Bigram implemented

exp

@@ -0,0 +1,10 @@
+ S
+ P
+ P
+ S
+ S
+ S
+ S
+ P
+ S
+ S

in

@@ -0,0 +1,10 @@
+In which case, tell them I'm in work, or dead, or down the shops. They can come back in when they've got Professor Brian Cox with them. If only cause I want to meet him and steal his hair.	1328302967
+Put me down as another for Mysterious Universe.  Those dudes are brilliant. 	1347836881
+The military of any country would never admit that UFO's have taken down our most powerful weapons.	1331905826
+An example would have been more productive than a downvote, I think.	1315584834
+sorry, but the authors of this article admit that the study is limited and flawed. Also, upon peer review, the study is found to be fallacious. It says this towards the end of the article, a point to which the majority of people usually fail to read. \n\nalso, 463 people is NOT, in any way, compelling evidence for a causal relationship. 	1347389166
+"Are you afraid of science in general, or just the kind you read about on the internet?"	1303864529
+Well, I know it can decrease intraocular pressure, but as for *cancer...*\n\nThe only data I'm aware of relating marijuana to cancer is that it's an antiemetic and an appetite increasing drug. Which, while very good for patient health, are not directly related to the cancer itself, as much as a treatment of the adverse reactions of chemotherapy/radiation treatment (unless if the cancer is in the GI system, or is making weird hormones...).\n\nIf you have any information indicating that marijuana helps in cancer/HIV treatment *other* than as an antiemetic/appetite increasing medication, could you please provide it? I mean, I'm all for it's responsible use, but I don't want to go around spreading misinformation, especially if I'm going to one day be in the medical field.	1318558797
+That could be anything. Why even bother... 	1285029343
+what was the joke?  he deleted it	1337651956
+Its the landing and taking off that/'s hard, and those things will do do that for you as well.  The pilots, for the most part, are there in case shit happens.\n\nBTW stalling a Cessna is fun as hell.	1336941346

predict.py

@@ -3,69 +3,61 @@
 import pickle
 import math
 import re
+import sys
 
-def clear_tokens(tokens, is_text=True):
-    tokens = tokens.replace('\\n', ' ')
-    return tokens
-    tokens = re.sub(r'\(((http)|(https)).*((\.com)|(\.net)|(\.jpg)|(\.html))\)'," ", tokens)
-    tokens = re.sub(r'[\n\&\"\?\\\'\*\[\]\,\;\.\=\+\(\)\!\/\:\`\~\%\^\$\#\@\’\＞\″\±]+', ' ', tokens)
-    tokens = re.sub(r'[\.\-][\.\-]+', ' ', tokens)
-    tokens = re.sub(r'[0-9]+', ' ', tokens)
-    tokens = re.sub(r'œ|·', '', tokens)
-    if is_text:
-        tokens = re.sub(r' +', ' ', tokens)
-    else:
-        tokens = re.sub(r' +', '', tokens)
-    return tokens
-
-def calc_post_prob(post, paranormal_class_logprob, sceptic_class_logprob, word_logprobs):
-    # dla kazdego tokenu z danego posta
+def calc_post_class(post, paranormal_class_logprob, sceptic_class_logprob, bigrams_logprobs):
     text, timestap = post.rstrip('\n').split('\t')
-    text =  clear_tokens(text, True)
+    text = clear_post(text)
     tokens = text.lower().split(' ')
-    #probs = {0.0 : 'sceptic', 0.0 : 'paranormal'}
     probs = {}
-    for class_ in word_logprobs.keys():
-        product = 1
-        for token in tokens:
-            token = clear_tokens(token, False)
+    for class_ in bigrams_logprobs.keys():
+        product = 0
+        for index in range(len(tokens)-1):
+            # we handle bigrams not in models as neutral
+            bigram = tokens[index] + " " + tokens[index + 1]
+            #print(bigram)
             try:
-                product *= word_logprobs[class_][token]
+                product +=  bigrams_logprobs[class_][bigram]
             except KeyError:
-                product *= 1
-            # tu wzoru uzyj
+                product +=0
         if class_ == 'sceptic':
-            product *=  sceptic_class_logprob
+            product += sceptic_class_logprob
         elif class_ == 'paranormal':
-            product *= paranormal_class_logprob
-        probs[abs(product)] = class_
+            product += paranormal_class_logprob
+        probs[product] = class_
     #print(probs)
-# mozna jeszcze zrobic aby bralo kluczowe slowa i wtedy decydowalo ze paranormal
-    if search_for_keywords(text):
-        return 'paranormal'
-    return probs[max(probs.keys())]
+    return probs[min(probs.keys())]
 
-def search_for_keywords(text):
-    keywords = ['paranormal', 'ufo', 'aliens', 'conspiracy', 'aliens', 'atlantis']
-    return any(keyword in text for keyword in keywords)
+def clear_post(post):
+    post = post.replace('\\n', ' ')
+    post = re.sub(r'(\(|)(http|https|www)[a-zA-Z0-9\.\:\/\_\=\&\;\-\?\+]+(\)|)', '', post)
+    post = re.sub(r'[\.\,]+', ' ', post)
+    post = re.sub(r'(&lt|&gt)','',post) 
+    post = re.sub(r'[\'\(\)\?\*\"\`\;0-9\[\]\:\%]+', '', post)
+    post = re.sub(r' \- ', ' ', post) 
+    post = re.sub(r' +', ' ', post)
+    post = post.rstrip(' ')
+    return post
 
 def main():
-    with open('naive_base_model.pkl', 'rb') as f:
+    if len(sys.argv) != 4:
+        print("syntax is ./predict.py in.tsv out.tsv model.pkl")
+        return
+    in_file = sys.argv[1]
+    out_file = sys.argv[2]
+    model = sys.argv[3]
+    with open(model, 'rb') as f:
         pickle_list = pickle.load(f)
+
     paranormal_class_logprob = pickle_list[0]
     sceptic_class_logprob = pickle_list[1]
-    word_logprobs = pickle_list[2]
-    in_file = "test-A/in.tsv"
-    #in_file = "dev-0/in.tsv"
-    out_file = "test-A/out.tsv"
-    #out_file = "dev-0/out.tsv"
-    print (f"in {in_file}")
-    print (f"out {out_file}")
+    bigrams_logprobs = pickle_list[2]
+
     with open(in_file) as in_f, open(out_file, 'w') as out_f:
         for line in  in_f:
-            hyp = calc_post_prob(line, paranormal_class_logprob, sceptic_class_logprob, word_logprobs)
+            hyp = calc_post_class(line, paranormal_class_logprob, sceptic_class_logprob, bigrams_logprobs)
             if hyp == 'sceptic':
-                out_f.write(" S\n")
+                out_f.write(' S\n')
             elif hyp == 'paranormal':
                 out_f.write(' P\n')
 main()

predict_baseline.py

@@ -0,0 +1,71 @@
+#!/usr/bin/python3
+
+import pickle
+import math
+import re
+
+def clear_tokens(tokens, is_text=True):
+    tokens = tokens.replace('\\n', ' ')
+    return tokens
+    tokens = re.sub(r'\(((http)|(https)).*((\.com)|(\.net)|(\.jpg)|(\.html))\)'," ", tokens)
+    tokens = re.sub(r'[\n\&\"\?\\\'\*\[\]\,\;\.\=\+\(\)\!\/\:\`\~\%\^\$\#\@\’\＞\″\±]+', ' ', tokens)
+    tokens = re.sub(r'[\.\-][\.\-]+', ' ', tokens)
+    tokens = re.sub(r'[0-9]+', ' ', tokens)
+    tokens = re.sub(r'œ|·', '', tokens)
+    if is_text:
+        tokens = re.sub(r' +', ' ', tokens)
+    else:
+        tokens = re.sub(r' +', '', tokens)
+    return tokens
+
+def calc_post_prob(post, paranormal_class_logprob, sceptic_class_logprob, word_logprobs):
+    # dla kazdego tokenu z danego posta
+    text, timestap = post.rstrip('\n').split('\t')
+    text =  clear_tokens(text, True)
+    tokens = text.lower().split(' ')
+    #probs = {0.0 : 'sceptic', 0.0 : 'paranormal'}
+    probs = {}
+    for class_ in word_logprobs.keys():
+        product = 1
+        for token in tokens:
+            token = clear_tokens(token, False)
+            try:
+                product *= word_logprobs[class_][token]
+            except KeyError:
+                product *= 1
+            # tu wzoru uzyj
+        if class_ == 'sceptic':
+            product *=  sceptic_class_logprob
+        elif class_ == 'paranormal':
+            product *= paranormal_class_logprob
+        probs[abs(product)] = class_
+        #print(probs)
+# mozna jeszcze zrobic aby bralo kluczowe slowa i wtedy decydowalo ze paranormal
+    if search_for_keywords(text):
+        return 'paranormal'
+    return probs[max(probs.keys())]
+
+def search_for_keywords(text):
+    keywords = ['paranormal', 'ufo', 'aliens', 'conspiracy', 'aliens', 'atlantis']
+    return any(keyword in text for keyword in keywords)
+
+def main():
+    with open('naive_base_model.pkl', 'rb') as f:
+        pickle_list = pickle.load(f)
+    paranormal_class_logprob = pickle_list[0]
+    sceptic_class_logprob = pickle_list[1]
+    word_logprobs = pickle_list[2]
+    in_file = "test-A/in.tsv"
+    #in_file = "dev-0/in.tsv"
+    out_file = "test-A/out.tsv"
+    #out_file = "dev-0/out.tsv"
+    print (f"in {in_file}")
+    print (f"out {out_file}")
+    with open(in_file) as in_f, open(out_file, 'w') as out_f:
+        for line in in_f:
+            hyp = calc_post_prob(line, paranormal_class_logprob, sceptic_class_logprob, word_logprobs)
+            if hyp == 'sceptic':
+                out_f.write(" S\n")
+            elif hyp == 'paranormal':
+                 out_f.write(' P\n')
+main()

train.py

@@ -3,89 +3,81 @@ from collections import defaultdict
 import math
 import pickle
 import re
+import sys
 
-# in expected.tsv
 def calc_class_logprob(expected_path):
-    paranolal_classcount=0
-    sceptic_classcount=0
+    paranormal_classcount = 0
+    sceptic_classcount = 0
+
     with open(expected_path) as f:
         for line in f:
             line = line.rstrip('\n').replace(' ','')
             if 'P' in line:
-                paranolal_classcount +=1
+                paranormal_classcount +=1
             elif 'S' in line:
                 sceptic_classcount +=1
 
-    paranol_prob = paranolal_classcount / (paranolal_classcount + sceptic_classcount)
-    sceptic_prob = sceptic_classcount / (paranolal_classcount + sceptic_classcount)
+    paranol_prob = paranormal_classcount / (paranormal_classcount + sceptic_classcount)
+    sceptic_prob = sceptic_classcount / (paranormal_classcount + sceptic_classcount)
 
     return math.log(paranol_prob), math.log(sceptic_prob)
 
-def clear_tokens(tokens, is_text=True):
-    tokens = tokens.replace('\\n', ' ')
-    return tokens
-    # delete links, special characters, kropki, and \n
-    tokens = re.sub(r'\(((http)|(https)).*((\.com)|(\.net)|(\.jpg)|(\.html))\)'," ", tokens)
-    tokens = re.sub(r'(|\-|\_)([a-z]+(\-|\_))+[a-z]+(|\-|\_)', ' ', tokens)
-    tokens = re.sub(r'[\n\&\"\?\\\'\*\[\]\,\;\.\=\+\(\)\!\/\:\`\~\%\^\$\#\@\’\＞\″\±]+', ' ', tokens)
-    tokens = re.sub(r'[\.\-][\.\-]+', ' ', tokens)
-    tokens = re.sub(r'[0-9]+', ' ', tokens)
-    tokens = re.sub(r'œ|·', '', tokens)
-    if is_text:
-        tokens = re.sub(r' +', ' ', tokens)
-    else:
-        tokens = re.sub(r' +', '', tokens)
-    return tokens
+def clear_post(post):
+    post = post.replace('\\n', ' ')
+    # delete links
+    post = re.sub(r'(\(|)(http|https|www)[a-zA-Z0-9\.\:\/\_\=\&\;\?\+]+(\)|)', '', post)
+    post = re.sub(r'[\.\,\/]+', ' ', post)
+    post = re.sub(r'(&lt|&gt)','',post)
+    post = re.sub(r'[\'\(\)\?\*\"\`\;0-9\[\]\:\%]+', '', post)
+    post = re.sub(r' \- ', ' ', post)
+    post = re.sub(r' +', ' ', post)
+    post = post.rstrip(' ')
+    return post
 
-# ile razy slowo wystepuje w dokumentach w danej klasie
-def calc_word_count(in_path, expected_path):
-    word_counts = {'paranormal':defaultdict(int), 'sceptic': defaultdict(int)} # dzienik zawierajacy slownik w ktorym s slowa i ile razy wystepuja
-    with open(in_path) as infile, open(expected_path)  as expectedfile:
-        for line, exp in zip(infile, expectedfile):
-            class_ = exp.rstrip('\n').replace(' ','')
-            text, timestap =line.rstrip('\n').split('\t')
-            #print(f"text  {type(text)}")
-            text = clear_tokens(text, True)
+def calc_bigram_count(in_path, expected_path):
+    bigram_counts = {'paranormal' : defaultdict(int), 'sceptic' : defaultdict(int)}
+    with open(in_path) as infile, open(expected_path) as expected_file:
+        for line, exp in zip(infile, expected_file):
+            class_ = exp.rstrip('\n').replace(' ', '')
+            text, timestap = line.rstrip('\n').split('\t')
+            text = clear_post(text)
             tokens = text.lower().split(' ')
-            #print(f"tokens {type(tokens)}")
-            for token in tokens:
-                clear_tokens(token,False)
+            for index in range(len(tokens)-1):
+                # if there is next token we append current and next
+                bigram = tokens[index] + " " + tokens[index + 1]
+                #print(bigram)
+                #print (f"bigram constructed from ;;;;{tokens[index]}:{tokens[index+1]};;;;;;;")
                 if class_ == 'P':
-                    word_counts['paranormal'][token] += 1
+                    bigram_counts['paranormal'][bigram] +=1
                 elif class_ == 'S':
-                    word_counts['sceptic'][token]+=1
+                    bigram_counts['sceptic'][bigram] +=1
+    return bigram_counts
 
-    return word_counts
+def calc_bigram_logprobs(bigram_counts):
+    total_sceptic = sum(bigram_counts['sceptic'].values()) + len(bigram_counts['sceptic'].keys())
+    total_paranormal = sum(bigram_counts['paranormal'].values()) + len(bigram_counts['paranormal'].keys())
+    bigram_logprobs = {'paranormal' : {}, 'sceptic' : {}}
+    for class_ in bigram_counts.keys():
+        for bigram, value in bigram_counts[class_].items():
+            if class_ == "sceptic":
+                bigram_prob = (value + 1) / total_sceptic
+            elif class_ == "paranormal":
+                bigram_prob = (value + 1) / total_paranormal
 
-def calc_word_logprobs(word_counts):
-    total_skeptic = sum(word_counts['sceptic'].values()) + len(word_counts['sceptic'].keys())
-    total_paranormal = sum(word_counts['paranormal'].values())+ len(word_counts['paranormal'].keys())
-    word_logprobs= {'paranormal': {}, 'sceptic': {}}
-    for class_ in word_counts.keys(): # sceptic paranormal
-        for token, value in word_counts[class_].items():
-            if class_ == 'sceptic':
-                word_prob = (value +1)/ total_skeptic
-            elif class_ == 'paranormal':
-                word_prob = (value+1)/ total_paranormal
+            bigram_logprobs[class_][bigram] = math.log(bigram_prob)
 
-            #print (token)
-            word_logprobs[class_][token] = math.log(word_prob)
-
-    return word_logprobs
+    return bigram_logprobs
 
 def main():
-    expected = './train/expected.tsv'
-    #expected = './dev-0/expected.tsv'
-    in_f = './train/in.tsv'
-    #in_f = './dev-0/in.tsv'
-    print (f"expected {expected}")
-    print (f"in {in_f}")
-    paranormal_class_lgprob, skeptic_class_logprob = calc_class_logprob(expected)
-    wordcounts =calc_word_count(in_f,expected)
-
-    word_logprobs = calc_word_logprobs(wordcounts)
-    with open('naive_base_model.pkl', 'wb') as f:
-        pickle.dump([paranormal_class_lgprob, skeptic_class_logprob, word_logprobs], f)
-     # w predict.py bierzemy ten wzor argmax P(w) iloczynP(w|c)
-
+    if len(sys.argv) != 4:
+        print("syntax is ./train.py expected.tsv in.tsv model.pkl")
+        return
+    expected_file = str(sys.argv[1])
+    in_file = str(sys.argv[2])
+    model = str(sys.argv[3])
+    paranormal_class_logprob, sceptic_class_logprob = calc_class_logprob(expected_file)
+    bigrams_count = calc_bigram_count(in_file, expected_file)
+    bigram_logprobs = calc_bigram_logprobs(bigrams_count)
+    with open(model, 'wb') as f:
+        pickle.dump([paranormal_class_logprob, sceptic_class_logprob, bigram_logprobs],f)
 main()

train_baseline.py

@@ -0,0 +1,91 @@
+#!/usr/bin/python3
+from collections import defaultdict
+import math
+import pickle
+import re
+
+# in expected.tsv
+def calc_class_logprob(expected_path):
+    paranolal_classcount=0
+    sceptic_classcount=0
+    with open(expected_path) as f:
+        for line in f:
+            line = line.rstrip('\n').replace(' ','')
+            if 'P' in line:
+                paranolal_classcount +=1
+            elif 'S' in line:
+                sceptic_classcount +=1
+
+    paranol_prob = paranolal_classcount / (paranolal_classcount + sceptic_classcount)
+    sceptic_prob = sceptic_classcount / (paranolal_classcount + sceptic_classcount)
+
+    return math.log(paranol_prob), math.log(sceptic_prob)
+
+def clear_tokens(tokens, is_text=True):
+    tokens = tokens.replace('\\n', ' ')
+    return tokens
+    # delete links, special characters, kropki, and \n
+    tokens = re.sub(r'\(((http)|(https)).*((\.com)|(\.net)|(\.jpg)|(\.html))\)'," ", tokens)
+    tokens = re.sub(r'(|\-|\_)([a-z]+(\-|\_))+[a-z]+(|\-|\_)', ' ', tokens)
+    tokens = re.sub(r'[\n\&\"\?\\\'\*\[\]\,\;\.\=\+\(\)\!\/\:\`\~\%\^\$\#\@\’\＞\″\±]+', ' ', tokens)
+    tokens = re.sub(r'[\.\-][\.\-]+', ' ', tokens)
+    tokens = re.sub(r'[0-9]+', ' ', tokens)
+    tokens = re.sub(r'œ|·', '', tokens)
+    if is_text:
+        tokens = re.sub(r' +', ' ', tokens)
+    else:
+        tokens = re.sub(r' +', '', tokens)
+    return tokens
+
+# ile razy slowo wystepuje w dokumentach w danej klasie
+def calc_word_count(in_path, expected_path):
+    word_counts = {'paranormal':defaultdict(int), 'sceptic': defaultdict(int)} # dzienik zawierajacy slownik w ktorym s slowa i ile razy wystepuja
+    with open(in_path) as infile, open(expected_path)  as expectedfile:
+        for line, exp in zip(infile, expectedfile):
+            class_ = exp.rstrip('\n').replace(' ','')
+            text, timestap =line.rstrip('\n').split('\t')
+            #print(f"text  {type(text)}")
+            text = clear_tokens(text, True)
+            tokens = text.lower().split(' ')
+            #print(f"tokens {type(tokens)}")
+            for token in tokens:
+                clear_tokens(token,False)
+                if class_ == 'P':
+                    word_counts['paranormal'][token] += 1
+                elif class_ == 'S':
+                    word_counts['sceptic'][token]+=1
+
+    return word_counts
+
+def calc_word_logprobs(word_counts):
+    total_skeptic = sum(word_counts['sceptic'].values()) + len(word_counts['sceptic'].keys())
+    total_paranormal = sum(word_counts['paranormal'].values())+ len(word_counts['paranormal'].keys())
+    word_logprobs= {'paranormal': {}, 'sceptic': {}}
+    for class_ in word_counts.keys(): # sceptic paranormal
+        for token, value in word_counts[class_].items():
+            if class_ == 'sceptic':
+                word_prob = (value +1)/ total_skeptic
+            elif class_ == 'paranormal':
+                word_prob = (value+1)/ total_paranormal
+
+            #print (token)
+            word_logprobs[class_][token] = math.log(word_prob)
+
+    return word_logprobs
+
+def main():
+    expected = './train/expected.tsv'
+    #expected = './dev-0/expected.tsv'
+    in_f = './train/in.tsv'
+    #in_f = './dev-0/in.tsv'
+    print (f"expected {expected}")
+    print (f"in {in_f}")
+    paranormal_class_lgprob, skeptic_class_logprob = calc_class_logprob(expected)
+    wordcounts =calc_word_count(in_f,expected)
+
+    word_logprobs = calc_word_logprobs(wordcounts)
+    with open('naive_base_model.pkl', 'wb') as f:
+        pickle.dump([paranormal_class_lgprob, skeptic_class_logprob, word_logprobs], f)
+     # w predict.py bierzemy ten wzor argmax P(w) iloczynP(w|c)
+
+main()