Transfer theoretical course (CM) code snippets and examples to this repo

cm-code/examples-bio.py

+#!/usr/bin/env python3
+
+from lib.conllulib import CoNLLUReader
+test="""# global.columns = ID FORM parseme:ne
+1	Le	1:PROD
+2	Petit	1
+3	Prince	1
+4	de	*
+5	Saint-Exupéry	2:PERS
+6	est	*
+7	entré	*
+8	à	*
+9	l'	*
+10	École	3:ORG
+11	Jules-Romains	3"""
+for sent in CoNLLUReader.readConlluStr(test):
+  print(CoNLLUReader.to_bio(sent))
+#['B-PROD', 'I-PROD', 'I-PROD', 'O', 'B-PERS', 'O', 'O', 'O', 'O', 'B-ORG', 'I-ORG']
+  
+s1 = ["B-PERS", "I-PERS", "I-PERS", "O", "B-LOC", "I-LOC"]
+s2 = ["I-PERS", "B-PERS", "I-PERS", "O", "I-LOC"]
+print(CoNLLUReader.from_bio(s1, bio_style='bio'))
+# ['1:PERS', '1', '1', '*', '2:LOC', '2']
+print(CoNLLUReader.from_bio(s1, bio_style='io'))
+# WARNING: Got B tag in spite of 'io' bio_style: interpreted as I
+# WARNING: Got B tag in spite of 'io' bio_style: interpreted as I
+# ['1:PERS', '1', '1', '*', '2:LOC', '2']
+print(CoNLLUReader.from_bio(s2, bio_style='bio'))
+# WARNING: Invalid I-initial tag I-PERS converted to B
+# WARNING: Invalid I-initial tag I-LOC converted to B
+# ['1:PERS', '2:PERS', '2', '*', '3:LOC']
+print(CoNLLUReader.from_bio(s2, bio_style='io'))
+# WARNING: Got B tag in spite of 'io' bio_style: interpreted as I
+# ['1:PERS', '1', '1', '*', '2:LOC']

cm-code/examples-conllu.py

+#!/usr/bin/env python
+
+################################################################################
+### CM1 exercise to get familiar with conllu format and library
+
+import conllu, sys
+conllufile = open(sys.argv[1], 'r', encoding='UTF-8')
+slens, wlens = [], []
+for sent in conllu.parse_incr(conllufile):
+  slens.append(len(sent))  
+  wlens.extend([len(token['form']) for token in sent])
+print("Avg sent len={:.2f}".format(sum(slens)/len(slens)))
+print("Avg word len={:.2f}".format(sum(wlens)/len(wlens)))
+
+import matplotlib.pyplot as plt
+f,(a1,a2) = plt.subplots(1,2)
+a1.hist(slens,bins=20)
+a1.set_title("Sentence length")
+a2.hist(wlens,bins=20)
+a2.set_title("Word length")
+plt.show()
+
+################################################################################
+### CM3 exercise to manipulate morphological features
+
+from collections import Counter, defaultdict
+
+conllufile = open(sys.argv[1], 'r', encoding='UTF-8')
+slens, wlens = [], []
+feats_pos_dict = defaultdict(lambda: Counter())
+feats_dict = defaultdict(lambda: Counter())
+no_feat = total_words = total_features = 0
+
+for sent in conllu.parse_incr(conllufile):
+  for w in sent :
+    total_words += 1
+    if w["feats"] :
+      for (key,value) in w["feats"].items():
+        feats_pos_dict[w["upos"]][key] += 1
+        feats_dict[key][value] += 1
+        total_features += 1
+    else:
+      no_feat += 1
+      
+print(f"Number of feature keys: {len(feats_dict)}")
+for feat in feats_dict:
+  print(f"  {feat}: {list(feats_dict[feat])}")
+print(' ' * 9 + '|' + '|'.join([f"{pos:5}" for pos in feats_pos_dict])+'|')
+for feat in feats_dict:
+  print(f"{feat:9}|" + '|'.join([f"{feats_pos_dict[pos][feat] if feats_pos_dict[pos][feat] else ' ':5}" for pos in feats_pos_dict]) + '|')
+print(f"Words with no feature: {no_feat}/{total_words} ({no_feat/total_words*100:.2f}%)")
+print(f"Average features per word: {total_features/total_words:.2f}")
+
+
+

cm-code/examples-torch-cm1.py

+#!/usr/bin/env python
+
+################################################################################
+# defaultdict examples
+
+from collections import defaultdict
+word_count = defaultdict(lambda: 0) # return 0 if absent key
+sentence = "the man and the dog and the tree"
+for word in sentence.split():
+  word_count[word] += 1
+print(dict(word_count)) 
+# {'the': 3, 'man': 1, 'and': 2, 'dog': 1, 'tree': 1}
+print(word_count["hi"]) 
+# 0  
+print(dict(word_count)) 
+# {'the': 3, 'man': 1, 'and': 2, 'dog': 1, 'tree': 1, 'hi': 0}
+
+################################################################################
+# Embedding layer examples
+
+from torch import LongTensor
+import torch.nn as nn
+
+embed = nn.Embedding(20, 4, padding_idx=0)
+print(embed.weight[:2])
+# tensor([[ 0.0000,  0.0000,  0.0000,  0.0000],
+#        [-0.9336, -0.0982,  0.2726,  1.9872], grad_fn=<...>)
+print(embed(LongTensor([3,4])).shape)
+# torch.Size([2, 4])
+print(embed(LongTensor([3, 4, 3, 0])).shape)
+# torch.Size([4, 4])
+print(embed(LongTensor([[3, 4, 3], [0, 1, 1]])).shape) # batched
+# torch.Size([2, 3, 4])
+
+################################################################################
+# RNN unit examples
+
+gru = nn.GRU(4, 10, batch_first=True)
+emb = nn.Embedding(20, 4, padding_idx=0)
+x   = emb(LongTensor([[3, 4, 3, 2, 5],   # B=2 (batch)
+                      [1, 12, 1, 0, 0]])) # L=5 (timesteps)
+print(x.shape)
+# torch.Size([2, 5, 4])
+y = gru(x)
+print(y[0].shape)
+# torch.Size([2, 5, 10])
+print(y[1].shape)
+# torch.Size([1, 2, 10]) # notice batch is dim1
+
+################################################################################
+# TensorDataset/DataLoader examples
+
+from torch.utils.data import TensorDataset, DataLoader
+import torch
+x = torch.rand(7,3)
+y = LongTensor([[i] for i in range(7)])
+tds = TensorDataset(x,y)
+print(len(tds))
+# 7
+print([(e.shape, e.dtype) for e in tds[5]])
+# [(torch.Size([3]), torch.float32), (torch.Size([1]), torch.int64)]
+dl = DataLoader(tds, batch_size=2)
+print(next(iter(dl))[0].shape)
+#torch.Size([2, 3])
+print(next(iter(dl))[1].shape)
+#torch.Size([2, 1])
+for (bx, by) in dl :
+  print(bx.shape, by.shape)
+  
+################################################################################
+
+

cm-code/fibo.py

+#!/usr/bin/env python3
+
+n_appels_rec = n_appels_dynprog = 0
+
+def fiboRec(n):
+  global n_appels_rec
+  n_appels_rec += 1
+  if n <= 1:
+    return n
+  else:
+    return fiboRec(n-1) + fiboRec(n-2)
+
+################################################################################
+
+def fiboDynProg(n):
+  global n_appels_dynprog
+  fib = [0] * (n+1)
+  fib[1] = 1
+  n_appels_dynprog += 2
+  for i in range(2,n+1):
+    n_appels_dynprog += 1
+    fib[i] = fib[i-1] + fib[i-2]
+  return fib[n]
+
+for n in range(1,20):
+  n_appels_rec = n_appels_dynprog = 0
+  resultRec = fiboRec(n)
+  resultDynProg = fiboDynProg(n)
+  print(f"n={n:2} fib_rec(n)={resultRec:4}, n_appels_rec={n_appels_rec:5} fib_dp(n)={resultDynProg:4}, n_appels_dp={n_appels_dynprog:5}")

cm-code/lib

+../lib
\ No newline at end of file