Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TextClassificationPipeline
|
3 |
+
import re
|
4 |
+
|
5 |
+
tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
|
6 |
+
model = AutoModelForSequenceClassification.from_pretrained("Qilex/colorpAI-monocolor")
|
7 |
+
|
8 |
+
def round_to_2(num):
|
9 |
+
return round(num, 2)
|
10 |
+
|
11 |
+
def format_output(out_list):
|
12 |
+
white = 0
|
13 |
+
for dictionary in out_list:
|
14 |
+
if dictionary["label"] =='W':
|
15 |
+
white = round_to_2(dictionary["score"])
|
16 |
+
for dictionary in out_list:
|
17 |
+
if dictionary["label"] =='U':
|
18 |
+
blue = round_to_2(dictionary["score"])
|
19 |
+
for dictionary in out_list:
|
20 |
+
if dictionary["label"] =='B':
|
21 |
+
black = round_to_2(dictionary["score"])
|
22 |
+
for dictionary in out_list:
|
23 |
+
if dictionary["label"] =='R':
|
24 |
+
red = round_to_2(dictionary["score"])
|
25 |
+
for dictionary in out_list:
|
26 |
+
if dictionary["label"] =='G':
|
27 |
+
green = round_to_2(dictionary["score"])
|
28 |
+
for dictionary in out_list:
|
29 |
+
if dictionary["label"] =='C':
|
30 |
+
colorless = round_to_2(dictionary["score"])
|
31 |
+
out= {}
|
32 |
+
out['White'] = white
|
33 |
+
out['Blue'] = blue
|
34 |
+
out['Black'] = black
|
35 |
+
out['Red'] = red
|
36 |
+
out['Green'] = green
|
37 |
+
out['Colorless'] = colorless
|
38 |
+
return out
|
39 |
+
|
40 |
+
def predict(card):
|
41 |
+
return predictor_lg(card)
|
42 |
+
|
43 |
+
def remove_colored_pips(text):
|
44 |
+
pattern = r'\{[W,U,B,R,G,C]+/*[W,U,B,R,G,C]*\}'
|
45 |
+
return(re.sub(pattern, '{?}', text))
|
46 |
+
|
47 |
+
def preprocess_text(text):
|
48 |
+
return remove_colored_pips(text)
|
49 |
+
|
50 |
+
def categorize(card):
|
51 |
+
text = preprocess_text(card)
|
52 |
+
prediction = predict(text)
|
53 |
+
|
54 |
+
return format_output(prediction)
|
55 |
+
|
56 |
+
title = "Color pAI Version 1.0"
|
57 |
+
description = """
|
58 |
+
Color pAI is trained on around 18,000 Magic: the Gathering cards made available under Wizards of the Coast's
|
59 |
+
<a href="https://company.wizards.com/en/legal/fancontentpolicy" target = 'blank'>fan content policy</a>.
|
60 |
+
<br>
|
61 |
+
Input a card text using Scryfall syntax, and the model will tell evaluate which color it is most likely to be.
|
62 |
+
<br>Replace any card names with the word CARDNAME.
|
63 |
+
<br>
|
64 |
+
<br>This only works on monocolored cards. Version 2 will also handle multicolored cards.
|
65 |
+
<br>
|
66 |
+
"""
|
67 |
+
article = '''
|
68 |
+
<br>
|
69 |
+
Magic: the Gathering is property of Wizards of the Coast.
|
70 |
+
'''
|
71 |
+
predictor_lg = TextClassificationPipeline(model=model, tokenizer=tokenizer, function_to_apply = 'softmax', top_k = 6)
|
72 |
+
|
73 |
+
gr.Interface(
|
74 |
+
fn=categorize,
|
75 |
+
inputs=gr.Textbox(lines=1, placeholder="Type card text here."),
|
76 |
+
outputs=gr.Label(num_top_classes=6),
|
77 |
+
title=title,
|
78 |
+
description=description,
|
79 |
+
article = article,
|
80 |
+
).launch()
|