python 3

hmm_class/frost.py

@@ -2,17 +2,36 @@
 # https://udemy.com/unsupervised-machine-learning-hidden-markov-models-in-python
 # http://lazyprogrammer.me
 # Model and generate Robert Frost poems.
+from __future__ import print_function, division
+from future.utils import iteritems
+from builtins import range
+# Note: you may need to update your version of future
+# sudo pip install -U future
+
+
 
 import numpy as np
 import string
+import sys
+
 
 initial = {} # start of a phrase
 second_word = {}
 transitions = {}
 
-def remove_punctuation(s):
+# unfortunately these work different ways
+def remove_punctuation_2(s):
     return s.translate(None, string.punctuation)
 
+def remove_punctuation_3(s):
+    return s.translate(str.maketrans('','',string.punctuation))
+
+if sys.version.startswith('2'):
+    remove_punctuation = remove_punctuation_2
+else:
+    remove_punctuation = remove_punctuation_3
+
+
 def add2dict(d, k, v):
     if k not in d:
         d[k] = []
@@ -22,7 +41,7 @@ def add2dict(d, k, v):
     tokens = remove_punctuation(line.rstrip().lower()).split()
 
     T = len(tokens)
-    for i in xrange(T):
+    for i in range(T):
         t = tokens[i]
         if i == 0:
             # measure the distribution of the first word
@@ -43,7 +62,7 @@ def add2dict(d, k, v):
 
 # normalize the distributions
 initial_total = sum(initial.values())
-for t, c in initial.iteritems():
+for t, c in iteritems(initial):
     initial[t] = c / initial_total
 
 def list2pdict(ts):
@@ -52,15 +71,15 @@ def list2pdict(ts):
     n = len(ts)
     for t in ts:
         d[t] = d.get(t, 0.) + 1
-    for t, c in d.iteritems():
+    for t, c in iteritems(d):
         d[t] = c / n
     return d
 
-for t_1, ts in second_word.iteritems():
+for t_1, ts in iteritems(second_word):
     # replace list with dictionary of probabilities
     second_word[t_1] = list2pdict(ts)
 
-for k, ts in transitions.iteritems():
+for k, ts in iteritems(transitions):
     transitions[k] = list2pdict(ts)
 
 # generate 4 lines
@@ -69,14 +88,14 @@ def sample_word(d):
     p0 = np.random.random()
     # print "p0:", p0
     cumulative = 0
-    for t, p in d.iteritems():
+    for t, p in iteritems(d):
         cumulative += p
         if p0 < cumulative:
             return t
     assert(False) # should never get here
 
 def generate():
-    for i in xrange(4):
+    for i in range(4):
         sentence =[]
 
         # initial word
@@ -95,7 +114,7 @@ def generate():
             sentence.append(w2)
             w0 = w1
             w1 = w2
-        print ' '.join(sentence)
+        print(' '.join(sentence))
 
 generate()
 

hmm_class/generate_ht.py

@@ -2,6 +2,13 @@
 # https://udemy.com/unsupervised-machine-learning-hidden-markov-models-in-python
 # http://lazyprogrammer.me
 # Generate discrete data from an HMM.
+from __future__ import print_function, division
+from future.utils import iteritems
+from builtins import range
+# Note: you may need to update your version of future
+# sudo pip install -U future
+
+
 import numpy as np
 
 
@@ -16,7 +23,7 @@ def generate_sequence(N):
     s = np.random.choice(xrange(M), p=pi) # initial state
     x = np.random.choice(xrange(V), p=B[s]) # initial observation
     sequence = [x]
-    for n in xrange(N-1):
+    for n in range(N-1):
         s = np.random.choice(xrange(M), p=A[s]) # next state
         x = np.random.choice(xrange(V), p=B[s]) # next observation
         sequence.append(x)
@@ -25,7 +32,7 @@ def generate_sequence(N):
 
 def main():
     with open('coin_data.txt', 'w') as f:
-        for n in xrange(50):
+        for n in range(50):
             sequence = generate_sequence(30)
             sequence = ''.join(symbol_map[s] for s in sequence)
             print sequence

hmm_class/hmm_classifier.py

@@ -2,11 +2,19 @@
 # https://udemy.com/unsupervised-machine-learning-hidden-markov-models-in-python
 # http://lazyprogrammer.me
 # Demonstrate how HMMs can be used for classification.
+from __future__ import print_function, division
+from future.utils import iteritems
+from builtins import range
+# Note: you may need to update your version of future
+# sudo pip install -U future
+
+
+
 import string
 import numpy as np
 import matplotlib.pyplot as plt
 
-from hmmd_theano import HMM
+from hmmd_theano2 import HMM
 from sklearn.utils import shuffle
 from nltk import pos_tag, word_tokenize
 
@@ -18,14 +26,14 @@ def fit(self, X, Y, V):
         K = len(set(Y)) # number of classes - assume 0..K-1
         self.models = []
         self.priors = []
-        for k in xrange(K):
+        for k in range(K):
             # gather all the training data for this class
             thisX = [x for x, y in zip(X, Y) if y == k]
             C = len(thisX)
             self.priors.append(np.log(C))
 
             hmm = HMM(5)
-            hmm.fit(thisX, V=V, p_cost=0.1, print_period=1, learning_rate=1e-4, max_iter=100)
+            hmm.fit(thisX, V=V, print_period=1, learning_rate=1e-2, max_iter=80)
             self.models.append(hmm)
 
     def score(self, X, Y):
@@ -56,7 +64,7 @@ def get_data():
         for line in open(fn):
             line = line.rstrip()
             if line:
-                print line
+                print(line)
                 # tokens = remove_punctuation(line.lower()).split()
                 tokens = get_tags(line)
                 if len(tokens) > 1:
@@ -69,26 +77,25 @@ def get_data():
                     X.append(sequence)
                     Y.append(label)
                     count += 1
-                    print count
+                    print(count)
                     if count >= 50:
                         break
-    print "Vocabulary:", word2idx.keys()
+    print("Vocabulary:", word2idx.keys())
     return X, Y, current_idx
 
 
 def main():
     X, Y, V = get_data()
-    # print "Finished loading data"
-    print "len(X):", len(X)
-    print "Vocabulary size:", V
+    print("len(X):", len(X))
+    print("Vocabulary size:", V)
     X, Y = shuffle(X, Y)
     N = 20 # number to test
     Xtrain, Ytrain = X[:-N], Y[:-N]
     Xtest, Ytest = X[-N:], Y[-N:]
 
     model = HMMClassifier()
     model.fit(Xtrain, Ytrain, V)
-    print "Score:", model.score(Xtest, Ytest)
+    print("Score:", model.score(Xtest, Ytest))
 
 
 if __name__ == '__main__':