Update app.py

app.py

@@ -7,18 +7,12 @@ import os
 import gradio as gr
 import requests
 
+# just for the sake of this demo, we use cloze prob to initialize the hypothesis 
+
 #url = "https://github.com/simonepri/lm-scorer/tree/master/lm_scorer/models"
 #resp = requests.get(url)
 
 from sentence_transformers import SentenceTransformer, util
-#from sentence_transformers import SentenceTransformer, util
-#from sklearn.metrics.pairwise import cosine_similarity
-#from lm_scorer.models.auto import AutoLMScorer as LMScorer
-#from sentence_transformers import SentenceTransformer, util
-#from sklearn.metrics.pairwise import cosine_similarity
-
-
-#model_sts = gr.Interface.load('huggingface/sentence-transformers/stsb-distilbert-base') 
 
 model_sts = SentenceTransformer('stsb-distilbert-base')
 #model_sts = SentenceTransformer('roberta-large-nli-stsb-mean-tokens')
@@ -68,7 +62,6 @@ tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
 def cloze_prob(text):
 
 	whole_text_encoding = tokenizer.encode(text)
-	# Parse out the stem of the whole sentence (i.e., the part leading up to but not including the critical word)
 	text_list = text.split()
 	stem = ' '.join(text_list[:-1])
 	stem_encoding = tokenizer.encode(stem)
@@ -80,7 +73,6 @@ def cloze_prob(text):
 		predictions = outputs[0]   
 
 	logprobs = []
-	# start at the stem and get downstream probabilities incrementally from the model(see above)
 	start = -1-len(cw_encoding)
 	for j in range(start,-1,1):
 			raw_output = []
@@ -93,8 +85,7 @@ def cloze_prob(text):
 	conditional_probs = []
 	for cw,prob in zip(cw_encoding,logprobs):
 			conditional_probs.append(prob[cw])
-	# now that you have all the relevant probabilities, return their product.
-	# This is the probability of the critical word given the context before it.
+
 
 	return np.exp(np.sum(conditional_probs))
 
@@ -106,38 +97,7 @@ def cos_sim(a, b):
     return np.inner(a, b) / (np.linalg.norm(a) * (np.linalg.norm(b)))
 
 
-  
-#def Visual_re_ranker(caption, visual_context_label, visual_context_prob):
-#def Visual_re_ranker(caption_man, caption_woman, visual_context_label, visual_context_prob):
-#    caption_man = caption_man  
-#    caption_woman = caption_woman
-#    visual_context_label= visual_context_label
-#    visual_context_prob = visual_context_prob
-#    caption_emb_man = model_sts.encode(caption_man, convert_to_tensor=True)
-#    caption_emb_woman = model_sts.encode(caption_woman, convert_to_tensor=True)
-#    visual_context_label_emb = model_sts.encode(visual_context_label, convert_to_tensor=True)
-
-#    sim_m =  cosine_scores = util.pytorch_cos_sim(caption_emb_man, visual_context_label_emb)
-#    sim_m = sim_m.cpu().numpy()
-#    sim_m = get_sim(sim_m)
-
-#    sim_w = cosine_scores = util.pytorch_cos_sim(caption_emb_woman, visual_context_label_emb) 
-#    sim_w = sim_w.cpu().numpy()
-#    sim_w = get_sim(sim_w)
-
-
-#    LM_man = cloze_prob(caption_man)
-#    LM_woman = cloze_prob(caption_woman)
-    #LM  = scorer.sentence_score(caption, reduce="mean")
-#    score_man     = pow(float(LM_man),pow((1-float(sim_m))/(1+ float(sim_m)),1-float(visual_context_prob)))
-#    score_woman   = pow(float(LM_woman),pow((1-float(sim_w))/(1+ float(sim_w)),1-float(visual_context_prob)))
-
-
-
 
-    #return {"LM": float(LM)/1, "sim": float(sim)/1, "score": float(score)/1 }
-#    return {"Man": float(score_man)/1, "Woman": float(score_woman)/1}
-    #return LM, sim, score 
 
 def Visual_re_ranker(caption_man, caption_woman, context_label, context_prob):
     caption_man = caption_man  
@@ -178,7 +138,7 @@ demo = gr.Interface(
     description="Demo for Women Wearing Lipstick: Measuring the Bias Between Object and Its Related Gender",
     inputs=[gr.Textbox(value="a man riding a motorcycle on a road") , gr.Textbox(value="a woman riding a motorcycle on a road"), gr.Textbox(value="motor scooter"),  gr.Textbox(value="0.2183")],
     
-     #inputs=[gr.Textbox(value="a man is blow drying his hair in the bathroom") , gr.Textbox(value="a woman is blow drying her hair in the bathroom"), gr.Textbox(value="hair spray"),  gr.Textbox(value="0.7385")],
+  
     
     outputs="label",
 )