roberta base noft

roberta_base_no_ft/04_predict.py

@@ -0,0 +1,48 @@
+from tqdm import tqdm
+from transformers import pipeline
+
+def get_formatted(text):
+    answers = unmasker(text, top_k=15)
+    answers = {x['token_str']:x['score'] for x in answers}
+    empty = 1 - sum(answers.values())
+    answers[''] = empty
+    answers_str =''
+    for k,v in answers.items():
+        answers_str += k.strip()+':'+str(v) + ' '
+    return answers_str.rstrip(' ').lstrip(' ')
+    
+def write(f_path_in, f_path_out):
+    with open(f_path_in) as f_in, open(f_path_out,'w') as f_out:
+        i = 0 
+        for line in tqdm(f_in,total=10_600):
+            char_context = 400
+            i+=1
+            #print(i)
+            is_ok = False
+            while not is_ok:
+                try:
+                    left_text = line.rstrip().split('\t')[-2]
+                    right_text = line.rstrip().split('\t')[-1]
+                    l_in = left_text[-char_context:] + ' <mask> ' + right_text[:char_context]
+                    a = get_formatted(l_in)
+                    is_ok = True
+                except:
+                    print('lowering context')
+                    char_context -= 50 
+                    if char_context < 60:
+                        a = ':1'
+                        print('lower threshold context exceeded')
+                        is_ok = True
+
+            f_out.write(a + '\n')
+            #left_text = line.rstrip().split('\t')[-2]
+            #right_text = line.rstrip().split('\t')[-1]
+            #l_in = left_text[-char_context:] + ' <mask> ' + right_text[:char_context]
+            #a = get_formatted(l_in)
+
+            #f_out.write(a + '\n')
+
+model = 'roberta-base'
+unmasker = pipeline('fill-mask', model=model, device=0)
+write('../dev-0/in.tsv', '../dev-0/out.tsv')
+write('../test-A/in.tsv', '../test-A/out.tsv')