gizemcaylak/deeplearning

OCaml implementation of fully connected deep neural network

DeepLearning

OCaml implementation of fully connected stochastic deep neural network

How to compile the code?

Just type make on the terminal!

$ make

How to run the code?

Parameters

@data_filepath			: file path to train file
@delimeter 			: a char that is field delimeter in the sequence of floating numbers for each 
								string in data e.g. ','
@test_ratio			: represent the proportion of the dataset to include in the test split, should be 								between 0.0 and 1.0  
@Learning_rate 			: initial learning rate, a floating point number generally small
@learning_rate_method		:0 = learning rate does not change througout program
				 1 = learning rate decreases after every sqrt(sample size) iteration
@alpha				: a floating point number between [0,1] to determine the effect of previous weight changes
@epochs				: the number of epochs	
@activation_func		: 0 = sigmoid function
				  1 = relu function					
@hidden_layer_no 		: the number of hidden layers in the network
@neuron_no_per_layer		: the number of neurons in each hidden layer

./ann data_filepath delimeter test_ratio learning_rate learning_rate_method alpha epochs activation_func hidden_layer_no neuron_number_1 neuron_number2 neuron_numbern  

Examples

*The results of the examples can be found in 'Results' subfolder

MNIST

(1)
The number of hidden layers 1.
	The number of neurons in hidden layers 
	0 : 8
(2)
The number of hidden layers 2.
	The number of neurons in hidden layers 
	0 : 10
	1 : 10 
The number of epochs	: 100
Delimeter 				: ',' -> comma
Test ratio 				: 0.2
Learning rate method 	: Learning rate decreases
Learning rate 			: starts with 0.5
Activation Function 	: sigmoid
Alpha 					: 0.01
Run on MNIST data

Terminal code

(1)

$ tar xf data.tar.xz data.txt
$ ./ann DATASET/MNIST/data.txt , 0.2 0.5 1 0.01 100 0 1 8 > Results/MNIST_1.txt	

(2)

$ ./ann DATASET/MNIST/data.txt , 0.2 0.5 1 0.01 100 0 2 10 10 > Results/MNIST_2.txt

Iris

The number of hidden layers 2.
	The number of neurons in hidden layers 
	0 : 8
	1 : 8
The number of epochs	: 100
Delimeter 				: ',' -> comma
Test ratio 				: 0.2
Learning rate method 	: Learning rate decreases
Learning rate 			: starts with 3 (1) and starts with 0.01 (2) 
Activation Function 	: sigmoid(1) and relu(2)
Alpha 					: 0.05
Run on Iris data

Terminal code

(1) with Sigmoid Function

$ ./ann DATASET/Iris/data.txt , 0.2 3 1 0.05 100 0 2 3 5  > Results/iris_sigmoid.txt	

(2) with Relu Function

$ ./ann DATASET/Iris/data.txt , 0.2 0.01 1 0.05 100 1 2 8 8 > Results/iris_relu.txt

Wine

The number of hidden layers 1.
	The number of neurons in hidden layers 
	0 : 6
The number of epochs	: 100
Delimeter 				: ',' -> comma
Test ratio 				: 0.2
Learning rate method 	: Learning rate decreases
Learning rate 			: starts with 3 
Activation Function 	: sigmoid
Alpha 					: 0.05
Run on Wine data

Terminal code

$ ./ann DATASET/Wine/data.txt , 0.2 3 0 0.05 100 0 1 6 > Results/wine.txt

Breast Cancer

The number of hidden layers 1.
	The number of neurons in hidden layers 
	0 : 10
The number of epochs	: 100
Delimeter 				: ',' -> comma
Test ratio 				: 0.2
Learning rate method 	: Learning rate decreases
Learning rate 			: starts with 0.5 
Activation Function 	: sigmoid
Alpha 					: 0.01
Run on Wine data

Terminal code

$ ./ann DATASET/Breast_cancer/data.txt , 0.2 0.5 1 0.01 100 0 1 10 > Results/breast_cancer.txt

Digits

The number of hidden layers 2.
	The number of neurons in hidden layers 
	0 : 10
	1 : 10
The number of epochs	: 100
Delimeter 				: ',' -> comma
Test ratio 				: 0.2
Learning rate method 	: Learning rate decreases
Learning rate 			: starts with 0.5 
Activation Function 	: sigmoid
Alpha 					: 0.05
Run on Wine data

Terminal code

$ ./ann DATASET/Digits/data.txt , 0.2 0.5 1 0.01 100 0 2 10 10 > Results/digits.txt