Naive Bayes Multinomial

Question

I am working with naive bayes multinomial model. I am supposed to use the pseudo code seen in the train method. So those are my questions:

1)I have put most code in but I have some problems mainly in extracting the vocabulary, counting the number of documents in the class, and concatenate the text of all docs in class.

2)I also noticed that the train method I need only requires the documents(aka train_doc). So I have no idea how to tweak to obtain C which is the class.

def train(self, documents):
    # TRAINMULTINOMIALNB(C,D)
    # 1 V <-- EXTRACTVOCABULARY(D)
    # 2 N <-- COUNTDOCS(D)
    # 3 for each c in C
        # 4 do Nc <-- COUNTDOCSINCLASS(D, c)
            # 5 prior[c] <-- Nc/N
            # 6 textc <-- CONCATENATETEXTOFALLDOCSINCLASS(D, c)
            # 7 for each t in V
            # 8 do Tct <-- COUNTTOKENSOFTERM(textc, t)
            # 9 for each t in V
            # 10 do condprob[t][c] <-- Tct+1
    # 11 return V, prior, condprob
    """
    prior={}
    N = len(documents)

    #Vocab
    V = Counter()
    for d in documents:
        V.update(doc[***])

    #COUNTDOCSINCLASS(C,D)
    cdic = Counter(C)
    for d2 in documents:
    for label in C:
            cdic.update({label:int(math.ceil(float(doc[***])))})

    #CONCATENATETEXTOFALLDOCSINCLASS(documents,C)
    ctoadic = defaultdict(Counter)
    for d3 in document:
        for label2 in C:
            if(float(***)>0):
                ctoadic[label].update(doc[***]) 

    #used to get term by class it is in
    tii = defaultdict(Counter)
    for label,word in ctoadic.iteritems():
        for w in word:
            tii[w].update({l:word[w]})

    #getCondProb(tii,ctofadic,C)
    gcp = defaultdict(lambda: defaultdict(float))
    tnw ={} #total number of words in that label
    for l,v inctofadic.iteritems():
        tnwl[l] = sum(v.values())
    for w,count in tii.iteritems():

    #for 0 occurences
    z = [zeroo for zeroo in C if zeroo not in count.keys()]
    for ling in z:
        gcp[w[ling]=1.0/(len(ctofadic[ling])+tnw[ling])
    for ling,val in count.iteritems():
        gcp[w][ling]=float(val+1)/(len(ctofadic[ling])+tnw[ling])

    #Prior    
    for c in C:
        prior[c] = cdic[c] / float(N)
    return V,prior,gcp

For this not to be closed, you'll need to provide an actual question, one with a quantifiable answer. — Hooked, Mar 31 '15 at 20:04

score 0 · Answer 1 · answered Sep 15 '15 at 21:04

For question 1

For vocabulary when you send your data to the classifier also send all the words you encounter under some common label. For example if you have model like this:

l=label1,W=word1 count

l=label1,W=word2 count

.

l=label2,W=word3 count

.

l=label3,W=word1 count

and so on Also add something like :-

Vocab,word1 count

Vocab,word2 count

Vocab,word3 count

here words1,word2,word3 are all the words encountered in training doc but unique. store them in hashmap and flush out.

and then in classifier whenever you encounter "Vocab" increment by 1. and the total will be vocab.

Do the Same for every document keep a different counter and increment whenever new document is encountered.

For Question 2

Are you taking all classes into consideration or a few?

Naive Bayes Multinomial

1 Answers1