Empezar a poner en carpeta aprolos8000

.gitignore

@@ -1,3 +1,4 @@
 __pycache__
 .venv
-.vscode
+.vscode
+A-POR-LOS-8000

app.py

@@ -24,7 +24,7 @@ def call(audiopath, model, dataset_path, filter_white_noise):
 
 def predict(audio_path_pred):
     with torch.no_grad():
-        logits = call(audio_path_pred, model=model_class, dataset_path="data/mixed_data", filter_white_noise=True)
+        logits = call(audio_path_pred, model=model_class, dataset_path="A-POR-LOS-8000/data/mixed_data", filter_white_noise=True)
         predicted_class_ids_class = torch.argmax(logits, dim=-1).item()
         label_class = id2label_class[predicted_class_ids_class]
         label_mapping = {0: 'Hambre', 1: 'Problemas para respirar', 2: 'Dolor', 3: 'Cansancio/Incomodidad'}
@@ -33,7 +33,7 @@ def predict(audio_path_pred):
 
 def predict_stream(audio_path_stream):
     with torch.no_grad():
-        logits = call(audio_path_stream, model=model_mon, dataset_path="data/baby_cry_detection", filter_white_noise=False)
+        logits = call(audio_path_stream, model=model_mon, dataset_path="A-POR-LOS-8000/data/baby_cry_detection", filter_white_noise=False)
         probabilities = torch.nn.functional.softmax(logits, dim=-1)
         crying_probabilities = probabilities[:, 1]
         avg_crying_probability = crying_probabilities.mean()*100
@@ -45,7 +45,7 @@ def predict_stream(audio_path_stream):
 
 def decibelios(audio_path_stream):
     with torch.no_grad():
-        logits = call(audio_path_stream, model=model_mon, dataset_path="data/baby_cry_detection", filter_white_noise=False)
+        logits = call(audio_path_stream, model=model_mon, dataset_path="A-POR-LOS-8000/data/baby_cry_detection", filter_white_noise=False)
         rms = torch.sqrt(torch.mean(torch.square(logits)))
         db_level = 20 * torch.log10(rms + 1e-6).item()
         return db_level

model.py

@@ -1,6 +1,7 @@
 import os
 import json
 import random
+import argparse
 import torch
 import torchaudio
 from torch.utils.data import Dataset, DataLoader
@@ -187,11 +188,18 @@ def load_config(model_name):
     return model_config
 
 if __name__ == "__main__":
-    # config = load_config(clasificador) # PARA CAMBIAR MODELOS
-    # filter_white_noise = True
-    config = load_config(monitor) # PARA CAMBIAR MODELOS
-    filter_white_noise = False
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--n", choices=["mon", "class"], 
+        required=True, help="Elegir qué modelo entrenar"
+        )
+    args = parser.parse_args()
+    config = load_config(args.n)
     training_args = config["training_args"]
     output_dir = config["output_dir"]
     dataset_path = config["dataset_path"]
+    if args.n == "mon":
+        filter_white_noise = False
+    elif args.n == "class":
+        filter_white_noise = True
     main(training_args, output_dir, dataset_path, filter_white_noise)

models_config.json

@@ -1,12 +1,12 @@
 {
     "mon": {
-        "dataset_path": "data/baby_cry_detection",
-        "output_dir": "distilhubert-finetuned-cry-detector",
+        "dataset_path": "A-POR-LOS-8000/data/baby_cry_detection",
+        "output_dir": "A-POR-LOS-8000/distilhubert-finetuned-cry-detector",
         "training_args": {
-            "num_train_epochs": 6, 
-            "learning_rate": 0.0001,
+            "num_train_epochs": 10, 
+            "learning_rate": 0.00003,
             "warmup_ratio": 0.001,
-            "output_dir": "distilhubert-finetuned-cry-detector",
+            "output_dir": "A-POR-LOS-8000/distilhubert-finetuned-cry-detector",
             "eval_strategy": "epoch",
             "save_strategy": "epoch",
             "lr_scheduler_type": "cosine",
@@ -27,13 +27,13 @@
         }
     },
     "class": {
-        "dataset_path": "data/mixed_data",
-        "output_dir": "distilhubert-finetuned-mixed-data",
+        "dataset_path": "A-POR-LOS-8000/data/mixed_data",
+        "output_dir": "A-POR-LOS-8000/distilhubert-finetuned-mixed-data",
         "training_args": {
-            "num_train_epochs": 10,
-            "learning_rate": 0.0001,
-            "warmup_ratio": 0.001,
-            "output_dir": "distilhubert-finetuned-mixed-data",
+            "num_train_epochs": 15,
+            "learning_rate": 0.0003,
+            "warmup_ratio": 0.4,
+            "output_dir": "A-POR-LOS-8000/distilhubert-finetuned-mixed-data",
             "eval_strategy": "epoch",
             "save_strategy": "epoch",
             "lr_scheduler_type": "cosine",
@@ -44,7 +44,6 @@
             "gradient_checkpointing": true,
             "load_best_model_at_end": true,
             "greater_is_better": true,
-            "metric_for_best_model": "accuracy",
             "optim": "adamw_torch",
             "hub_strategy": "checkpoint",
             "report_to": "tensorboard",