Linear Regression Mini-batch Gradient Descent in python: split training and validation data by batch size

Question

I was trying to split training and validation data by batch size, and finding the validation rmse. The size for training feature data is 11000. The size for validation feature data is 2750. However, when I use the zip function in for loop. It limits the for loop in the smaller range, which is 2750.

Anyone know how could I solve this problem? I have been stuck for two days.

  for epoch in range (0, maximum_iterations, 1):
      for i, j in zip(range(0, training_feature_data.shape[0], batch_size), range(0, validation_feature_data.shape[0], batch_size)):
         training_feature_data_i = training_feature_data[i:i+batch_size]
         training_target_data_i = training_target_data[i:i+batch_size]
         validation_feature_data_j = validation_feature_data[j:j+batch_size]
         validation_target_data_j = validation_target_data[j:j+batch_size]

You can split your second for loop into two loops, one for the training data and one for the validation data — Thomas Schillaci, Nov 15 '20 at 21:20

score 0 · Answer 1 · answered Nov 15 '20 at 21:28

Since the training and validation don't have any relation here (index i doesn't seem to affect the validation, and j doesn't affect training), it makes more sense to generate the mini-batches separately. This is what Thomas Schillaci is saying in his comment. Here's one suggestion on how to reformat your code.

for epoch in range (0, maximum_iterations, 1):
  for i in range(0, training_feature_data.shape[0], batch_size):
     training_feature_data_i = training_feature_data[i:i+batch_size]
     training_target_data_i = training_target_data[i:i+batch_size]
  for j in range(0, validation_feature_data.shape[0], batch_size):
     validation_feature_data_j = validation_feature_data[j:j+batch_size]
     validation_target_data_j = validation_target_data[j:j+batch_size]

Linear Regression Mini-batch Gradient Descent in python: split training and validation data by batch size

1 Answers1