[ADD] seed_everything for reproducibility

evaluate/_utils_compute.py

@@ -1,6 +1,5 @@
 import numpy as np
 import pandas as pd
-import pytorch_lightning as pl
 import torch
 
 from tqdm import tqdm
@@ -9,6 +8,10 @@
 from prototypicalbeats.prototraining import ProtoBEATsModel
 from datamodules.TestDCASEDataModule import DCASEDataModule, AudioDatasetDCASE
 
+import pytorch_lightning as pl
+from pl.utilities.seed import seed_everything
+seed_everything(42, workers=True)
+
 def to_dataframe(features, labels):
     # Load the saved array and map the features and labels into a single dataframe
     input_features = np.load(features)
@@ -92,7 +95,7 @@ def reshape_support(support_samples, tensor_length=128, n_subsample=1):
 def train_model(
     model_type=None,
     datamodule_class=DCASEDataModule,
-    max_epochs=1,
+    max_epochs=5,
     enable_model_summary=False,
     num_sanity_val_steps=0,
     seed=42,
@@ -102,7 +105,7 @@ def train_model(
 ):
     # create the lightning trainer object
     trainer = pl.Trainer(
-        max_epochs=1,
+        max_epochs=max_epochs,
         enable_model_summary=enable_model_summary,
         num_sanity_val_steps=num_sanity_val_steps,
         deterministic=True,

evaluate/evaluateDCASE.py

@@ -21,6 +21,8 @@
 from datamodules.TestDCASEDataModule import DCASEDataModule, AudioDatasetDCASE
 
 import pytorch_lightning as pl
+from pl.utilities.seed import seed_everything
+seed_everything(42, workers=True)
 
 from callbacks.callbacks import MilestonesFinetuning
 
@@ -151,6 +153,12 @@ def compute(
 
         # Detect POS samples
         detected_pos_indices = np.where(p_values_pos == 1)[0]
+        print(f"[INFO] SELF DETECTED {detected_pos_indices} POS SAMPLES")
+
+        # BECAUSE CUDA ERROR WHEN RESAMPLING TOO MANY SAMPLES
+        if len(detected_pos_indices) > 40:
+            detected_pos_indices = np.random.choice(detected_pos_indices, size=40, replace=False)
+
         df_extension_pos = df_query.iloc[detected_pos_indices].copy()
         df_extension_pos["category"] = "POS"
 
@@ -159,6 +167,7 @@ def compute(
 
         # Randomly sample NEG samples to match the number of POS samples
         num_pos_samples = len(detected_pos_indices)
+
         if num_pos_samples > 0 and len(detected_neg_indices) > num_pos_samples:
             sampled_neg_indices = np.random.choice(detected_neg_indices, size=num_pos_samples, replace=False)
         else:
@@ -190,21 +199,19 @@ def compute(
         # CHANGE THE NUMBER OF SUPPORTS AND QUERY TO A CERTAIN RATIO #
         ##############################################################
 
-        support_to_query_ratio = 2
+        support_to_query_ratio = 1.5
 
         # Calculate the total number of samples for each category (POS and NEG)
         num_pos_samples = len(df_support_extended[df_support_extended['category'] == 'POS'])
         num_neg_samples = len(df_support_extended[df_support_extended['category'] == 'NEG'])
-        print(num_pos_samples)
-        print(num_neg_samples)
+
         # Assuming an equal distribution of samples between support and query for simplicity
         total_samples_per_class = min(num_pos_samples, num_neg_samples)
         print(f"TOTAL SLOT PER CLASS {total_samples_per_class}")
 
         # Calculate n_shot and n_query based on the desired ratio
         # The total should not exceed the total_samples_per_class for each category
         total_slots_per_class = total_samples_per_class / (1 + support_to_query_ratio)
-        print(f"TOTAL SLOT PER CLASS {total_samples_per_class}")
         n_query = int(total_slots_per_class)  # Round down to ensure we do not exceed the available samples
         n_shot = int(total_slots_per_class * support_to_query_ratio)
 
@@ -484,6 +491,7 @@ def main(cfg: DictConfig):
 
         results = results.append(result)
         results_raw = results_raw.append(result_raw)
+
         if cfg["predict"]["wav_save"]:
             write_wav(
                 files,