[prev in list] [next in list] [prev in thread] [next in thread]
List: wekalist
Subject: [Wekalist] Re: Hidden Layers
From: Nandita Tripathi <tnandita () yahoo ! com>
Date: 2011-03-28 11:56:02
Message-ID: 76092.44367.qm () web112119 ! mail ! gq1 ! yahoo ! com
[Download RAW message or body]
Thanks Mark,
That was really helpful.
Regards,
Nandita
--- On Mon, 3/28/11, wekalist-request@list.scms.waikato.ac.nz \
<wekalist-request@list.scms.waikato.ac.nz> wrote:
> From: wekalist-request@list.scms.waikato.ac.nz \
> <wekalist-request@list.scms.waikato.ac.nz>
> Subject: Wekalist Digest, Vol 97, Issue 68
> To: wekalist@list.scms.waikato.ac.nz
> Date: Monday, March 28, 2011, 7:54 AM
> Send Wekalist mailing list
> submissions to
> wekalist@list.scms.waikato.ac.nz
>
> To subscribe or unsubscribe via the World Wide Web, visit
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
> or, via email, send a message with subject or body 'help'
> to
> wekalist-request@list.scms.waikato.ac.nz
>
> You can reach the person managing the list at
> wekalist-owner@list.scms.waikato.ac.nz
>
> When replying, please edit your Subject line so it is more
> specific
> than "Re: Contents of Wekalist digest..."
>
>
> Today's Topics:
>
> 1. Re: Language classification with Bayer
> filter (Mark Hall)
> 2. Re: Different ways of using ensembles?
> (Mark Hall)
> 3. Re: MLP hidden layers (Mark Hall)
> 4. Filter didn't make the test instance
> immediately available!
> (Vo Nhu Tuan)
> 5. Is there any analysis tool which would
> give output like: (ascoril)
> 6. Filter didn't make the test instance
> immediately available!
> (Vo Nhu Tuan)
> 7. Re: SMO with IDS (Mark Hall)
> 8. Re: Quoting strings from command line
> (Mark Hall)
> 9. Re: how to add starting time and
> execution time in weka
> algorithms (Mark Hall)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Mon, 28 Mar 2011 09:31:46 +1300
> From: Mark Hall <mhall@pentaho.com>
> Subject: Re: [Wekalist] Language classification with Bayer
> filter
> To: "Weka machine learning workbench list."
> <wekalist@list.scms.waikato.ac.nz>
> Message-ID: <4D8F9EB2.1070906@pentaho.com>
> Content-Type: text/plain; charset="ISO-8859-1";
> format=flowed
>
> On 23/03/11 1:38 AM, alex pappas wrote:
> > Hi Alex,
> >
> > I don't see any function in StringToWordVector
> for converting a string
> > to a Vector. From examples I founded online I see
> that
> > StringToWordVector converts from Instances.
> >
> > I want to get a string from a database and to classify
> it.
> >
> > I'm sure I miss something here.
> >
> > Can please give me an example of converting a string
> to a Vector ?
>
> If you have columns in your database that are contain the
> strings that
> you want to convert using StringToWordVector, and their
> types are mapped
> to "text" via the DatabaseUtils.props file, then you can
> connect to your
> database and load the data into the Explorer. String
> attributes will be
> created for these columns. Then you can use the
> FilteredClassifier to
> encapsulate the StringToWordVector filter and your chosen
> classifier in
> order to learn a model. Any test instances, assuming that
> they come from
> a similar database (same column names/types), will then get
> converted by
> the "trained" StringToWordVector filter before reaching the
> classifier
> for prediction.
>
> Cheers,
> Mark.
>
>
>
> ------------------------------
>
> Message: 2
> Date: Mon, 28 Mar 2011 09:37:57 +1300
> From: Mark Hall <mhall@pentaho.com>
> Subject: Re: [Wekalist] Different ways of using ensembles?
> To: "Weka machine learning workbench list."
> <wekalist@list.scms.waikato.ac.nz>
> Message-ID: <4D8FA025.7080603@pentaho.com>
> Content-Type: text/plain; charset="windows-1256";
> format=flowed
>
> On 27/03/11 2:28 AM, Hamad Ali wrote:
> >
> >
> >
> > From: mhall@pentaho.com
> > Subject: Re: [Wekalist] Different ways of using
> ensembles?
> > > I'd suggest to just use the
> Stacking meta classifier. Stacking's level-1
> > > classifier learns from the
> predictions made by level-0 classifier(s).
> > > Predictions (probability
> distributions) generated from level-0
> > > classifier(s) generated via
> cross-validation on the training data become
> > > input to the level-1 classifier
> along with the original class labels.
> >
> > the problem with stacking is that it takes all the
> input from one classifier and adjusts it.
> > what I want to do is to just send one leg/branch to
> another classifier. we might call it some sort of
> class-based stacking.
> > side note: in my dataset, stacking/boosting/bagging
> don't add much of added value. are there any recent addition
> in ensembles that are horribly awesome? I'm somehow unsure
> whether I need to enhance ensembles or select better
> features.
> > thanks for the support,-- H
>
> I can't really see that your approach of having essentially
> "hard-coded"
> rules to send instances that one classifier classifies as a
> certain
> class to another classifier to learn from will outperform
> Stacking.
> basically, this idea uses a fixed threshold (0.5) on the
> probability of
> classification from the first classifier, whereas Stacking
> uses more
> information (the actual probability distribution). If the
> meta
> classifier for Stacking is a rule or tree learner then it
> will
> automatically learn the rules that you propose to have
> hard-coded - if
> they actually lead to improved performance that is.
>
> Cheers,
> Mark.
>
>
>
> ------------------------------
>
> Message: 3
> Date: Mon, 28 Mar 2011 09:40:49 +1300
> From: Mark Hall <mhall@pentaho.com>
> Subject: Re: [Wekalist] MLP hidden layers
> To: "Weka machine learning workbench list."
> <wekalist@list.scms.waikato.ac.nz>
> Message-ID: <4D8FA0D1.5020707@pentaho.com>
> Content-Type: text/plain; charset="ISO-8859-1";
> format=flowed
>
> On 23/03/11 6:31 PM, Nandita Tripathi wrote:
> > Hi Mark,
> >
> > I want to use the weka experimenter for multilayer
> perceptron.
> > If I set the option hidden layers = 50, does this mean
> that there
> > will be one hidden layer with 50 units. How do I set
> for multiple hidden layers?
> >
> > Thanks,
> >
> > Nandita
> >
> >
> >
> >
>
> From the help on options for MultilayerPerceptron:
>
> -H <comma seperated numbers for nodes on each layer>
> The hidden layers to be created for the
> network.
> (Value should be a list of comma
> separated Natural
> numbers or the letters 'a' = (attribs +
> classes) / 2,
> 'i' = attribs, 'o' = classes, 't' =
> attribs .+ classes)
> for wildcard values, Default = a).
>
> Cheers,
> Mark.
>
>
>
> ------------------------------
>
> Message: 4
> Date: Mon, 28 Mar 2011 05:52:53 +0800
> From: Vo Nhu Tuan <vonhutuan@gmail.com>
> Subject: [Wekalist] Filter didn't make the test instance
> immediately
> available!
> To: Wekalist@list.scms.waikato.ac.nz
> Message-ID: <0AD1D9C3-0C04-40D8-95A2-E39F4B37B7CB@gmail.com>
> Content-Type: text/plain; charset=us-ascii
>
> Hi,
>
> Can someone help me with this please?
>
> I'm getting the Exception "Filter didn't make the test
> instance immediately available!" when using
> FilteredClassifier to classify my test instances. The filter
> I used is MultiFilter with 2 filters StringToWordVector and
> StringToNominal. I'm using weka 3.7.3.
>
> Do you know how can I fix this, or any workaround?
>
> Thanks
> Tuan
>
>
> ------------------------------
>
> Message: 5
> Date: Sun, 27 Mar 2011 20:52:57 -0700 (PDT)
> From: ascoril <ascoril12@gmail.com>
> Subject: [Wekalist] Is there any analysis tool which would
> give output
> like:
> To: wekalist@list.scms.waikato.ac.nz
> Message-ID: <31253628.post@talk.nabble.com>
> Content-Type: text/plain; charset=us-ascii
>
>
> for the rule:
>
> [modality=game subject=2] ==> [level=l1 2] ( :conf(1))
>
> if modality is game and subject is 2 then output will come
> for level as l1
>
> In other words, it would interpret the rule without the
> confidenc and
> support and without Java Coding!!
> --
> View this message in context: \
> http://old.nabble.com/Is-there-any-analysis-tool-which-would-give-output-like%3A-tp31253628p31253628.html
> Sent from the WEKA mailing list archive at Nabble.com.
>
>
>
>
> ------------------------------
>
> Message: 6
> Date: Mon, 28 Mar 2011 12:15:07 +0800
> From: Vo Nhu Tuan <vonhutuan@gmail.com>
> Subject: [Wekalist] Filter didn't make the test instance
> immediately
> available!
> To: wekalist@list.scms.waikato.ac.nz
> Message-ID: <E22FDBCE-1350-4E85-86CA-58B08A956813@gmail.com>
> Content-Type: text/plain; charset=us-ascii
>
> Hi,
>
> Can someone help me with this please?
>
> I'm getting the Exception "Filter didn't make the test
> instance immediately available!" when using
> FilteredClassifier to classify my test instances. The filter
> I used was MultiFilter with 2 filters StringToWordVector and
> StringToNominal, and the classifier was Naive Bayes. I'm
> using weka 3.7.3.
>
> Do you know how can I fix this, or any workaround?
>
> Thanks
> Tuan
>
>
> ------------------------------
>
> Message: 7
> Date: Mon, 28 Mar 2011 19:43:55 +1300
> From: Mark Hall <mhall@pentaho.com>
> Subject: Re: [Wekalist] SMO with IDS
> To: "Weka machine learning workbench list."
> <wekalist@list.scms.waikato.ac.nz>
> Message-ID: <4D902E2B.9010600@pentaho.com>
> Content-Type: text/plain; charset="ISO-8859-1";
> format=flowed
>
> On 03/23/2011 11:47 PM, Chintan Bhatt wrote:
> > To develop (classify) Intrusion Detection System with
> SMO, which part of
> > SMO (weka SMO) code i have to use?
>
> I would think that you'd need to use the whole classifier.
>
> Cheers,
> Mark.
>
>
>
>
> ------------------------------
>
> Message: 8
> Date: Mon, 28 Mar 2011 19:56:05 +1300
> From: Mark Hall <mhall@pentaho.com>
> Subject: Re: [Wekalist] Quoting strings from command line
> To: "Weka machine learning workbench list."
> <wekalist@list.scms.waikato.ac.nz>
> Message-ID: <4D903105.8060707@pentaho.com>
> Content-Type: text/plain; charset="ISO-8859-1";
> format=flowed
>
> On 03/25/2011 08:12 AM, andyatnabble wrote:
> >
> >
> > andyatnabble wrote:
> > >
> > > I have a bunch of raw text files that I need to
> convert into tf-idf
> > > vectors. Thus far, I have stored the entire
> contents of the file as a
> > > string attribute, and the class label of the file
> as a nominal attribute.
> > > Now I use StringToWordVector to convert it to
> tf-idf.
> > >
> > > The files that I have got now contain quotes in
> them. So I use Utils.quote
> > > to properly quote the text and convert to tf-idf.
> > >
> > > The question I have is, is there a way I could use
> quote strings from
> > > command line? The rest of my codebase is in Perl,
> so I would rather not
> > > use Java for creating 1 single file to call
> Utils.quote() just to quote my
> > > strings.
> > >
> > > Thanks.
>
> Why not just convert the quotes() routine in Perl?
>
>
>
> ------------------------------
>
> Message: 9
> Date: Mon, 28 Mar 2011 20:03:11 +1300
> From: Mark Hall <mhall@pentaho.com>
> Subject: Re: [Wekalist] how to add starting time and
> execution time in
> weka algorithms
> To: "Weka machine learning workbench list."
> <wekalist@list.scms.waikato.ac.nz>
> Message-ID: <4D9032AF.9020500@pentaho.com>
> Content-Type: text/plain; charset="ISO-8859-1";
> format=flowed
>
> On 03/25/2011 06:47 PM, SHRUTI AGGARWAL wrote:
> > Hello
> > I want to alter a classification algorithms in weka
> and add starting
> > time and execution time to it.
> > Kindly help how can I do so.
> > Thanks
> >
>
> The Explorer and the experiment classes use
> System.currentTimeMillis()
> to measure execution time.
>
> Cheers,
> Mark.
>
>
>
> ------------------------------
>
> _______________________________________________
> Wekalist mailing list
> Wekalist@list.scms.waikato.ac.nz
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>
>
> End of Wekalist Digest, Vol 97, Issue 68
> ****************************************
>
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@list.scms.waikato.ac.nz
List info and subscription status: \
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist List etiquette: \
http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic