hacky bp generation stuff for anyone who wants it

Specific_Language_Generation.py

@@ -0,0 +1,93 @@
+### hacky stuff in the name of sharing ###
+
+import random
+import string
+from Training_Functions import make_train_set_for_target
+
+bp_other_letters = string.ascii_lowercase  #probably avoid putting '$' in here because that's my dummy letter somewhere in the network (todo: something more general)
+alphabet_bp = "()"+bp_other_letters
+
+
+def make_similar(w, alphabet):
+    new = list(w)
+    indexes = list(range(len(new)))
+    # switch characters
+    num_switches = random.choice(range(3))
+    shuffle(indexes)
+    indexes_to_switch = indexes[:num_switches]
+    for i in indexes_to_switch:
+        new[i] = random.choice(alphabet)
+    # insert characters
+    num_inserts = random.choice(range(3))
+    indexes = indexes + [len(new)]
+    indexes_to_insert = indexes[:num_inserts]
+    for i in indexes_to_insert:
+        new = new[:i] + [random.choice(alphabet)] + new[i:]
+    num_changes = num_switches + num_inserts
+    # duplicate letters
+    while ((num_changes == 0) or (random.choice(range(3)) == 0)) and len(new) > 0:
+        index = random.choice(range(len(new)))
+        new = new[:index + 1] + new[index:]
+        num_changes += 1
+    # omissions
+    while ((num_changes == 0) or random.choice(range(3)) == 0) and len(new) > 0:
+        index = random.choice(range(len(new)))
+        new = new[:index] + new[index + 1:]
+        num_changes +=1
+    return ''.join(new)
+
+def balanced_parantheses(w):
+    open_counter = 0
+    while len(w)>0:
+        c = w[0]
+        w = w[1:]
+        if c == "(":
+            open_counter += 1
+        elif c==")":
+            open_counter -= 1
+            if open_counter <0:
+                return False
+    return open_counter == 0
+
+def random_balanced_word(start_closing):
+    count = 0
+    word = ""
+    while len(word)<start_closing:
+        paran = (random.choice(range(3))==0)
+        next_letter = random.choice("()") if paran else random.choice(bp_other_letters)
+        if next_letter == ")" and count <= 0:
+            continue
+        word+=next_letter
+        if next_letter == "(":
+            count += 1
+        if next_letter == ")":
+            count -= 1
+    while True:
+        paran = (random.choice(range(3))==0)
+        next_letter = random.choice(")") if paran else random.choice(bp_other_letters)
+        if next_letter == ")":
+            count -= 1
+            if count<0:
+                break
+        word+=next_letter
+    return word
+
+def n_balanced_words_around_lengths(n,short,longg):
+    words = set()
+    while len(words)<n:
+        for l in range(short,longg):
+            words.add(random_balanced_word(l))
+#     print('\n'.join(sorted(list(words),key=len)))
+    return words
+
+def get_balanced_parantheses_train_set(n,short,longg,lengths=None,max_train_samples_per_length=300,search_size_per_length=200): # eg 15000, 2, 30
+    balanced_words = list(n_balanced_words_around_lengths(n,short,longg))
+    almost_balanced = [make_similar(w,alphabet_bp) for w in balanced_words][:int(2*n/3)]
+    less_balanced = [make_similar(w,alphabet_bp) for w in almost_balanced]
+    barely_balanced = [make_similar(w,alphabet_bp) for w in less_balanced]
+
+    all_words = balanced_words + almost_balanced + less_balanced + barely_balanced
+    return make_train_set_for_target(balanced_parantheses,alphabet_bp,lengths=lengths,\
+        max_train_samples_per_length=max_train_samples_per_length,\
+        search_size_per_length=search_size_per_length,\
+        provided_examples=all_words):