the big regression

api/app.py

@@ -0,0 +1,121 @@
+import gradio as gr
+import pandas as pd
+import numpy as np
+import torch
+import joblib
+import json
+import os
+from transformers import AutoTokenizer, AutoModel
+from torch import nn
+
+class RegressionTransformer(nn.Module):
+    def __init__(self, model_name='neuralmind/bert-base-portuguese-cased'):
+        super(RegressionTransformer, self).__init__()
+        self.bert = AutoModel.from_pretrained(model_name)
+        self.regressor = nn.Sequential(
+            nn.Dropout(p=0.2),
+            nn.Linear(self.bert.config.hidden_size, 128),
+            nn.ReLU(),
+            nn.Dropout(p=0.2),
+            nn.Linear(128, 1)
+        )
+    def forward(self, input_ids, attention_mask):
+        outputs = self.bert(input_ids=input_ids, attention_mask=attention_mask)
+        pooled_output = outputs.pooler_output
+        return self.regressor(pooled_output)
+
+print("Carregando artefatos do modelo...")
+device = torch.device("cpu")
+
+MODEL_PATH = './modelo/best_model_state.pth'
+model = RegressionTransformer()
+model.load_state_dict(torch.load(MODEL_PATH, map_location=device))
+model.to(device)
+model.eval()
+print("Modelo carregado com sucesso.")
+
+TOKENIZER_NAME = 'neuralmind/bert-base-portuguese-cased'
+tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_NAME)
+print("Tokenizador carregado com sucesso.")
+
+FEATURES_PATH = './modelo/model_features.json'
+with open(FEATURES_PATH, 'r') as f:
+    model_features = json.load()['text_features']
+print("Features do modelo carregadas.")
+
+def predict_price(make, model_name, year, odometer, trim, body, transmission, color, interior):
+    print(f"Recebida nova predição: {make}, {model_name}, {year}, {odometer}km")
+    
+    input_data_dict = {
+        'make': make, 'model': model_name, 'year': year, 'odometer': odometer,
+        'trim': trim, 'body': body, 'transmission': transmission, 
+        'color': color, 'interior': interior
+    }
+    input_df = pd.DataFrame([input_data_dict])
+    
+    ano_referencia = 2024
+    input_df['age'] = ano_referencia - input_df['year']
+    input_df['sale_month'] = pd.to_datetime('now').month
+    input_df['sale_dayofweek'] = pd.to_datetime('now').dayofweek
+    
+    def criar_representacao_textual(row):
+        partes = [f"{coluna}[{str(row[coluna])}]" for coluna in model_features if col in row]
+        return " | ".join(partes)
+        
+    text_input = input_df.apply(criar_representacao_textual, axis=1).iloc[0]
+    
+    encoded_text = tokenizer.encode_plus(
+        text_input, max_length=128, add_special_tokens=True,
+        return_token_type_ids=False, padding='max_length',
+        truncation=True, return_attention_mask=True, return_tensors='pt',
+    )
+
+    input_ids = encoded_text['input_ids'].to(device)
+    attention_mask = encoded_text['attention_mask'].to(device)
+    
+    with torch.no_grad():
+        prediction_log = model(input_ids, attention_mask)
+
+    predicted_price = np.expm1(prediction_log.cpu().numpy()[0][0])
+
+    return f"R$ {predicted_price:,.2f}".replace(",", "X").replace(".", ",").replace("X", ".")
+
+with gr.Blocks(theme=gr.themes.Soft()) as iface:
+    gr.Markdown("# 🚗 FipeFinder AI: Previsão de Preços de Carros Usados")
+    gr.Markdown("Preencha as características do veículo para receber uma estimativa de preço de mercado baseada em nosso modelo de IA.")
+    
+    with gr.Row():
+        with gr.Column():
+            make_input = gr.Dropdown(label="Marca", choices=["Ford", "Chevrolet", "Honda", "Toyota", "Nissan", "Hyundai", "Kia"])
+            model_input = gr.Textbox(label="Modelo", placeholder="Ex: Ka, Onix, Civic...")
+            year_input = gr.Slider(label="Ano do Modelo", minimum=2000, maximum=2015, step=1, value=2012)
+            odo_input = gr.Number(label="Quilometragem (km)", value=80000)
+            trim_input = gr.Textbox(label="Versão", placeholder="Ex: SE 1.0, LTZ, EXL...")
+
+        with gr.Column():
+            body_input = gr.Dropdown(label="Carroceria", choices=["Sedan", "SUV", "Hatchback", "Pickup", "Minivan", "Coupe"])
+            trans_input = gr.Radio(label="Transmissão", choices=["automatic", "manual"], value="automatic")
+            color_input = gr.Textbox(label="Cor", placeholder="Ex: preto, branco, prata...")
+            interior_input = gr.Textbox(label="Cor Interior", placeholder="Ex: preto, cinza, bege...")
+            
+    predict_btn = gr.Button("Estimar Preço", variant="primary")
+    
+    with gr.Row():
+        output_price = gr.Label(label="Preço Estimado")
+
+    predict_btn.click(
+        fn=predict_price,
+        inputs=[make_input, model_input, year_input, odo_input, trim_input, body_input, trans_input, color_input, interior_input],
+        outputs=output_price
+    )
+    
+    gr.Examples(
+        examples=[
+            ["Ford", "Focus", 2013, 75000, "SE 2.0", "Hatchback", "automatic", "prata", "preto"],
+            ["Chevrolet", "Onix", 2014, 90000, "LT 1.0", "Hatchback", "manual", "branco", "cinza"],
+        ],
+        inputs=[make_input, model_input, year_input, odo_input, trim_input, body_input, trans_input, color_input, interior_input]
+    )
+
+if __name__ == "__main__":
+    iface.launch()

model/best_model_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48c0c48e29f53c023e16f98357f670bdb947cacf5195a70f3d64de6bf7bed6db
+size 436171458

model/model_features.json

@@ -0,0 +1 @@
+{"numeric": ["age", "odometer", "sale_month", "sale_dayofweek", "sale_dayofyear"], "categorical": ["make", "model", "trim", "body", "transmission", "color", "interior"]}

model/preprocessor.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36e18e8602fe9477c0fde1100d402befa84dd74508e61f0241684b25b1f4c5dd
+size 769

requirements.txt

@@ -0,0 +1,11 @@
+gradio
+torch
+torchvision
+torchaudio
+transformers
+sentence-transformers
+scikit-learn==1.3.2
+joblib
+pandas
+numpy
+python-dotenv