add barplot

app.py

@@ -1,5 +1,7 @@
 import gradio as gr
+import numpy as np
 import pandas as pd
+import plotly.express as px
 import torch
 from transformers import AutoTokenizer, AutoModelForMaskedLM
 
@@ -18,8 +20,9 @@ def add_mask(target_word, text):
 
 def eval_prob(target_word, text):
     text_mask = add_mask(target_word, text)
+
     # Get index of target_word
-    idx = tokenizer.encode(target_word)[2]
+    target_idx = tokenizer.encode(target_word)[2]
 
     # Get logits
     inputs = tokenizer(text_mask, return_tensors="pt")
@@ -34,54 +37,54 @@ def eval_prob(target_word, text):
     probs = torch.nn.functional.softmax(torch.tensor([logits]), dim=1)[0]
 
     # Get probability of target word filling the MASK
-    result = float(probs[idx])
+    # result = float(probs[target_idx])
+
+    return probs, target_idx
 
-    return result
 
+def plot_results(target_word, text):
+    probs, target_idx = eval_prob(target_word, text)
 
-# test barplot
-simple = pd.DataFrame(
-    {
-        "item": ["A", "B", "C", "D", "E", "F", "G", "H", "I"],
-        "inventory": [28, 55, 43, 91, 81, 53, 19, 87, 52],
-    }
-)
+    # Sort tokens based on probability scores
+    words = [
+        tokenizer.decode(idx) for idx in torch.sort(probs, descending=True).indices
+    ]
+    scores = torch.sort(probs, descending=True).values
 
-css = (
-    "footer {display: none !important;} .gradio-container {min-height: 0px !important;}"
-)
+    # Consolidate results in dataframe
+    d = {"word": words, "score": scores}
+    df = pd.DataFrame(data=d)
 
-with gr.Blocks(css=css) as demo:
-    gr.BarPlot(value=simple, x="item", y="inventory", title="Simple Bar Plot").style(
-        container=False,
+    # Get score rank of target word
+    result_rank = words.index(target_word)
+    target_col = [0] * len(scores)
+    target_col[result_rank] = 1
+    df["target"] = target_col
+
+    # Plot
+    fig = px.bar(
+        df[:100],
+        x="word",
+        y="score",
+        color="target",
+        color_continuous_scale=px.colors.sequential.Bluered,
     )
-demo.launch(share=True)
+    # fig.update(layout_coloraxis_showscale=False)
+    fig.show()
+    return fig
 
 
 gr.Interface(
-    fn=eval_prob,
+    fn=plot_results,
     inputs=[
-        gr.Textbox(label="词语", placeholder="夸大"),
-        gr.Textbox(label="造句", placeholder=f"我们使用生成式人工智能已经很长时间了，所以最近的媒体报道可能被夸大了。"),
+        gr.Textbox(label="词语", placeholder="标准"),
+        gr.Textbox(label="造句", placeholder="小明朗读课文时发音标准，被老师评为优秀。"),
     ],
     examples=[
-        ["夸大", "我们使用生成式人工智能已经很长时间了，所以最近的媒体报道可能被夸大了。"],
+        ["聪明", "小明很聪明，每年考班上第一名。"],
+        ["尴尬", "小明去朋友的生日庆祝会，忘了带礼物，感到很尴尬。"],
+        ["标准", "小明朗读课文时发音标准，被老师评为优秀。"],
     ],
-    outputs="number",
+    outputs=["plot"],
     title="Chinese Sentence Grading",
-).launch(share=True)
-
-# Plot bar chart of probs x target_words to find optimal cutoff
-
-# pipeline = pipeline(task="image-classification", model="julien-c/hotdog-not-hotdog")
-
-# def predict(image):
-#     predictions = pipeline(image)
-#     return {p["label"]: p["score"] for p in predictions}
-
-# gr.Interface(
-#     predict,
-#     inputs=gr.inputs.Image(label="Upload hot dog candidate", type="filepath"),
-#     outputs=gr.outputs.Label(num_top_classes=2),
-#     title="Hot Dog? Or Not?",
-# ).launch()
+).launch()

requirements.txt

@@ -1,4 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/cu113
 torch
 transformers
-pandas
+numpy 
+pandas
+plotly.express