Yes they could have used MAPE or other relative loss but... Sometimes, you may wonder if the effort is worth it

Also calcium is only a secondary macronutrient, it is not anywhere remotely as important as nitrogen/phosphorus/potassium for plant health, yet is being weighted several orders of magnitude higher...

Maybe @MICADEE you're not calling the RMSE right. I had the same issue in the beginning. Instead of doing

mean_squared_error(y_true, y_pred)

Rather use:

mean_squared_error(y_true.reshape(-1), y_pred.reshape(-1)  ) 

I hope it will work. When I dit it I went from CV around 67x.xxx to CV aroud 125x.xxx

@marching_learning Very thoughtful.... May be ? Will definitely revisit, look at it again and get back to you on this. Thanks for the hint.👍

@marching_learning, this gives me the same result.

# Evaluate model

mae = mean_absolute_error(y_val, y_pred)

mse = mean_squared_error(y_val, y_pred)

rmse = np.sqrt(mse)

print(f'MAE: {mae:.4f}, RMSE: {rmse:.4f}')

MAE: 140.7893, RMSE: 391.2219

# Convert to NumPy arrays

y_val_array = np.array(y_val).reshape(-1)

y_pred_array = np.array(y_pred).reshape(-1)

# Evaluate model

mae = mean_absolute_error(y_val_array, y_pred_array)

mse = mean_squared_error(y_val_array, y_pred_array)

rmse = np.sqrt(mse)

print(f'MAE: {mae:.4f}, RMSE: {rmse:.4f}')

MAE: 140.7893, RMSE: 391.2219

Am I making a mistake?

Hello @CodeJoe I'm a bit suprised, bu try this function to settle it once for all:

import numpy as np

def RMSE(y_pred, y_true):

    return np.sqrt( np.mean((y_pred-y_true)**2) )

#

Just did. I just trained another. Same result:

def RMSE(y_pred, y_true):

return np.sqrt( np.mean((y_pred-y_true)**2) )

#

RMSE(y_pred, y_val)

481.73341583738716

Sorry @CodeJoe that I cannot help. I'm puzzled. This function really works for me

Oh no worries. It is all part of the experiments. Let me try k folds now.