Create app.py

This work is based on the research work presented in the papers “Unsupervised law article mining based on deep pre-trained language representation models with application to the Italian civil code” , “LamBERTa: Law Article Mining Based on Bert Architecture for the Italian Civil Code” and “Exploring domain and task adaptation of LamBERTa models for article retrieval on the Italian Civil Code” [90]

app.py

@@ -0,0 +1,69 @@
+import pandas as pd
+import re
+import gradio as gr
+import torch
+from transformers import BertTokenizerFast, BertForSequenceClassification
+
+if torch.cuda.is_available():
+    device = torch.device("cuda")
+    print('There are %d GPU(s) available.' % torch.cuda.device_count())
+    print('We will use the GPU:', torch.cuda.get_device_name(0))
+else:
+    print('No GPU available, using the CPU instead.')
+    device = torch.device("cpu")
+
+dataset_path = './codice_civile_ITA_LIBRI_2_withArtRef_v2.csv'
+input_model_path = './MODELLO_LOCALE_LIBRI_2_v5_2_subset60UniRRemphT4'
+
+
+def load_CC_from_CSV(path):
+  NUM_ART = 0
+  cc = pd.read_csv(path, header=None, sep='|', usecols=[1,2,3], names=['art','title','text'], engine='python')
+  article_id={}
+  id_article={}
+  article_text={}
+  for i in range(len(cc)):
+    NUM_ART +=1
+    art = re.sub('(\s|\.|\-)*', '', str(cc['art'][i]).lower())
+    article_id[art] = i
+    id_article[i] = art
+    article_text[art] = str(cc['title'][i]).lower() + " -> " + str(cc['text'][i]).lower()
+    if i == 59:
+        break
+  return article_id, id_article, article_text, NUM_ART
+
+article_id, id_article, article_text, NUM_ART = load_CC_from_CSV(dataset_path)
+
+model = BertForSequenceClassification.from_pretrained(input_model_path)
+tokenizer = BertTokenizerFast.from_pretrained(input_model_path)
+
+def LamBERTa_v5_placeholder(query):
+    n = 345
+    predictions = torch.softmax(torch.randn(n), dim=0)
+    values, indices = torch.topk(predictions, 5)
+    confidences = {id_article[i.item()] : v.item() for i, v in zip(indices, values)}
+    # confidences = {id_article[i] : float(predictions[i]) for i in range(n)}
+    return confidences
+
+def LamBERTa(query):
+    texts = []
+    input_ids = torch.tensor(tokenizer.encode(query, add_special_tokens=True)).unsqueeze(0)  # Batch size 1
+    labels = torch.tensor([1]).unsqueeze(0)  # Batch size 1
+    outputs = model(input_ids, labels=labels)
+    loss, logits = outputs[:2]
+    log_probs = torch.softmax(logits, dim=1)
+    values, indices = torch.topk(log_probs, 5, dim=1)
+    confidences = {id_article[i.item()] : v.item() for i, v in zip(indices[0], values[0])}
+    for art, prob in confidences.items():
+        texts.append(
+            {
+                "art": art,
+                "text": article_text[art],
+            }
+        )
+    return confidences, texts
+
+demo = gr.Interface(fn=LamBERTa, inputs="text", outputs=["label", "json"], examples=["Quando si apre la successione","Dove si apre la successione","In quali casi, alla morte, non spetta l'eredità"], live=True)
+
+demo.launch()
+demo.launch(share=True)