On Mon, Ju
I wouldn't use the in memory JDBC solution.
I was wondering do most people choose the JDBC backed solutions or the
File backed?
On 7/4/11 10:17 AM, Sean Owen wrote:
Yes. Both are just fine to use in production. For speed and avoiding abuse
of the database, I'd load into memory and tell it to
A look into a recent blogpost of mine might maybe be helpful with
choosing the appropriate data access strategies for your recommender
setup. It covers a very common usecase in great detail:
http://ssc.io/deploying-a-massively-scalable-recommender-system-with-apache-mahout/
--sebastian
2011/7/4
Hi,
I'm trying to model the following training data that's targeting the gender
and from what I've been reading in the archives of the mailing list, the
OnlineLogisticRegression classifier is the easiest to get up and running..
sex height (feet) weight (lbs) foot size(inches) male 6 180 12 male
May I ask why you choose to go with
AllSimilarItemsCandidateItemsStrategy over the default
PreferredItemsNeighborhoodCandidateItemsStrategy?
On 7/4/11 10:23 AM, Sebastian Schelter wrote:
A look into a recent blogpost of mine might maybe be helpful with
choosing the appropriate data access stra
If the item similarities are already precomputed there's no sense in
fetching them from the data model, you can just read use the already
precomputed set of possibly similar items as no other items can be
recommended anyway and it's faster to fetch them from a similarity
implementation that holds t
The mahout implementation of Naive_Bayes does not use continuous variables
well. The best bet is to discretize these variables either individually or
together using k-means. Then use the discrete version for the classifier.
The random forest implementation and the SGD implementation are both hap
The wikipedia page recommends binning if you have a large amount of data and
a supervised variable extraction method if not. These are both ways of
preprocessing to discretize continuous variables.
On Mon, Jul 4, 2011 at 11:28 AM, Ted Dunning wrote:
> The mahout implementation of Naive_Bayes do
Thank you Ted
However, even with using the default OnlineLogisiticRegression I'm unable to
get acceptable results when trying to replicate the gender-guesser discussed
in the example of http://en.wikipedia.org/wiki/Naive_Bayes_classifier
For that particular problem, do you recommend I take a
binn
Hi all,
I have trained a bayes model using mahout on my hadoop cluster, and I
want to use this model with my c++ online application. So I will implement the
classifier as mahout did, but I don’t know how to load the model using c++ as
the model are sequence files.
I want a bayes mod
read the java source code and implemenet it in c++
我也不明白为啥你要用阿里巴巴的邮箱
2011-07-05
beneo_7
发件人: 刘逸哲
发送时间: 2011-07-05 10:55
主 题: How could I use bayse model with my C++ online classifier
收件人: "user@mahout.apache.org"
Hi all,
I have trained a bayes model using mahout on my hadoop
You should use AdaptiveLogisticRegression, and you can try the MAHOUT-696.
Xiaobo Gu
On Tue, Jul 5, 2011 at 1:30 AM, Vijay Santhanam
wrote:
> Hi,
>
> I'm trying to model the following training data that's targeting the gender
> and from what I've been reading in the archives of the mailing list
What could be the reason of a poor Lanczos SVD scalability on cluster? I
don't observe any speed-up at all increasing the number of nodes. What am I
doing wrong?
I'm processing a 1x1000 matrix with 1% non-zeros. The elapsed CPU time
scales like this:
1 slave node - 89m39.399s
2 slave nodes - 9
41 matches
Mail list logo
