| Type: | Package |
| Title: | Neural Network Numerai |
| Version: | 1.0.0 |
| Author: | Damian Siniakowicz |
| Maintainer: | Damian Siniakowicz <DamianSiniakowicz@gmail.com> |
| Date: | 2016-09-13 |
| Packaged: | 2016-09-13 |
| Description: | Interactively train neural networks on Numerai, https://numer.ai/, data. Generate tournament predictions and write them to a CSV. |
| Imports: | caret, methods, testthat |
| License: | GPL-3 |
| LazyData: | FALSE |
| RoxygenNote: | 5.0.1 |
| NeedsCompilation: | no |
| Repository: | CRAN |
| Date/Publication: | 2016-09-14 18:50:32 |
cost
Description
get the logarithmic loss for a set of predictions
Usage
## S4 method for signature 'Neural_Network,numeric'
Get_Cost(object, target)
Arguments
object |
... a Neural_Network that has run forward_prop at least once |
target |
... a numeric vector ... the target ... |
Value
Numeric
log loss
Description
get log loss
Usage
Get_LogLoss(predictions, target)
Arguments
predictions |
is a numeric vector |
target |
is a numeric vector |
Value
Numeric
num observs
Description
returns the number of observations that the network has processed
Usage
## S4 method for signature 'Neural_Network'
Get_Number_Observations(object)
Arguments
object |
... a Neural Network that has called fprop. ie. that has called train/predict |
Value
Numeric
Neural Network implementation
Description
Neural Network implementation
predict stuff
Description
returns predictions
Usage
## S4 method for signature 'Neural_Network,data.frame'
Predict(object, dataset)
Arguments
object |
: a neural network |
dataset |
: a dataframe of features and observations |
Value
Numeric
start script
Description
main function that runs the interactive script
Usage
Start()
Details
takes your numerai training data and trains a neural network to your architectural specifications. provides you with the out of sample error offers to retrain with a new architecture or predict on a numerai tournament dataset. Can then write the predictions to a CSV
train the NN
Description
gets NN parameters that minimize cost on dataset using optimization_method
Usage
## S4 method for signature 'Neural_Network,data.frame,numeric,numeric,numeric'
Train(object,
dataset, regularization_constant, learning_rate, tolerable_error)
Arguments
object |
is a Neural Network |
dataset |
is a data.frame, the original data frame that includes the target |
regularization_constant |
is a numeric |
learning_rate |
is a numeric |
tolerable_error |
is a numeric, units : log loss |
Value
Neural_Network
back prop
Description
updates connection strengths using results of last forward prop
Usage
## S4 method for signature 'Neural_Network,numeric,numeric,numeric'
back_propogation(object,
target, regularization_parameter, learning_rate)
Arguments
object |
is a Neural_Network |
target |
is a numeric vector |
regularization_parameter |
is non-negative number punishes strong connections |
learning_rate |
is a positive number that controls the rate at which connections are adjusted |
Value
Neural_Network
f_prop
Description
... part of the training program
Usage
## S4 method for signature 'Neural_Network,matrix'
forward_propogation(object, dataset)
Arguments
object |
is a Neural_Network |
dataset |
is a matrix not containing the target vector |
Value
Neural_Network
init
Description
initalizes a neural network capable of studying datasets with ncol = to the ncol(sample_dataset) and making predictions on such datasets
Usage
## S4 method for signature 'Neural_Network'
initialize(.Object, number_predictors,
hidden_layer_lengths)
Arguments
.Object |
... a Neural_Network object |
number_predictors |
... a numeric telling how many preditors there are |
|
... a numeric telling the number of layers and the number of neurons in each layer |
Details
NN is parametrized by its connection_strength matrices
Value
Neural_Network