find_lr script finished advances #26

scripts/find_learning_rates.sh

@@ -1,23 +1,28 @@
 #!/bin/bash
 
-#SBATCH --job-name LR_EfficientNetB0    # Nombre del proceso
-
-#SBATCH --partition dios                # Cola para ejecutar
-
-#SBATCH --gres=gpu:1                    # Numero de gpus a usar
-
-
+#SBATCH --job-name=train_EfficientNetB0    # Process name
+#SBATCH --partition=dios                  # Queue for execution
+#SBATCH --gres=gpu:1                      # Number of GPUs to use
+#SBATCH --mail-type=END,FAIL              # Notifications for job done & fail
+#SBATCH [email protected]  # Where to send notification
 
+# Load necessary paths
 export PATH="/opt/anaconda/anaconda3/bin:$PATH"
-
 export PATH="/opt/anaconda/bin:$PATH"
 
+# Setup Conda environment
 eval "$(conda shell.bash hook)"
+conda activate /mnt/homeGPU/polivares/tda-nn/tda-nn-separability
+export TFHUB_CACHE_DIR=.
 
-conda activate /mnt/homeGPU/polivares/tda-nn-separability
+# Check if correct number of arguments is passed
+if [ "$#" -ne 1 ]; then
+    echo "Usage: $0 <config_file>"
+    exit 1
+fi
 
-export TFHUB_CACHE_DIR=.
+config_file=$1
 
-python find_learning_rates.py          
+python find_learning_rates.py $config_file
 
-mail -s "Proceso finalizado" [email protected] <<< "El proceso ha finalizado"
+# mail -s "Proceso finalizado" [email protected] <<<"El proceso ha finalizado"

utils/plotting.py

@@ -1,5 +1,52 @@
 import matplotlib.pyplot as plt
 import numpy as np
+import pandas as pd
+
+def plot_loss_statistics_with_seaborn(loss_lists):
+    """
+    Plots the mean loss per epoch and the standard deviation as a shaded area for multiple experiments with variable epoch lengths using seaborn for improved aesthetics.
+    
+    Args:
+        loss_lists (list of list of lists): Each element is a list of lists where each sublist represents 
+                                            the loss per epoch for a single experiment, not necessarily all the same length.
+    """
+    # Setting up seaborn for better aesthetics
+    sns.set(style="whitegrid")
+
+    # Determine the maximum number of epochs across all experiments
+    max_epochs = max(max(len(losses) for losses in experiment) for experiment in loss_lists)
+
+    num_experiments = len(loss_lists)
+    mean_losses = []
+    std_losses = []
+
+    for experiment_losses in loss_lists:
+        # Create an array with shape (num_of_experiments, max_epochs), initialized with NaN
+        losses_array = np.full((len(experiment_losses), max_epochs), np.nan)
+
+        # Fill the array with loss values
+        for i, losses in enumerate(experiment_losses):
+            losses_array[i, :len(losses)] = losses
+
+        # Compute mean and std deviation along the experiment axis, ignoring NaNs
+        mean_losses.append(np.nanmean(losses_array, axis=0))
+        std_losses.append(np.nanstd(losses_array, axis=0))
+
+    # Plotting with seaborn
+    plt.figure(figsize=(10, 6))
+    epochs_x = np.arange(1, max_epochs + 1)
+    colors = sns.color_palette("hsv", num_experiments)  # Using seaborn color palette
+
+    for i, (mean_loss, std_loss) in enumerate(zip(mean_losses, std_losses)):
+        plt.plot(epochs_x, mean_loss, label=f'Experiment Group {i+1}', color=colors[i])
+        plt.fill_between(epochs_x, mean_loss - std_loss, mean_loss + std_loss, color=colors[i], alpha=0.3)
+
+    plt.title('Mean Loss Per Epoch With Standard Deviation')
+    plt.xlabel('Epoch')
+    plt.ylabel('Mean Loss')
+    plt.xticks(epochs_x)  # Set x-ticks to show integer values for epochs
+    plt.legend()
+    plt.show()
 
 def plot_loss(training_epoch_losses, validation_epoch_losses, plot_path):
     """
@@ -45,7 +92,7 @@ def plot_lr_vs_loss(log_lrs, losses, plot_path):
     """
     plt.figure(figsize=(10, 5))
 
-    plt.plot(log_lrs, losses, label="Learning Rate vs Loss", marker='o')
+    plt.plot(log_lrs, losses, label="Learning Rate vs Loss")
 
     plt.title("Learning Rate vs Loss")
     plt.xlabel("Log Learning Rate")
@@ -54,3 +101,11 @@ def plot_lr_vs_loss(log_lrs, losses, plot_path):
     plt.grid(True)
     plt.savefig(plot_path)
     plt.close()
+
+    # Deriving CSV path from the plot path
+    csv_path = plot_path.replace(".png", ".csv")
+
+    # Saving data to a CSV file
+    data = {'Log Learning Rate': log_lrs, 'Loss': losses}
+    df = pd.DataFrame(data)
+    df.to_csv(csv_path, index=False)