eleanor91888388 / neural-networks Goto Github PK

********* Running N-Layer Neural Network ************

1. Open nn-start.py or nn-start.ipynb files
2. Modify model parameters:

• 'dataset': set path to dataset. For now, dataset have to be 1 file with 1 target column (but equal or more labels, or text or integers)

• 'target_column': name of target column in dataset

• 'lambda': set regularization lambda value to "penalize" overfitting.

• 'learning_rate': set the learning rate for gradient descent update steps.

• 'iterations': set the number of training iterations

• 'test_split': set percent of test split, which will be slice from loaded dataset (after shuffling).

• 'hidden_layers': set the structure for hidden layers of the model.

  Format: list of integers. Each number - the amount of neurons per layer. Several numbers - several hidden layers.

  For example, input '[25]' will create:

  • #1 Input Layer (number of neurons == number of feauters/columns (excluding target column)) ===>>>
  • #2 Hidden Layer (one hidden layer with 25 neurons). Fully connected to 1# Input layer
  • #3 Output Layer (number of neurons, equals the number of classes in target column). Fully connected to 2# Hidden layer

  For example, input '[25, 10]' will create:

  • #1 Input Layer (number of neurons == number of feauters/columns (excluding target column)) ===>>>
  • #2 Hidden Layer (first hidden layer with 25 neurons). Fully connected to 1# Input layer
  • #3 Hidden Layer (second hidden layer with 10 neurons). Fully connected to 2# Hidden layer
  • #4 Output Layer (number of neurons, equals the number of classes in target column). Fully connected to 3# Hidden layer

• 'print_split': while training proccess, it will print training results each "print_split" iteration.

• 'func_type': type of activation function to use. Option (for now): 'sigmoid', 'tanh'.

3. Run file.

********* Information ************

Implementation of algorythm completed based on numpy vectorization of forward and back propagations.