I have several different .weight files that were outputted in training. The reason I did this, is I noticed the model trained better with fewer classes than if I combined all 35 together. Could it possible to loop through code and have multiple model.load_weights()?
Any help is appreciated!
Thanks
I don't see the code, but I can say, that you can try to create multiple class instances of the model class, each of them with their own weights and configs, and than run each of them in the way you want
Related
I am working on a project in which I have about 18 classes with about 4,000 total instances. I have 7 attributes, 1 being string data, the rest nominal. I am currently using StringToWordVector on the string attribute with Platt's SMO classifier, achieving good results. We are about to implement this, but I would like to try other classifiers in case there maybe one I could get better results from. Any suggestions?
Also, should I be using MultiClassClassifier with so many classes? If so, what settings should I try within that?
Any advice is appreciated!
An AdaBoosted J48 Decision Tree yielded the best results has been well established in our division
I have a Presenter class which has references to objects of other classes.These objects are large in number roughly 18-25.I have not injected any of the objects. During the GWT compile, this Presenter is converted to a n number of *.cache.js files, out which one cache.js seems to be very large and this file is downloaded when Application starts, making the start up time longer than usual, Is their a way to efficiently use GWT.runAsync() to divide this file into multiple split points?
Thanks in advance
Thanks,
Rohit
i need to merge two instances(those are output of StrinToWordVector) with different set of attribute but overlapping and completely different instances in weka .is there any easy way to do it ?(in other word,i have two tfidf matrix with diferent doc and different word (but overlapped) and i want to have a tfidf matrix that is union of all of them)
You can use InputMappedClassifier, although there are two other options if you still have your documents in raw format, as discused here.
Please check if weka.join could help you, this is an extended weka.core.Instaces class with methods like innerJoin, leftJoin, fullJoin, update and union.
Any fast ways to merge two nodes into one, without traverse the properties and relatioships myself?
Any advice would be appreciated!
You have to implement that yourself - including conflict resolution if e.g. both nodes contains the same property with different values etc.
Neo4j does not offer any support for this (since this is probably not a very common use case).
I'm using Weka to develop a classifier for a medical problem. This dataset has a class imbalance situation and I want to know if there is also a problem of class overlapping. Each record has 30 attributes, how can I discover if there is class overlapping using Weka features?
Class Overlapping happens when some samples from different classes have very similar characteristics
Cluster your data set.
If your instances belong to same cluster then they are very similar.
Then find error rate using actual class membership.
If your instances belong to same cluster but their classes are different, then you found what you are asking.
To solve the class imbalance problem, you can use SMOTE. It is in the Weka supervised filter (instance). But can you explain what do you mean by class overlapping?
I think you mean by 'class overlapping', Exist similar instances that belong to different classes. Simply, you can remove them. In awk, you could do the following:
awk '!NF || !seen[$0]++' inputFile > outputFile