Python utility for gradient pushing

Uses only numpy.

Written from scratch using only numpy (no Keras, TF, PyT, …)

Autograd based differentiator

numpy override by autograd

Not SGD or Batch, only GD

Configurable knobs

NN_EPOCHS = 200000 NN_LEARNING_RATE = 0.205 NN_PARAMS_NPY_FILE= None NN_NORMALIZE = True NN_LOG_TARGET = False NN_MULTI_ENCODE_TEXT_VARS = False NN_APPLY_DATA_SCIENCE = True NN_SHUFFLE_ROWS_EPOCH = True

NN_BATCH_ROWS_EPOCH = False NN_INPUT_TO_HIDDEN_MULTIPLIER = 3 NN_ZERO_MEAN_NORMALIZE = False # True will make zero mean set(with +,- values) so will not work rmsle NN_RUN_MODE = "kaggle_home" NN_SHAPE = "wide" # long, wide NN_DROPOUT = False

## Debug configs NN_DEBUG_PRINT_EPOCH_COUNT = 100 NN_DEBUG_EXIT_EPOCH_ONE = False NN_DEBUG_SHAPES = False

Remove outliers, Sorted (Manual in csv)