[Scikit-learn-general] Latent Dirichlet Allocation

[Rockenkamm, Christian]

Hallo,

I have question concerning the Latent Dirichlet Allocation. The results I get from using it are a bit confusing.
At first I use about 3000 documents. In the preparation with the CountVectorizrt I use the following parameters : max_df=0.95 and min_df=0.05.
For the LDA fit I use the bath learning method. For the other parameters I have tried many different values. However regardless of which configuration I used, I face one common problem. I get topics that are never used in any of the docs and said topics all show the same structure (topic-word-distribution). I even tried gensim with the same configuration as scikit, yet I still encountered this problem. I also tried lowering the number of topics in the model, but this did not lead to the expected results either. For 100 topics, 20-27 were still affected by this problem, for 50 topics, there were still 2-8 of them being affected, depending on the parameter setting.
Does anybody have an idea as to what might be causing this problem and how to resolve it?

Best regards,
Christian Rockenkamm

[Rockenkamm, Christian]

Hallo,

I have question concerning the Latent Dirichlet Allocation. The results I get from using it are a bit confusing.
At first I use about 3000 documents. In the preparation with the CountVectorizrt I use the following parameters : max_df=0.95 and min_df=0.05.
For the LDA fit I use the bath learning method. For the other parameters I have tried many different values. However regardless of which configuration I used, I face one common problem. I get topics that are never used in any of the docs and said topics all show the same structure (topic-word-distribution). I even tried gensim with the same configuration as scikit, yet I still encountered this problem. I also tried lowering the number of topics in the model, but this did not lead to the expected results either. For 100 topics, 20-27 were still affected by this problem, for 50 topics, there were still 2-8 of them being affected, depending on the parameter setting.
Does anybody have an idea as to what might be causing this problem and how to resolve it?

Best regards,
Christian Rockenkamm

[Andreas Mueller]

Hi Christian.
Can you provide the data and code to reproduce?
Best,
Andy

On 01/26/2016 08:21 AM, Rockenkamm, Christian wrote:
>
> Hallo,
>
>
> I have question concerning the Latent Dirichlet Allocation. The
> results I get from using it are a bit confusing.
>
> At first I use about 3000 documents. In the preparation with the
> CountVectorizrt I use the following parameters : max_df=0.95 and
> min_df=0.05.
>
> For the LDA fit I use the bath learning method. For the other
> parameters I have tried many different values. However regardless of
> which configuration I used, I face one common problem. I get topics
> that are never used in any of the docs and said topics all show the
> same structure (topic-word-distribution). I even tried gensim with the
> same configuration as scikit, yet I still encountered this problem. I
> also tried lowering the number of topics in the model, but this did
> not lead to the expected results either. For 100 topics, 20-27 were
> still affected by this problem, for 50 topics, there were still 2-8 of
> them being affected, depending on the parameter setting.
>
> Does anybody have an idea as to what might be causing this problem and
> how to resolve it?
>
>
> Best regards,
>
> Christian Rockenkamm
>
>
>
> ------------------------------------------------------------------------------
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
>
>
> _______________________________________________
> Scikit-learn-general mailing list
> Scikit-learn-***@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

[Joel Nothman]

How many distinct words are in your dataset?

On 27 January 2016 at 00:21, Rockenkamm, Christian <
***@stud.uni-goettingen.de> wrote:

> Hallo,
>
>
> I have question concerning the Latent Dirichlet Allocation. The results I
> get from using it are a bit confusing.
>
> At first I use about 3000 documents. In the preparation with the
> CountVectorizrt I use the following parameters : max_df=0.95 and
> min_df=0.05.
>
> For the LDA fit I use the bath learning method. For the other parameters I
> have tried many different values. However regardless of which configuration
> I used, I face one common problem. I get topics that are never used in any
> of the docs and said topics all show the same structure
> (topic-word-distribution). I even tried gensim with the same configuration
> as scikit, yet I still encountered this problem. I also tried lowering the
> number of topics in the model, but this did not lead to the expected
> results either. For 100 topics, 20-27 were still affected by this problem,
> for 50 topics, there were still 2-8 of them being affected, depending on
> the parameter setting.
>
> Does anybody have an idea as to what might be causing this problem and how
> to resolve it?
>
>
> Best regards,
>
> Christian Rockenkamm
>
>
> ------------------------------------------------------------------------------
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
> _______________________________________________
> Scikit-learn-general mailing list
> Scikit-learn-***@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
>

[Rockenkamm, Christian]

I used more datasets in a range from 2200 to 3500 distinct words in the tf for training the LDA. This data are preprocessed with lemmatizing before CountVectorizrt.
________________________________
Von: Joel Nothman [***@gmail.com]
Gesendet: Dienstag, 26. Januar 2016 23:35
An: scikit-learn-general
Betreff: Re: [Scikit-learn-general] Latent Dirichlet Allocation

How many distinct words are in your dataset?

On 27 January 2016 at 00:21, Rockenkamm, Christian <***@stud.uni-goettingen.de<mailto:***@stud.uni-goettingen.de>> wrote:
Hallo,

I have question concerning the Latent Dirichlet Allocation. The results I get from using it are a bit confusing.
At first I use about 3000 documents. In the preparation with the CountVectorizrt I use the following parameters : max_df=0.95 and min_df=0.05.
For the LDA fit I use the bath learning method. For the other parameters I have tried many different values. However regardless of which configuration I used, I face one common problem. I get topics that are never used in any of the docs and said topics all show the same structure (topic-word-distribution). I even tried gensim with the same configuration as scikit, yet I still encountered this problem. I also tried lowering the number of topics in the model, but this did not lead to the expected results either. For 100 topics, 20-27 were still affected by this problem, for 50 topics, there were still 2-8 of them being affected, depending on the parameter setting.
Does anybody have an idea as to what might be causing this problem and how to resolve it?

Best regards,
Christian Rockenkamm

[Manjush Vundemodalu]

I think you have most of words filtered out of tf because of the condition
min_df=0.05,

I faced similar problems while working with chat data and I tried min_df=2
instead of using float value and it worked

Regards,
Manjush



On Wed, Jan 27, 2016 at 4:31 AM, Rockenkamm, Christian <
***@stud.uni-goettingen.de> wrote:

> I used more datasets in a range from 2200 to 3500 distinct words in the tf
> for training the LDA. This data are preprocessed with lemmatizing before
> CountVectorizrt.
> ------------------------------
> *Von:* Joel Nothman [***@gmail.com]
> *Gesendet:* Dienstag, 26. Januar 2016 23:35
> *An:* scikit-learn-general
> *Betreff:* Re: [Scikit-learn-general] Latent Dirichlet Allocation
>
> How many distinct words are in your dataset?
>
> On 27 January 2016 at 00:21, Rockenkamm, Christian <
> ***@stud.uni-goettingen.de> wrote:
>
>> Hallo,
>>
>>
>> I have question concerning the Latent Dirichlet Allocation. The results I
>> get from using it are a bit confusing.
>>
>> At first I use about 3000 documents. In the preparation with the
>> CountVectorizrt I use the following parameters : max_df=0.95 and
>> min_df=0.05.
>>
>> For the LDA fit I use the bath learning method. For the other parameters
>> I have tried many different values. However regardless of which
>> configuration I used, I face one common problem. I get topics that are
>> never used in any of the docs and said topics all show the same structure
>> (topic-word-distribution). I even tried gensim with the same configuration
>> as scikit, yet I still encountered this problem. I also tried lowering the
>> number of topics in the model, but this did not lead to the expected
>> results either. For 100 topics, 20-27 were still affected by this problem,
>> for 50 topics, there were still 2-8 of them being affected, depending on
>> the parameter setting.
>>
>> Does anybody have an idea as to what might be causing this problem and
>> how to resolve it?
>>
>>
>> Best regards,
>>
>> Christian Rockenkamm
>>
>>
>> ------------------------------------------------------------------------------
>> Site24x7 APM Insight: Get Deep Visibility into Application Performance
>> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
>> Monitor end-to-end web transactions and take corrective actions now
>> Troubleshoot faster and improve end-user experience. Signup Now!
>> http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
>> _______________________________________________
>> Scikit-learn-general mailing list
>> Scikit-learn-***@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>>
>>
>
>
> ------------------------------------------------------------------------------
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
> _______________________________________________
> Scikit-learn-general mailing list
> Scikit-learn-***@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
>

[Vlad Niculae]

I usually use an absolute threshold for min_df and a relative one for
max_df. I find it very useful to look at the histogram of word dfs for
choosing the latter, it varies a lot from dataset to dataset. For
short texts, like tweets, words such as "the" can have a df of 0.1.

It's very easy to look at dfs, just get a transformed X out of your
vectorizer and do:

>>> df = (X > 0).sum(axis=0)
>>> df = df.A.ravel().astype(np.double)
>>> df /= X.shape[0]


My 2c,
Vlad

On Tue, Feb 9, 2016 at 3:05 AM, Manjush Vundemodalu <***@gmail.com> wrote:
> I think you have most of words filtered out of tf because of the condition
> min_df=0.05,
>
> I faced similar problems while working with chat data and I tried min_df=2
> instead of using float value and it worked
>
> Regards,
> Manjush
>
>
>
> On Wed, Jan 27, 2016 at 4:31 AM, Rockenkamm, Christian
> <***@stud.uni-goettingen.de> wrote:
>>
>> I used more datasets in a range from 2200 to 3500 distinct words in the tf
>> for training the LDA. This data are preprocessed with lemmatizing before
>> CountVectorizrt.
>> ________________________________
>> Von: Joel Nothman [***@gmail.com]
>> Gesendet: Dienstag, 26. Januar 2016 23:35
>> An: scikit-learn-general
>> Betreff: Re: [Scikit-learn-general] Latent Dirichlet Allocation
>>
>> How many distinct words are in your dataset?
>>
>> On 27 January 2016 at 00:21, Rockenkamm, Christian
>> <***@stud.uni-goettingen.de> wrote:
>>>
>>> Hallo,
>>>
>>>
>>> I have question concerning the Latent Dirichlet Allocation. The results I
>>> get from using it are a bit confusing.
>>>
>>> At first I use about 3000 documents. In the preparation with the
>>> CountVectorizrt I use the following parameters : max_df=0.95 and
>>> min_df=0.05.
>>>
>>> For the LDA fit I use the bath learning method. For the other parameters
>>> I have tried many different values. However regardless of which
>>> configuration I used, I face one common problem. I get topics that are never
>>> used in any of the docs and said topics all show the same structure
>>> (topic-word-distribution). I even tried gensim with the same configuration
>>> as scikit, yet I still encountered this problem. I also tried lowering the
>>> number of topics in the model, but this did not lead to the expected results
>>> either. For 100 topics, 20-27 were still affected by this problem, for 50
>>> topics, there were still 2-8 of them being affected, depending on the
>>> parameter setting.
>>>
>>> Does anybody have an idea as to what might be causing this problem and
>>> how to resolve it?
>>>
>>>
>>> Best regards,
>>>
>>> Christian Rockenkamm
>>>
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Site24x7 APM Insight: Get Deep Visibility into Application Performance
>>> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
>>> Monitor end-to-end web transactions and take corrective actions now
>>> Troubleshoot faster and improve end-user experience. Signup Now!
>>> http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
>>> _______________________________________________
>>> Scikit-learn-general mailing list
>>> Scikit-learn-***@lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>>>
>>
>>
>>
>> ------------------------------------------------------------------------------
>> Site24x7 APM Insight: Get Deep Visibility into Application Performance
>> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
>> Monitor end-to-end web transactions and take corrective actions now
>> Troubleshoot faster and improve end-user experience. Signup Now!
>> http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
>> _______________________________________________
>> Scikit-learn-general mailing list
>> Scikit-learn-***@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>>
>
>
> ------------------------------------------------------------------------------
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=272487151&iu=/4140
> _______________________________________________
> Scikit-learn-general mailing list
> Scikit-learn-***@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>