Create app.py

app.py

@@ -0,0 +1,101 @@
+import gradio as gr
+from transformers import pipeline
+import torch
+import pandas as pd
+from openprompt.plms import load_plm
+from openprompt import PromptDataLoader
+from openprompt.prompts import ManualVerbalizer
+from openprompt.prompts import ManualTemplate
+from openprompt.data_utils import InputExample
+from openprompt import PromptForClassification
+
+# def sentiment_analysis(sentence, template, positive, neutral, nagetive ):
+#     model_name = "bert-base-uncased" 
+#     nlp = pipeline("fill-mask", model=model_name)
+#     text = template.replace("[SENTENCE]",sentence)
+#     # return str(nlp(text))
+#     return sentence + template + positive + neutral + nagetive
+
+# demo = gr.Interface(fn=sentiment_analysis,
+#                     inputs = [gr.Textbox(placeholder="Enter sentence here.",label="sentence"),
+#                             gr.Textbox(placeholder="Your template must have a [SENTENCE] token and a [MASK] token.",label="template"),
+#                             gr.Textbox(placeholder="Separate words with Spaces.",label="positive"),
+#                             gr.Textbox(placeholder="Separate words with Spaces.",label="neutral"),
+#                             gr.Textbox(placeholder="Separate words with Spaces.",label="negative")
+#                             ],
+#                     outputs="text")
+    
+# demo.launch(server_port=8080)  
+
+def readLMwords():
+    alldata = pd.read_csv("LoughranMcDonald_MasterDictionary_2020.csv")
+    positive = list(alldata[alldata["Positive"]!=0]["Word"].str.lower())
+    negative = list(alldata[alldata["Negative"]!=0]["Word"].str.lower())
+    uncertainty = list(alldata[alldata["Uncertainty"]!=0]["Word"].str.lower())
+    return positive,negative,uncertainty
+
+
+def sentiment_analysis(sentence, model_name):
+    model_name = model_name
+    sentences = sentence.split('\n')
+    template = '{"placeholder":"text_a"} Shares are {"mask"}.'
+    classes = ['positive', 'neutral', 'negative']
+    positive,negative,neutral = readLMwords()
+    label_words = {
+        "positive": positive,
+        "neutral": neutral,
+        "negative": negative,
+    }
+    type_dic = {
+        "CCCC/ARCH_tuned_bert":"bert",
+        "bert-base-uncased":"bert",
+        "roberta-base":"roberta",
+        "yiyanghkust/finbert-pretrain":"bert",
+        "facebook/opt-125m":"opt",
+        "facebook/opt-350m":"opt",
+    }
+    testdata = []
+    for i,sentence in enumerate(sentences):
+        testdata.append(InputExample(guid=i,text_a=sentence,label=0))
+
+    plm, tokenizer, model_config, WrapperClass = load_plm(type_dic[model_name], model_name)
+
+    promptTemplate = ManualTemplate(
+        text = template,
+        tokenizer = tokenizer,
+    )
+    promptVerbalizer = ManualVerbalizer(
+        classes = classes,
+        label_words = label_words,
+        tokenizer = tokenizer,
+    )
+    test_dataloader = PromptDataLoader(
+        dataset = testdata,
+        tokenizer = tokenizer, 
+        template = promptTemplate, 
+        tokenizer_wrapper_class = WrapperClass,
+        batch_size = 4,
+        max_seq_length = 512,
+    )
+    prompt_model = PromptForClassification(
+        plm=plm,
+        template=promptTemplate, 
+        verbalizer=promptVerbalizer, 
+        freeze_plm=False #whether or not to freeze the pretrained language model
+    )
+    result = []
+    for step, inputs in enumerate(test_dataloader):
+        logits = prompt_model(inputs)
+        result.extend(torch.argmax(logits, dim=-1))
+    output = '\n'.join([classes[i] for i in result])
+    return output #classes[torch.argmax(logits, dim=-1)[0]]
+
+
+demo = gr.Interface(fn=sentiment_analysis,
+                    inputs = [gr.Textbox(placeholder="Enter sentence here.",label="sentence",lines=5),
+                            gr.Radio(choices=["ARCH_tuned_robert","FNCH_tuned_robert","AREN_tuned_robert",
+                                              "FNEN_tuned_robert","bert-base-uncased"], label="model choics")],
+                    outputs=gr.Textbox(label="output",lines=5),
+        )
+    
+demo.launch()