initial commit

app.py

@@ -0,0 +1,98 @@
+import streamlit as st
+from presidio_analyzer import AnalyzerEngine
+from presidio_anonymizer import AnonymizerEngine
+from transformers import AutoTokenizer, AutoModel
+from torch.nn import functional as F
+import matplotlib.pyplot as plt
+import torch
+model = AutoModel.from_pretrained("aarnow/distilbert-base-uncased-1212-test")
+tokenizer = AutoTokenizer.from_pretrained("aarnow/distilbert-base-uncased-1212-test")
+
+
+with st.sidebar:
+    st.title('Technical Demonstration')
+    st.header('powered by rascal')
+    st.markdown('''
+    ## About
+    This is a tool that shows the classification and PII redaction capabilities of the auditory skills model.  PII redaction is powered by Microsoft's presidio tool and the text classification model is trained on a combination of synthetic and human annotated data from the HATCH (Helping Adults Talk to Children) Lab at Idaho State University.  Erber's Hierarchy is used to benchmark the text classification model.
+
+    ''')
+
+
+
+def main():
+    st.subheader("Enter Text for Evaluation")
+
+    sentence = st.text_input('Type text to classify below')
+    if sentence != "":
+        #with PII redacted
+        analyzer = AnalyzerEngine()
+        # Call analyzer to get results
+        results = analyzer.analyze(text=sentence,
+                                   language='en')
+
+        # Analyzer results are passed to the AnonymizerEngine for anonymization
+        anonymizer = AnonymizerEngine()
+        anonymized_text = anonymizer.anonymize(text=sentence,analyzer_results=results)
+        st.markdown("**Your text with PII redacted:**  "+anonymized_text.text)
+        st.text(results)
+        st.subheader("Classification Details")
+        #use classification model below
+        #sentence = 'My child is able to comprehend a voice when the TV is on'
+        labels = ['DETECTION', 'DISCRIMINATION', 'IDENTIFICATION','COMPREHENSION']
+
+        # run inputs through model and mean-pool over the sequence
+        # dimension to get sequence-level representations
+        inputs = tokenizer.batch_encode_plus([sentence] + labels,
+                                             return_tensors='pt',
+                                             pad_to_max_length=True)
+        input_ids = inputs['input_ids']
+        attention_mask = inputs['attention_mask']
+        output = model(input_ids, attention_mask=attention_mask)[0]
+        sentence_rep = output[:1].mean(dim=1)
+        label_reps = output[1:].mean(dim=1)
+
+        # now find the labels with the highest cosine similarities to
+        # the sentence
+        similarities = F.cosine_similarity(sentence_rep, label_reps)
+        closest = similarities.argsort(descending=True)
+        st.markdown("The classification that best fits your entry is: "+labels[closest[0]])
+
+
+        #map the labels
+        tensor_datalbl = label_reps.detach()
+        x_values = tensor_datalbl[:, 0].numpy()
+        y_values = tensor_datalbl[:, 1].numpy()
+
+        # Create a scatter plot for labels
+        plt.scatter(x_values, y_values)
+
+        # Add labels to specific points (adjust indices as needed)
+        for i in range(len(tensor_datalbl)):
+            plt.text(x_values[i], y_values[i], str(labels[i]), fontsize=8, ha='right', va='bottom')
+
+
+        #map the sentence
+        tensor_datasen = sentence_rep.detach()
+
+        # Extract the individual dimensions for the scatter plot
+        x_values = tensor_datasen[:, 0].numpy()
+        y_values = tensor_datasen[:, 1].numpy()
+
+        plt.scatter(x_values, y_values)
+
+        plt.title('2D Representation of Similarity Estimates (2D)')
+        plt.xlabel('X-axis')
+        plt.ylabel('Y-axis')
+        #plt.show()
+        plt.savefig('foo.png', bbox_inches='tight')
+        st.image("foo.png")
+        st.subheader("Classification Details")
+        for ind in closest:
+            #print(f'label: {labels[ind]} \t similarity: {similarities[ind]}')
+            st.write(f'label: {labels[ind]} \t similarity: {similarities[ind]}')
+
+
+#run main
+if __name__ == '__main__':
+    main()