Unbias - Version one push

app.py

@@ -1,7 +1,30 @@
+import combine_modules
 import gradio as gr
 
-def greet(name):
-    return "Hello " + name + "!!"
+with gr.Blocks() as demo:
+    gr.Markdown("Welcome to Bias Checking Demo !")
+    with gr.Tab("Text"):
+        text_input = gr.Textbox()
+        text_output = gr.Textbox()
+        text_output_interpret = gr.TextArea()
+        text_button = gr.Button("Check Bias in your Text")
+    
+    with gr.Tab("Video"):
+        vid_input = gr.Video()
+        vid_output = gr.Textbox()
+        vid_output_interpret = gr.TextArea()
+        vid_button = gr.Button("Check Bias in your Video")
 
-iface = gr.Interface(fn=greet, inputs="text", outputs="text")
-iface.launch()
+    with gr.Tab("Image"):
+        img_input = gr.Image()
+        img_output = gr.Textbox()
+        img_output_interpret = gr.TextArea()
+        img_button = gr.Button("Check Bias in your Image")
+
+    text_button.click(combine_modules.app_nlp_start, inputs=text_input, outputs=[text_output, text_output_interpret])
+    vid_button.click(combine_modules.app_video_start, inputs=vid_input, outputs=[vid_output, vid_output_interpret])
+    img_button.click(combine_modules.app_image_start, inputs=img_input, outputs=[img_output, img_output_interpret])
+
+
+if __name__=="__main__":
+    demo.launch()

bias_check.py

@@ -0,0 +1,47 @@
+import load_model_pt
+import interpret_model_pt
+
+
+def sub_pipeline(raw_input, pretrained_model):
+    tokenizer, model = load_model_pt.load_models_from_pretrained(pretrained_model)
+    output_ = load_model_pt.load_pipeline(raw_input, pretrained_model)
+    words_weightages = interpret_model_pt.explainer(raw_input, model, tokenizer)
+    return output_, words_weightages
+
+def bias_checker(input_statement):
+    pretrained_model_basic_check = "valurank/distilroberta-bias"
+    pretrained_model_political = "valurank/distilroberta-mbfc-bias"
+    pretrained_model_gender = "monologg/koelectra-base-v3-gender-bias"
+
+    raw_input = input_statement
+    # print("Checking if the input has any primary bias ?..")
+    output_stmt_zero, words_interpreted = sub_pipeline(raw_input, pretrained_model_basic_check)
+    print(output_stmt_zero)
+    return_var = " "
+    interpret_var = " "
+
+    if (output_stmt_zero["label"] == "BIASED" and output_stmt_zero["score"] >= 0.7) or (output_stmt_zero["label"] == "NEUTRAL" and output_stmt_zero["score"] < 0.6):
+        # print(output_stmt_zero)
+        # print("\n The statement seems biased, lets investigate ! \n")
+        # print(words_interpreted)
+        # print("\n Checking for political propaganda... \n")
+        output_stmt_political, words_interpreted_political = sub_pipeline(raw_input, pretrained_model_political)
+        # print(output_stmt_political, "\n")
+        # print(words_interpreted_political, "\n")
+        # print("\n Let's check for gender bias, shall we ? \n")
+        output_stmt_gender, words_interpreted_gender = sub_pipeline(raw_input, pretrained_model_gender)
+        # print(output_stmt_gender, "\n")
+        # print(words_interpreted_gender, "\n")
+        return_var = ("Generic:", output_stmt_zero,"\n","Gender:", output_stmt_gender,"\n","Political:", output_stmt_political)
+        interpret_var = ("Generic:", words_interpreted, "\n", "Gender:", words_interpreted_gender, "\n","Political:", words_interpreted_political)
+    else:
+        # print("The statement seems ok as of now, please input another statement!")
+        return_var = "The statement seems ok as of now, please input another statement!"
+        interpret_var = " "
+
+    return return_var, interpret_var
+
+
+if __name__=="__main__":
+    input_stmt = "Nevertheless, Trump and other Republicans have tarred the protests as havens for terrorists intent on destroying property."
+    bias_checker(input_stmt)

combine_modules.py

@@ -0,0 +1,17 @@
+from bias_check import bias_checker
+from video2txt import read_video_file_and_return_text as rtxt
+from img2txt import load_image_pipeline as img_pipe
+
+def app_nlp_start(statement):
+    output_txt, interpreted_txt = bias_checker(statement)
+    return output_txt, interpreted_txt
+
+def app_video_start(video_path):
+    return_text = rtxt(video_path)
+    output_txt = bias_checker(return_text)
+    return output_txt
+
+def app_image_start(image_path):
+    text_generated = img_pipe(image_path)
+    output_txt = bias_checker(text_generated)
+    return output_txt

img2txt.py

@@ -0,0 +1,19 @@
+from transformers import pipeline
+from PIL import Image
+import os
+pretrained_img_model = "nlpconnect/vit-gpt2-image-captioning"
+
+
+def load_image_pipeline(img_path):
+    img_path_read = Image.fromarray(img_path) 
+    img_path_read.save("temp_img.jpg")
+    image_to_text = pipeline("image-to-text", model=pretrained_img_model, framework="pt")
+    generated_text = image_to_text("temp_img.jpg")[0]["generated_text"]
+    os.remove("temp_img.jpg")
+    return generated_text
+
+
+if __name__=="__main__":
+    imgpath = r"C:\Users\Shringar\Pictures\ar.jpg"
+    img_text_generated = load_image_pipeline(imgpath)
+    print(img_text_generated)

interpret_model_pt.py

@@ -0,0 +1,8 @@
+from transformers_interpret import SequenceClassificationExplainer, MultiLabelClassificationExplainer
+
+def explainer(input_statement, model, tokenizer):
+    # cls_explainer = SequenceClassificationExplainer(model, tokenizer)
+    cls_explainer = MultiLabelClassificationExplainer(model, tokenizer)
+    word_attributions = cls_explainer(input_statement)
+    return dict(word_attributions)
+    

load_model_pt.py

@@ -0,0 +1,15 @@
+from transformers import pipeline
+from transformers import AutoTokenizer
+from transformers import AutoModelForSequenceClassification
+
+
+def load_pipeline(input_statement, pretrained_model_name):
+    classifier_ = pipeline("text-classification", model=pretrained_model_name, framework="pt")
+    cls_output = classifier_(input_statement)[0]    
+    return cls_output
+
+def load_models_from_pretrained(checkpoint):
+    checkpoint_local = checkpoint
+    tokenizer = AutoTokenizer.from_pretrained(checkpoint_local)
+    model = AutoModelForSequenceClassification.from_pretrained(checkpoint_local)
+    return tokenizer, model

video2txt.py

@@ -0,0 +1,22 @@
+import speech_recognition as sr
+import moviepy.editor as mp
+import os
+
+
+def convert_video_to_audio(file_path):
+    vid_clip = mp.VideoFileClip(file_path)
+    vid_clip.audio.write_audiofile("temp_converted_mp3.wav")
+
+def convert_audio_to_text():
+    recogniser = sr.Recognizer()
+    audio_file = sr.AudioFile(r"temp_converted_mp3.wav")
+    with audio_file as source:
+        audio_source = recogniser.record(source)
+    audio_result = recogniser.recognize_google(audio_source)
+    os.remove("temp_converted_mp3.wav")
+    return audio_result
+
+def read_video_file_and_return_text(filepath=r"C:\Users\Shringar\Documents\Python Scripts\hface\course\emma_1.mp4"):
+    convert_video_to_audio(filepath)
+    converted_text = convert_audio_to_text()
+    return converted_text