Spaces:
Running
Running
kkruel8100
commited on
Commit
·
0c76a0d
1
Parent(s):
6ddce78
commit gradio app
Browse files- app.py +406 -0
- requirements.txt +9 -0
app.py
ADDED
@@ -0,0 +1,406 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python
|
2 |
+
# coding: utf-8
|
3 |
+
|
4 |
+
# In[17]:
|
5 |
+
|
6 |
+
|
7 |
+
import pickle
|
8 |
+
from PIL import Image
|
9 |
+
import numpy as np
|
10 |
+
import gradio as gr
|
11 |
+
from pathlib import Path
|
12 |
+
from transformers import pipeline
|
13 |
+
from tensorflow.keras.models import load_model
|
14 |
+
import tensorflow as tf
|
15 |
+
from transformers import MBartForConditionalGeneration, MBart50TokenizerFast
|
16 |
+
from dotenv import load_dotenv
|
17 |
+
import openai
|
18 |
+
import os
|
19 |
+
from langchain.schema import HumanMessage, SystemMessage
|
20 |
+
from langchain_openai import ChatOpenAI
|
21 |
+
|
22 |
+
|
23 |
+
# In[18]:
|
24 |
+
|
25 |
+
|
26 |
+
# Set the model's file path
|
27 |
+
file_path = Path("models/model_adam_scaled.h5")
|
28 |
+
|
29 |
+
# Load the model to a new object
|
30 |
+
adam_5 = tf.keras.models.load_model(file_path)
|
31 |
+
|
32 |
+
# Load env variables
|
33 |
+
load_dotenv()
|
34 |
+
|
35 |
+
# Add your OpenAI API key here
|
36 |
+
openai_api_key = os.getenv("OPENAI_API_KEY")
|
37 |
+
|
38 |
+
print(f"OpenAI API Key Loaded: {openai_api_key is not None}")
|
39 |
+
|
40 |
+
|
41 |
+
# Load the model and tokenizer for translation
|
42 |
+
model = MBartForConditionalGeneration.from_pretrained(
|
43 |
+
"facebook/mbart-large-50-many-to-many-mmt"
|
44 |
+
)
|
45 |
+
tokenizer = MBart50TokenizerFast.from_pretrained(
|
46 |
+
"facebook/mbart-large-50-many-to-many-mmt"
|
47 |
+
)
|
48 |
+
|
49 |
+
# Set source language
|
50 |
+
tokenizer.src_lang = "en_XX"
|
51 |
+
|
52 |
+
|
53 |
+
# In[22]:
|
54 |
+
|
55 |
+
|
56 |
+
# Constants
|
57 |
+
# Language information MBart
|
58 |
+
language_info = [
|
59 |
+
"English (en_XX)",
|
60 |
+
"Arabic (ar_AR)",
|
61 |
+
"Czech (cs_CZ)",
|
62 |
+
"German (de_DE)",
|
63 |
+
"Spanish (es_XX)",
|
64 |
+
"Estonian (et_EE)",
|
65 |
+
"Finnish (fi_FI)",
|
66 |
+
"French (fr_XX)",
|
67 |
+
"Gujarati (gu_IN)",
|
68 |
+
"Hindi (hi_IN)",
|
69 |
+
"Italian (it_IT)",
|
70 |
+
"Japanese (ja_XX)",
|
71 |
+
"Kazakh (kk_KZ)",
|
72 |
+
"Korean (ko_KR)",
|
73 |
+
"Lithuanian (lt_LT)",
|
74 |
+
"Latvian (lv_LV)",
|
75 |
+
"Burmese (my_MM)",
|
76 |
+
"Nepali (ne_NP)",
|
77 |
+
"Dutch (nl_XX)",
|
78 |
+
"Romanian (ro_RO)",
|
79 |
+
"Russian (ru_RU)",
|
80 |
+
"Sinhala (si_LK)",
|
81 |
+
"Turkish (tr_TR)",
|
82 |
+
"Vietnamese (vi_VN)",
|
83 |
+
"Chinese (zh_CN)",
|
84 |
+
"Afrikaans (af_ZA)",
|
85 |
+
"Azerbaijani (az_AZ)",
|
86 |
+
"Bengali (bn_IN)",
|
87 |
+
"Persian (fa_IR)",
|
88 |
+
"Hebrew (he_IL)",
|
89 |
+
"Croatian (hr_HR)",
|
90 |
+
"Indonesian (id_ID)",
|
91 |
+
"Georgian (ka_GE)",
|
92 |
+
"Khmer (km_KH)",
|
93 |
+
"Macedonian (mk_MK)",
|
94 |
+
"Malayalam (ml_IN)",
|
95 |
+
"Mongolian (mn_MN)",
|
96 |
+
"Marathi (mr_IN)",
|
97 |
+
"Polish (pl_PL)",
|
98 |
+
"Pashto (ps_AF)",
|
99 |
+
"Portuguese (pt_XX)",
|
100 |
+
"Swedish (sv_SE)",
|
101 |
+
"Swahili (sw_KE)",
|
102 |
+
"Tamil (ta_IN)",
|
103 |
+
"Telugu (te_IN)",
|
104 |
+
"Thai (th_TH)",
|
105 |
+
"Tagalog (tl_XX)",
|
106 |
+
"Ukrainian (uk_UA)",
|
107 |
+
"Urdu (ur_PK)",
|
108 |
+
"Xhosa (xh_ZA)",
|
109 |
+
"Galician (gl_ES)",
|
110 |
+
"Slovene (sl_SI)",
|
111 |
+
]
|
112 |
+
|
113 |
+
# Convert the information into a dictionary
|
114 |
+
language_dict = {}
|
115 |
+
for info in language_info:
|
116 |
+
name, code = info.split(" (")
|
117 |
+
code = code[:-1]
|
118 |
+
language_dict[name] = code
|
119 |
+
|
120 |
+
# Get the language names for choices in the dropdown
|
121 |
+
languages = list(language_dict.keys())
|
122 |
+
first_language = languages[0]
|
123 |
+
sorted_languages = sorted(languages[1:])
|
124 |
+
sorted_languages.insert(0, first_language)
|
125 |
+
|
126 |
+
default_language = "English"
|
127 |
+
|
128 |
+
# Prediction responses
|
129 |
+
malignant_text = "Malignant. Please consult a doctor for further evaluation."
|
130 |
+
benign_text = "Benign. Please consult a doctor for further evaluation."
|
131 |
+
|
132 |
+
|
133 |
+
# In[23]:
|
134 |
+
|
135 |
+
|
136 |
+
# Create instance
|
137 |
+
llm = ChatOpenAI(
|
138 |
+
openai_api_key=openai_api_key, model_name="gpt-3.5-turbo", temperature=0
|
139 |
+
)
|
140 |
+
|
141 |
+
|
142 |
+
# In[24]:
|
143 |
+
|
144 |
+
|
145 |
+
# Method to get system and human messages for ChatOpenAI - Predictions
|
146 |
+
def get_prediction_messages(prediction_text):
|
147 |
+
# Create a HumanMessage object
|
148 |
+
human_message = HumanMessage(content=f"skin lesion that appears {prediction_text}")
|
149 |
+
|
150 |
+
# Get the system message
|
151 |
+
system_message = SystemMessage(
|
152 |
+
content="You are a medical professional chatting with a patient. You want to provide helpful information and give a preliminary assessment."
|
153 |
+
)
|
154 |
+
|
155 |
+
# Return the system message
|
156 |
+
return [system_message, human_message]
|
157 |
+
|
158 |
+
|
159 |
+
# In[25]:
|
160 |
+
|
161 |
+
|
162 |
+
# Method to get system and human messages for ChatOpenAI - Help
|
163 |
+
def get_chat_messages(chat_prompt):
|
164 |
+
# Create a HumanMessage object
|
165 |
+
human_message = HumanMessage(content=chat_prompt)
|
166 |
+
|
167 |
+
# Get the system message
|
168 |
+
system_message = SystemMessage(
|
169 |
+
content="You are a medical professional chatting with a patient. You want to provide helpful information."
|
170 |
+
)
|
171 |
+
# Return the system message
|
172 |
+
return [system_message, human_message]
|
173 |
+
|
174 |
+
|
175 |
+
# In[26]:
|
176 |
+
|
177 |
+
|
178 |
+
# Method to predict the image
|
179 |
+
def predict_image(language, img):
|
180 |
+
try:
|
181 |
+
try:
|
182 |
+
# Process the image
|
183 |
+
img = img.resize((224, 224))
|
184 |
+
img_array = np.array(img) / 255.0
|
185 |
+
img_array = np.expand_dims(img_array, axis=0)
|
186 |
+
except Exception as e:
|
187 |
+
print(f"Error: {e}")
|
188 |
+
return "There was an error processing the image. Please try again."
|
189 |
+
|
190 |
+
# Get prediction from model
|
191 |
+
prediction = adam_5.predict(img_array)
|
192 |
+
text_prediction = "Malignant" if prediction[0][0] > 0.5 else "Benign"
|
193 |
+
|
194 |
+
try:
|
195 |
+
# Get the system and human messages
|
196 |
+
messages = get_prediction_messages(text_prediction)
|
197 |
+
|
198 |
+
# Get the response from ChatOpenAI
|
199 |
+
result = llm(messages)
|
200 |
+
|
201 |
+
# Get the text prediction
|
202 |
+
text_prediction = (
|
203 |
+
f"Prediction: {text_prediction} Explanation: {result.content}"
|
204 |
+
)
|
205 |
+
|
206 |
+
except Exception as e:
|
207 |
+
print(f"Error: {e}")
|
208 |
+
print(f"Prediction: {text_prediction}")
|
209 |
+
text_prediction = (
|
210 |
+
malignant_text if text_prediction == "Malignant" else benign_text
|
211 |
+
)
|
212 |
+
|
213 |
+
# Get selected language code
|
214 |
+
selected_code = language_dict[language]
|
215 |
+
|
216 |
+
# Check if the target and source languages are the same
|
217 |
+
if selected_code == "en_XX":
|
218 |
+
return (
|
219 |
+
text_prediction,
|
220 |
+
gr.update(visible=False),
|
221 |
+
gr.update(visible=True),
|
222 |
+
gr.update(visible=True),
|
223 |
+
gr.update(visible=True),
|
224 |
+
)
|
225 |
+
|
226 |
+
try:
|
227 |
+
# Encode, generate tokens, decode the prediction
|
228 |
+
encoded_text = tokenizer(text_prediction, return_tensors="pt")
|
229 |
+
generated_tokens = model.generate(
|
230 |
+
**encoded_text,
|
231 |
+
forced_bos_token_id=tokenizer.lang_code_to_id[selected_code],
|
232 |
+
)
|
233 |
+
result = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
|
234 |
+
|
235 |
+
# Return the result
|
236 |
+
return (
|
237 |
+
result[0],
|
238 |
+
gr.update(visible=False),
|
239 |
+
gr.update(visible=True),
|
240 |
+
gr.update(visible=True),
|
241 |
+
gr.update(visible=True),
|
242 |
+
)
|
243 |
+
except Exception as e:
|
244 |
+
print(f"Error: {e}")
|
245 |
+
return (
|
246 |
+
f"""There was an error processing the translation.
|
247 |
+
In English:
|
248 |
+
{text_prediction}
|
249 |
+
""",
|
250 |
+
gr.update(visible=False),
|
251 |
+
gr.update(visible=True),
|
252 |
+
gr.update(visible=True),
|
253 |
+
gr.update(visible=True),
|
254 |
+
)
|
255 |
+
|
256 |
+
except Exception as e:
|
257 |
+
print(f"Error: {e}")
|
258 |
+
return (
|
259 |
+
"There was an error processing the request. Please try again.",
|
260 |
+
gr.update(visible=True),
|
261 |
+
gr.update(visible=False),
|
262 |
+
gr.update(visible=False),
|
263 |
+
gr.update(visible=False),
|
264 |
+
)
|
265 |
+
|
266 |
+
|
267 |
+
# In[27]:
|
268 |
+
|
269 |
+
|
270 |
+
# Method for on submit
|
271 |
+
def on_submit(language, img):
|
272 |
+
print(f"Language: {language}")
|
273 |
+
if language is None or len(language) == 0:
|
274 |
+
language = default_language
|
275 |
+
if img is None:
|
276 |
+
return (
|
277 |
+
"No image uploaded. Please try again.",
|
278 |
+
gr.update(visible=True),
|
279 |
+
gr.update(visible=False),
|
280 |
+
gr.update(visible=False),
|
281 |
+
gr.update(visible=False),
|
282 |
+
)
|
283 |
+
return predict_image(language, img)
|
284 |
+
|
285 |
+
|
286 |
+
# In[28]:
|
287 |
+
|
288 |
+
|
289 |
+
# Method for on clear
|
290 |
+
def on_clear():
|
291 |
+
return (
|
292 |
+
gr.update(),
|
293 |
+
gr.update(),
|
294 |
+
gr.update(),
|
295 |
+
gr.update(visible=True),
|
296 |
+
gr.update(value=None, visible=False),
|
297 |
+
gr.update(value=None, visible=False),
|
298 |
+
gr.update(visible=False),
|
299 |
+
)
|
300 |
+
|
301 |
+
|
302 |
+
# In[29]:
|
303 |
+
|
304 |
+
|
305 |
+
# Method for on chat
|
306 |
+
def on_chat(language, chat_prompt):
|
307 |
+
try:
|
308 |
+
# Get the system and human messages
|
309 |
+
messages = get_chat_messages(chat_prompt)
|
310 |
+
# Get the response from ChatOpenAI
|
311 |
+
result = llm(messages)
|
312 |
+
# Get the text prediction
|
313 |
+
chat_response = result.content
|
314 |
+
|
315 |
+
except Exception as e:
|
316 |
+
print(f"Error: {e}")
|
317 |
+
return gr.update(
|
318 |
+
value="There was an error processing your question. Please try again.",
|
319 |
+
visible=True,
|
320 |
+
), gr.update(visible=False)
|
321 |
+
|
322 |
+
# Get selected language code
|
323 |
+
if language is None or len(language) == 0:
|
324 |
+
language = default_language
|
325 |
+
selected_code = language_dict[language]
|
326 |
+
# Check if the target and source languages are the same
|
327 |
+
if selected_code == "en_XX":
|
328 |
+
return gr.update(value=chat_response, visible=True), gr.update(visible=False)
|
329 |
+
|
330 |
+
try:
|
331 |
+
# Encode, generate tokens, decode the prediction
|
332 |
+
encoded_text = tokenizer(chat_response, return_tensors="pt")
|
333 |
+
generated_tokens = model.generate(
|
334 |
+
**encoded_text, forced_bos_token_id=tokenizer.lang_code_to_id[selected_code]
|
335 |
+
)
|
336 |
+
result = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
|
337 |
+
|
338 |
+
# Return the result
|
339 |
+
return gr.update(value=result[0], visible=True), gr.update(visible=False)
|
340 |
+
except Exception as e:
|
341 |
+
print(f"Error: {e}")
|
342 |
+
return (
|
343 |
+
gr.update(
|
344 |
+
value=f"""There was an error processing the translation.
|
345 |
+
In English:
|
346 |
+
{chat_response}
|
347 |
+
""",
|
348 |
+
visible=True,
|
349 |
+
),
|
350 |
+
gr.update(visible=False),
|
351 |
+
)
|
352 |
+
|
353 |
+
|
354 |
+
# In[30]:
|
355 |
+
|
356 |
+
|
357 |
+
# Gradio app
|
358 |
+
|
359 |
+
with gr.Blocks(theme=gr.themes.Default(primary_hue="green")) as demo:
|
360 |
+
intro = gr.Markdown(
|
361 |
+
"""
|
362 |
+
# Welcome to Skin Lesion Image Classifier!
|
363 |
+
Select prediction language and upload image to start.
|
364 |
+
"""
|
365 |
+
)
|
366 |
+
language = gr.Dropdown(
|
367 |
+
label="Response Language - Default English", choices=sorted_languages
|
368 |
+
)
|
369 |
+
img = gr.Image(image_mode="RGB", type="pil")
|
370 |
+
output = gr.Textbox(label="Results", show_copy_button=True)
|
371 |
+
chat_prompt = gr.Textbox(
|
372 |
+
label="Do you have a question about the results or skin cancer?",
|
373 |
+
placeholder="Enter your question here...",
|
374 |
+
visible=False,
|
375 |
+
)
|
376 |
+
chat_response = gr.Textbox(
|
377 |
+
label="Chat Response", visible=False, show_copy_button=True
|
378 |
+
)
|
379 |
+
submit_btn = gr.Button("Submit", variant="primary", visible=True)
|
380 |
+
chat_btn = gr.Button("Submit Question", variant="primary", visible=False)
|
381 |
+
submit_btn.click(
|
382 |
+
fn=on_submit,
|
383 |
+
inputs=[language, img],
|
384 |
+
outputs=[output, submit_btn, chat_prompt, chat_btn, chat_response],
|
385 |
+
)
|
386 |
+
chat_btn.click(
|
387 |
+
fn=on_chat, inputs=[language, chat_prompt], outputs=[chat_response, chat_btn]
|
388 |
+
)
|
389 |
+
clear_btn = gr.ClearButton(
|
390 |
+
components=[language, img, output, chat_response], variant="stop"
|
391 |
+
)
|
392 |
+
clear_btn.click(
|
393 |
+
fn=on_clear,
|
394 |
+
outputs=[
|
395 |
+
language,
|
396 |
+
img,
|
397 |
+
output,
|
398 |
+
submit_btn,
|
399 |
+
chat_prompt,
|
400 |
+
chat_response,
|
401 |
+
chat_btn,
|
402 |
+
],
|
403 |
+
)
|
404 |
+
|
405 |
+
if __name__ == "__main__":
|
406 |
+
demo.launch(share=True)
|
requirements.txt
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
gradio==4.4.1
|
2 |
+
langchain==0.1.17
|
3 |
+
langchain-community==0.0.36
|
4 |
+
langchain-core==0.1.50
|
5 |
+
langchain-openai==0.1.6
|
6 |
+
langchain-text-splitters==0.0.1
|
7 |
+
python-dotenv==1.0.0
|
8 |
+
tensorflow==2.12.0
|
9 |
+
transformers==4.40.1
|