Hyperparameters of LDA
Can someone please interpret this graph for me? I find out the alpha value for the different number of topics to find the optimal number of topics which would be required to fir the corpus of data I have.What I could make of this graph is that my corpus is stabilizing at 50 topics. 1 value of alpha means 1 topic per document. Do we want to have 1 topic for each document for our corpus?
Find more posts tagged with
Sort by:
1 - 3 of
31
Thank you for your reply. Can you please tell me how to conduct perplexity test in LDA?
Altair EmployeeHi @lambamanika07
perplexity is part of the performance vector which is returned at the per port of LDA.
BR,
MArtin
perplexity is part of the performance vector which is returned at the per port of LDA.
BR,
MArtin
Sort by:
1 - 1 of
11
Altair EmployeeHey,
Neither Alpha nor Beta are "performance" metrics of your model (even though I've added it to the performance output).
Alpha or Beta are parameters of the model. It basically tells you how much either a word is allowed to be associated with more than one topic or one document is allowed to be associated with one topic.
By default, RapidMiner uses heuristics to set it (see documentation of the operators). Further, the algorithm is automatically tuning the hyperparameters internally (you can deactivate this). I think what you see is one of these two effects. You cannot judge on the setting of the number of topics nor on the alpha/beta settings. To do this you would need to check the other performance metrics for different topics (maybe with different alpha/beta or auto-tuning options).
BR,
Martin
Neither Alpha nor Beta are "performance" metrics of your model (even though I've added it to the performance output).
Alpha or Beta are parameters of the model. It basically tells you how much either a word is allowed to be associated with more than one topic or one document is allowed to be associated with one topic.
By default, RapidMiner uses heuristics to set it (see documentation of the operators). Further, the algorithm is automatically tuning the hyperparameters internally (you can deactivate this). I think what you see is one of these two effects. You cannot judge on the setting of the number of topics nor on the alpha/beta settings. To do this you would need to check the other performance metrics for different topics (maybe with different alpha/beta or auto-tuning options).
BR,
Martin
Neither Alpha nor Beta are "performance" metrics of your model (even though I've added it to the performance output).
Alpha or Beta are parameters of the model. It basically tells you how much either a word is allowed to be associated with more than one topic or one document is allowed to be associated with one topic.
By default, RapidMiner uses heuristics to set it (see documentation of the operators). Further, the algorithm is automatically tuning the hyperparameters internally (you can deactivate this). I think what you see is one of these two effects. You cannot judge on the setting of the number of topics nor on the alpha/beta settings. To do this you would need to check the other performance metrics for different topics (maybe with different alpha/beta or auto-tuning options).
BR,
Martin