YasinProDebian/mental-health-disorder

Model Card for Model ID

Model Details

Model Description

• Developed by: [More Information Needed]
• Funded by [optional]: [More Information Needed]
• Shared by [optional]: [More Information Needed]
• Model type: [More Information Needed]
• Language(s) (NLP): [More Information Needed]
• License: [More Information Needed]
• Finetuned from model [optional]: [More Information Needed]

Model Sources [optional]

• Repository: [More Information Needed]
• Paper [optional]: [More Information Needed]
• Demo [optional]: [More Information Needed]

Uses

Direct Use

[More Information Needed]

Downstream Use [optional]

[More Information Needed]

Out-of-Scope Use

[More Information Needed]

Bias, Risks, and Limitations

[More Information Needed]

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Use the code below to get started with the model.

import torch
from transformers import AutoTokenizer, AutoModelForMaskedLM
from sentence_transformers import SentenceTransformer
from huggingface_hub import hf_hub_download
from datasets import load_dataset
import pandas as pd
from peft import PeftModel

# تنظیمات اولیه
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
category_repo_id = "YasinProDebian/mental-health-category"  # مخزن دسته‌بندی
disorder_repo_id = "YasinProDebian/mental-health-disorder"  # مخزن اختلال
base_model_id = "xlm-roberta-large"  # مدل پایه برای اختلال

# اطلاعات GPU
if torch.cuda.is_available():
    print(f"GPU: {torch.cuda.get_device_name(0)}, حافظه: {torch.cuda.get_device_properties(0).total_memory / 1e9:.2f} GB")
else:
    print("GPU در دسترس نیست، از CPU استفاده می‌شود.")

# بارگذاری دیتاست و نگاشت دسته‌بندی‌ها و اختلال‌ها
dataset = load_dataset("YasinProDebian/mental-health")
df = pd.DataFrame(dataset['train'])

# استخراج دسته‌بندی‌ها و اختلال‌ها
category_labels = df["دسته بندی"].unique().tolist()
disorder_labels = df["اختلال"].unique().tolist()

# فیلتر کردن اختلال‌های کم‌تعداد (مانند کد آموزشی)
min_samples = 5
class_counts = df["اختلال"].value_counts().reindex(disorder_labels, fill_value=0)
valid_disorders = class_counts[class_counts >= min_samples].index.tolist()
disorder_labels = valid_disorders

# ایجاد نگاشت بین دسته‌بندی‌ها و اختلال‌ها
category_to_disorders = {category: df[df["دسته بندی"] == category]["اختلال"].unique().tolist() for category in category_labels}
print("نگاشت دسته‌بندی‌ها و اختلال‌ها:")
for category, disorders in category_to_disorders.items():
    print(f"{category}: {disorders}")

# بارگذاری توکنایزرها و مدل‌ها
category_tokenizer = AutoTokenizer.from_pretrained(category_repo_id)
disorder_tokenizer = AutoTokenizer.from_pretrained(disorder_repo_id)

# بارگذاری Sentence-Transformer
sentence_model = SentenceTransformer('paraphrase-multilingual-mpnet-base-v2')

# بارگذاری مدل دسته‌بندی
category_model = AutoModelForMaskedLM.from_pretrained(category_repo_id).to(device)
category_projection_path = hf_hub_download(repo_id=category_repo_id, filename="projection_layer.pth")
category_classifier_path = hf_hub_download(repo_id=category_repo_id, filename="classifier.pth")

combined_dim = category_model.config.hidden_size + sentence_model.get_sentence_embedding_dimension()
hidden_size = category_model.config.hidden_size

category_projection_layer = torch.nn.Linear(combined_dim, hidden_size)
category_projection_layer.load_state_dict(torch.load(category_projection_path, map_location=device, weights_only=True))
category_projection_layer.to(device)

category_classifier = torch.nn.Linear(hidden_size, len(category_labels))
category_classifier.load_state_dict(torch.load(category_classifier_path, map_location=device, weights_only=True))
category_classifier.to(device)

# بارگذاری مدل اختلال (با LoRA)
disorder_base_model = AutoModelForMaskedLM.from_pretrained(base_model_id)
disorder_model = PeftModel.from_pretrained(disorder_base_model, disorder_repo_id).to(device)
disorder_projection_path = hf_hub_download(repo_id=disorder_repo_id, filename="projection_layer.pth")
disorder_classifier_path = hf_hub_download(repo_id=disorder_repo_id, filename="classifier.pth")

disorder_projection_layer = torch.nn.Linear(combined_dim, hidden_size)
disorder_projection_layer.load_state_dict(torch.load(disorder_projection_path, map_location=device, weights_only=True))
disorder_projection_layer.to(device)

disorder_classifier = torch.nn.Linear(hidden_size, len(disorder_labels))
disorder_classifier.load_state_dict(torch.load(disorder_classifier_path, map_location=device, weights_only=True))
disorder_classifier.to(device)

# تابع پیش‌بینی دسته‌بندی با آستانه 0.005
def predict_category(input_sentence):
    category_model.eval()
    category_projection_layer.eval()
    category_classifier.eval()

    encoding = category_tokenizer(input_sentence, padding="max_length", truncation=True, max_length=128, return_tensors="pt")
    input_ids = encoding["input_ids"].to(device)
    attention_mask = encoding["attention_mask"].to(device)

    decoded_text = category_tokenizer.decode(input_ids[0], skip_special_tokens=True)
    sentence_embedding = sentence_model.encode(decoded_text, convert_to_tensor=True).to(device)

    with torch.no_grad():
        outputs = category_model.roberta(input_ids=input_ids, attention_mask=attention_mask, output_hidden_states=True)
        hidden_states = outputs.last_hidden_state

        sentence_embedding_expanded = sentence_embedding.unsqueeze(0).expand(hidden_states.size(1), -1).unsqueeze(0)
        combined_input = torch.cat((hidden_states, sentence_embedding_expanded), dim=-1)

        projected_output = category_projection_layer(combined_input)
        pooled_output = projected_output.mean(dim=1)

        logits = category_classifier(pooled_output)
        probabilities = torch.softmax(logits, dim=1).squeeze().cpu().tolist()

    # فیلتر کردن دسته‌بندی‌ها با احتمال بالای 0.005 و انتخاب حداکثر 5 مورد
    category_prob_pairs = [(cat, prob) for cat, prob in zip(category_labels, probabilities) if prob > 0.005]
    category_prob_pairs = sorted(category_prob_pairs, key=lambda x: x[1], reverse=True)[:5]

    return category_prob_pairs

# تابع پیش‌بینی اختلال
def predict_disorder(input_sentence):
    disorder_model.eval()
    disorder_projection_layer.eval()
    disorder_classifier.eval()

    encoding = disorder_tokenizer(input_sentence, padding="max_length", truncation=True, max_length=128, return_tensors="pt")
    input_ids = encoding["input_ids"].to(device)
    attention_mask = encoding["attention_mask"].to(device)

    decoded_text = disorder_tokenizer.decode(input_ids[0], skip_special_tokens=True)
    sentence_embedding = sentence_model.encode(decoded_text, convert_to_tensor=True).to(device)

    with torch.no_grad():
       

        outputs = disorder_model(input_ids=input_ids, attention_mask=attention_mask, output_hidden_states=True)
        hidden_states = outputs.hidden_states[-1]

        sentence_embedding_expanded = sentence_embedding.unsqueeze(0).unsqueeze(0).expand(1, hidden_states.size(1), -1)
        combined_input = torch.cat((hidden_states, sentence_embedding_expanded), dim=-1)

        projected_output = disorder_projection_layer(combined_input)
        pooled_output = projected_output.mean(dim=1)

        logits = disorder_classifier(pooled_output)
        probabilities = torch.softmax(logits, dim=1).squeeze().cpu().tolist()

    disorder_prob_pairs = list(zip(disorder_labels, probabilities))
    disorder_prob_pairs = sorted(disorder_prob_pairs, key=lambda x: x[1], reverse=True)

    return disorder_prob_pairs

# تابع ترکیبی برای پیش‌بینی دسته‌بندی و اختلال با آستانه‌ها
def predict_category_and_disorders(input_sentence):
    print(f"جمله ورودی: {input_sentence}")
    
    # پیش‌بینی دسته‌بندی‌ها با آستانه 0.005
    top_categories = predict_category(input_sentence)
    if not top_categories:
        print("هیچ دسته‌بندی با احتمال بالای 0.005 یافت نشد.")
        return
    
    print("دسته‌بندی‌های برتر (بالای 0.005، حداکثر 5 مورد):")
    for category, prob in top_categories:
        print(f"{category}: {prob:.4f}")
    
    # پیش‌بینی اختلال‌ها
    all_disorder_preds = predict_disorder(input_sentence)
    
    # فیلتر کردن اختلال‌ها برای هر دسته‌بندی با آستانه 0.0075
    print("\nاختلال‌های پراحتمال در هر دسته‌بندی (بالای 0.0075، حداکثر 4 مورد):")
    for category, _ in top_categories:
        relevant_disorders = category_to_disorders[category]
        category_disorder_preds = [pred for pred in all_disorder_preds if pred[0] in relevant_disorders and pred[1] > 0.0075]
        category_disorder_preds = sorted(category_disorder_preds, key=lambda x: x[1], reverse=True)[:4]
        
        print(f"\nدسته‌بندی: {category}")
        if category_disorder_preds:
            for disorder, prob in category_disorder_preds:
                print(f"  {disorder}: {prob:.4f}")
        else:
            print("  هیچ اختلالی با احتمال بالای 0.0075 یافت نشد.")

# تست با یک جمله
test_sentence = "نمی توانم ارضا شوم و رابطه جنسی به من لذتی نمی دهد"
predict_category_and_disorders(test_sentence)

Training Details

Training Data

[More Information Needed]

Training Procedure

Preprocessing [optional]

[More Information Needed]

Training Hyperparameters

• Training regime: [More Information Needed]

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Testing Data, Factors & Metrics

Testing Data

[More Information Needed]

Factors

[More Information Needed]

Metrics

[More Information Needed]

Results

[More Information Needed]

Summary

Model Examination [optional]

[More Information Needed]

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: [More Information Needed]
• Hours used: [More Information Needed]
• Cloud Provider: [More Information Needed]
• Compute Region: [More Information Needed]
• Carbon Emitted: [More Information Needed]

Technical Specifications [optional]

Model Architecture and Objective

[More Information Needed]

Compute Infrastructure

[More Information Needed]

Hardware

[More Information Needed]

Software

[More Information Needed]

Citation [optional]

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

Glossary [optional]

[More Information Needed]

More Information [optional]

[More Information Needed]

Model Card Authors [optional]

[More Information Needed]

Model Card Contact

[More Information Needed]

Framework versions

• PEFT 0.15.0