matesoft's picture
Update app.py
6f36701 verified
raw
history blame
11.2 kB
# -*- coding: utf-8 -*-
"""2_preprocessing_test.ipynb
Automatically generated by Colab.
Original file is located at
https://colab.research.google.com/drive/10c3x9G9z70J73l0LJDA8_VDZphQmHEZB
"""
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.preprocessing import LabelEncoder
import os
from sklearn.model_selection import train_test_split
import pickle
import warnings
warnings.filterwarnings('ignore')
df1 = pd.read_csv("/content/drive/MyDrive/Google Colab/disease-symptom-prediction/data/dataset.csv")
print(df1.shape)
df1.head()
df1.sort_values(by='Disease', inplace=True)
df1.head()
df1.drop_duplicates(inplace=True)
df1.shape
df1['Disease'].value_counts()
df1[df1['Disease']=="Fungal infection"]
df1.fillna("none", inplace=True)
df1[df1['Disease']=="Fungal infection"]
df1.columns = df1.columns.str.strip().str.lower()
for col in df1.columns:
df1[col] = df1[col].astype(str).str.strip().str.lower()
symptom_cols = [col for col in df1.columns if col.startswith('symptom')]
print(symptom_cols)
all_symptoms = set()
for col in symptom_cols:
for val in df1[col].unique():
if val != 'none':
all_symptoms.add(val)
print(f"Unique symptoms: {len(all_symptoms)}")
print(all_symptoms)
df1.head()
df1_num = pd.DataFrame(df1['disease'])
for symptom in all_symptoms:
df1_num[symptom] = df1[symptom_cols].apply(lambda row: int(symptom in row.values), axis=1)
df1_num
X = df1_num.drop('disease', axis=1)
y = df1_num['disease']
X.shape, y.shape
X.sum(axis=1)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, stratify=y)
print(np.unique(y_train, return_counts=True))
print(np.unique(y_test, return_counts=True))
from sklearn.ensemble import RandomForestClassifier
model = RandomForestClassifier(n_estimators=100,random_state=42)
model.fit(X_train, y_train)
model.fit(X_train, y_train)
import pickle
# Save model
with open("disease_model.pkl", "wb") as f:
pickle.dump(model, f)
# Save symptom list (to use in the app later)
with open("symptoms.pkl", "wb") as f:
pickle.dump(list(all_symptoms), f)
# Original symptoms (keys)
all_symptoms = sorted(all_symptoms)
# Create display labels by replacing '_' with ' ' and capitalizing each word
display_symptoms = [symptom.replace('_', ' ').title() for symptom in all_symptoms]
# Create a mapping from display label back to original symptom key
label_to_symptom = dict(zip(display_symptoms, all_symptoms))
from sklearn.metrics import accuracy_score, f1_score
y_train_pred = model.predict(X_train)
train_accuracy = accuracy_score(y_train, y_train_pred)
train_f1_score = f1_score(y_train, y_train_pred,average="weighted")
print("Train Accuracy:", train_accuracy)
print("Train f1 score:", train_f1_score)
y_test_pred = model.predict(X_test)
test_accuracy = accuracy_score(y_test, y_test_pred)
test_f1_score = f1_score(y_test, y_test_pred, average="weighted")
print("Train Accuracy:", test_accuracy)
print("Train f1 score:", test_f1_score)
import numpy as np
# Example user symptoms
user_symptoms = ['nausea', 'vomiting', 'abdominal_pain', 'diarrhoea']
# Tip for the user
if len(user_symptoms) < 4:
print("Tip: The model performs better if you enter at least 4 symptoms.\n")
# Convert symptoms to input vector
input_vector = [1 if symptom in user_symptoms else 0 for symptom in all_symptoms]
input_vector = np.array([input_vector])
# Make prediction and get probabilities
probas = model.predict_proba(input_vector)[0]
max_proba = np.max(probas)
predicted = model.classes_[np.argmax(probas)]
# Confidence threshold
threshold = 0.5
# Print predicted disease and confidence
if max_proba < threshold:
print("Warning: The model is not confident about this prediction.")
print(f"Predicted disease: {predicted} (Confidence: {max_proba * 100:.1f}%)")
else:
print(f"Predicted disease: {predicted} (Confidence: {max_proba * 100:.1f}%)")
# Function to print top N diseases
def print_top_diseases(probas, model, top_n=5):
classes = model.classes_
sorted_indices = np.argsort(probas)[::-1]
print(f"\nTop {top_n} possible diseases:")
for i in range(min(top_n, len(classes))):
disease = classes[sorted_indices[i]]
probability = probas[sorted_indices[i]]
print(f"{i+1}. {disease}: {probability:.4f}")
# Show top 5 possible diseases
print_top_diseases(probas, model, top_n=5)
import gradio as gr
import pickle
import numpy as np
# --- 1. Load Disease Prediction Model ---
with open("disease_model.pkl", "rb") as f:
model = pickle.load(f)
with open("symptoms.pkl", "rb") as f:
all_symptoms = pickle.load(f)
# Preprocess symptoms
all_symptoms = sorted(all_symptoms)
display_symptoms = [s.replace('_', ' ').title() for s in all_symptoms]
label_to_symptom = dict(zip(display_symptoms, all_symptoms))
# --- 2. Medical Knowledge Base ---
MEDICAL_KNOWLEDGE = {
"migraine": [
"For migraines: (1) Rest in dark room (2) OTC pain relievers (ibuprofen/acetaminophen) (3) Apply cold compress (4) Consult neurologist if frequent",
"Migraine treatment options include triptans (prescription) and caffeine. Avoid triggers like bright lights or strong smells."
],
"allergy": [
"Allergy management: (1) Antihistamines (cetirizine/loratadine) (2) Nasal sprays (3) Allergy shots (immunotherapy) for severe cases",
"For food allergies: Strict avoidance, carry epinephrine auto-injector (EpiPen), read food labels carefully"
],
"cold": [
"Treat colds with rest, fluids, and OTC pain relievers. See doctor if fever lasts >3 days",
"Most colds resolve in 7-10 days. Use decongestants for nasal congestion"
],
"headache": [
"For headaches: Hydrate, rest, and use OTC pain relievers sparingly",
"Persistent headaches require medical evaluation - consult your doctor"
],
"fever": [
"For fever: Rest, fluids, and acetaminophen/ibuprofen. Seek help if >39Β°C or lasts >3 days",
"High fever warning: Seek emergency care if fever >40Β°C or with stiff neck"
]
}
SPECIAL_RESPONSES = {
"general approaches": "I can provide specific guidance for: allergies, migraines, colds, fever, back pain, rashes. What condition are you asking about?",
"consult a doctor": "For these symptoms, seek medical care: severe pain, difficulty breathing, sudden weakness, high fever (>103Β°F), or symptoms lasting >7 days"
}
def get_medical_response(user_query):
user_query = user_query.lower()
# First check for special cases
for phrase, response in SPECIAL_RESPONSES.items():
if phrase in user_query:
return response
# Then check medical conditions
for condition, responses in MEDICAL_KNOWLEDGE.items():
if condition in user_query:
return np.random.choice(responses)
# Final improvement - suggest related conditions
related = [cond for cond in MEDICAL_KNOWLEDGE.keys() if cond in user_query]
if related:
return f"Are you asking about {', '.join(related)}? {np.random.choice(MEDICAL_KNOWLEDGE[related[0]])}"
return "I can advise on: " + ", ".join(MEDICAL_KNOWLEDGE.keys()) + ". Please be more specific."
# --- 3. Disease Prediction Function ---
def predict_disease(selected_labels):
if not selected_labels or len(selected_labels) < 4:
return "⚠️ Please select at least 4 symptoms for accurate results."
user_symptoms = [label_to_symptom[label] for label in selected_labels]
input_vector = [1 if symptom in user_symptoms else 0 for symptom in all_symptoms]
input_vector = np.array([input_vector])
probas = model.predict_proba(input_vector)[0]
max_proba = np.max(probas)
predicted = model.classes_[np.argmax(probas)]
sorted_indices = np.argsort(probas)[::-1]
top_diseases = [
f"<b>{i+1}. {model.classes_[idx]}</b> β€” {probas[idx]*100:.1f}%"
for i, idx in enumerate(sorted_indices[:3])
]
prediction_result = (
f"<div style='background: #001a33; padding: 15px; border-radius: 8px; margin-bottom: 15px;'>"
f"<h3 style='color: #4fc3f7; margin-top: 0;'>🩺 Predicted Disease</h3>"
f"<p style='font-size: 18px; color: white;'>{predicted} <span style='color: #4fc3f7'>({max_proba*100:.1f}% confidence)</span></p>"
"</div>"
"<div style='background: #001a33; padding: 15px; border-radius: 8px;'>"
"<h3 style='color: #4fc3f7; margin-top: 0;'>πŸ” Top 3 Possible Diseases</h3>"
"<ul style='color: white; padding-left: 20px;'>" +
"".join([f"<li>{d}</li>" for d in top_diseases]) +
"</ul>"
"</div>"
)
return prediction_result
# --- 4. Chat Responder ---
def chatbot_respond(message, chat_history):
response = get_medical_response(message)
return chat_history + [(message, response)], ""
# --- 5. UI Setup ---
custom_css = """
:root {
--primary: #4fc3f7;
--secondary: #001a33;
--text: #ffffff;
--bg: #0a192f;
--card-bg: #0a2342;
--error: #ff6b6b;
}
body, .gradio-container {
background: var(--bg) !important;
color: var(--text) !important;
font-family: 'Segoe UI', Roboto, sans-serif;
}
/* [Keep all your existing CSS styles] */
"""
with gr.Blocks(css=custom_css) as demo:
gr.Markdown("""
<div style="text-align: center; margin-bottom: 20px;">
<h1 style="margin-bottom: 5px;">🧬 Medical Diagnosis Assistant</h1>
<p style="color: #4fc3f7; font-size: 16px;">Select symptoms for diagnosis and get medical advice</p>
</div>
""")
with gr.Row(equal_height=True):
with gr.Column(scale=1, min_width=300):
gr.Markdown("### πŸ” Symptom Checker")
symptoms_input = gr.CheckboxGroup(
choices=display_symptoms,
label="Select your symptoms:",
interactive=True
)
predict_btn = gr.Button("Analyze Symptoms", variant="primary")
prediction_output = gr.Markdown(
label="Diagnosis Results",
value="Your results will appear here..."
)
with gr.Column(scale=1, min_width=400):
gr.Markdown("### πŸ’¬ Medical Advisor")
chatbot = gr.Chatbot(
label="Chat with Medical Advisor",
show_label=False,
bubble_full_width=False
)
with gr.Row():
user_input = gr.Textbox(
placeholder="Ask about symptoms or treatments...",
label="",
show_label=False,
container=False,
scale=7
)
send_btn = gr.Button("Send", scale=1, min_width=80)
# Event handlers
predict_btn.click(
fn=predict_disease,
inputs=symptoms_input,
outputs=prediction_output
)
send_btn.click(
fn=chatbot_respond,
inputs=[user_input, chatbot],
outputs=[chatbot, user_input]
)
user_input.submit(
fn=chatbot_respond,
inputs=[user_input, chatbot],
outputs=[chatbot, user_input]
)
demo.launch()