aml_make_model: Function to train Deep Learning regression model for a single...
In lazytrade: Learn Computer and Data Science using Algorithmic Trading

Description Usage Arguments Details Value Author(s) Examples

Function is training h2o deep learning model to match future prices of the asset to the indicator pattern. Main idea is to be able to predict future prices by solely relying on the recently retrieved indicator pattern. This is to mimic traditional algorithmic systems based on the indicator rule and to attempt automated optimization with AI.

aml_make_model(
  symbol,
  timeframe = 60,
  path_model,
  path_data,
  force_update = FALSE,
  objective_test = FALSE,
  num_nn_options = 12,
  fixed_nn_struct = c(100, 100),
  num_epoch = 100,
  num_bars_test = 600,
  num_bars_ahead = 34,
  num_cols_used = 16,
  min_perf = 0.3
)

`symbol`	Character symbol of the asset for which to train the model
`timeframe`	Integer, value in minutes, e.g. 60 min
`path_model`	Path where the models shall be stored
`path_data`	Path where the aggregated historical data is stored, if exists in rds format
`force_update`	Boolean, by setting this to TRUE function will generate new model (useful after h2o engine update)
`objective_test`	Boolean, option to use trading objective test as a parameter to validate best model, defaults to FALSE
`num_nn_options`	Integer, value from 0 to 24 or more. Used to change number of variants of the random neural network structures, when value is 0 uses fixed structure Higher number may lead to long code execution. Select value multiple of 3 otherwise function will generate warning. E.g. 12, 24, 48, etc
`fixed_nn_struct`	Integer vector with numeric elements, see par hidden in ?h2o.deeplearning, default value is c(100,100). Note this will only work if num_nn_options is 0
`num_epoch`	Integer, see parameter epochs in ?h2o.deeplearning, default value is 100 Higher number may lead to long code execution
`num_bars_test`	Integer, value of bars used for model testing
`num_bars_ahead`	Integer, value to specify how far should the function predict. Default 34 bars.
`num_cols_used`	Integer, number of columns to use for training the model, defaults to 16
`min_perf`	Double, value greater than 0. Used to set minimum value of model performance. Higher value will increase computation time

Deep learning model structure is obtained from several random combinations of neurons within 3 hidden layers of the network. The most accurate model configuration will be automatically selected based either RMSE or Objective Test. In addition, the function will check if there is a need to update the model. To do that function will check results of the function aml_test_model.R.

Function is using the dataset prepared by the function aml_collect_data.R. Note that function will start to train the model as soon as there are more than 1000 rows in the dataset

Function is writing a file object with the best Deep Learning Regression model

library(dplyr)
library(readr)
library(h2o)
library(lazytrade)
library(lubridate)
library(magrittr)

path_model <- normalizePath(tempdir(),winslash = "/")
path_data <- normalizePath(tempdir(),winslash = "/")

ind = system.file("extdata", "AI_RSIADXUSDJPY60.csv",
                  package = "lazytrade") %>% read_csv(col_names = FALSE)

ind$X1 <- ymd_hms(ind$X1)

tick = system.file("extdata", "TickSize_AI_RSIADX.csv",
                  package = "lazytrade") %>% read_csv(col_names = FALSE)

write_csv(tick, file.path(path_data, "TickSize_AI_RSIADX.csv"), col_names = FALSE)

# data transformation using the custom function for one symbol
aml_collect_data(indicator_dataset = ind,
                 symbol = 'USDJPY',
                 timeframe = 60,
                 path_data = path_data)

# dataset will be written to the temp directory

# start h2o engine
h2o.init(nthreads = 2)


# performing Deep Learning Regression using 2 random neural network structures and objective test
aml_make_model(symbol = 'USDJPY',
               timeframe = 60,
               path_model = path_model,
               path_data = path_data,
               force_update=FALSE,
               objective_test = TRUE,
               num_nn_options = 6,
               num_epoch = 10,
               min_perf = 0,
               num_bars_test = 600,
               num_bars_ahead = 34,
               num_cols_used = 16)

# performing DL Regression using 2 random neural network structures
# with objective test, all columns
aml_make_model(symbol = 'USDJPY',
               timeframe = 60,
               path_model = path_model,
               path_data = path_data,
               force_update=FALSE,
               objective_test = TRUE,
               num_nn_options = 6,
               num_epoch = 10,
               min_perf = 0,
               num_bars_test = 600,
               num_bars_ahead = 34,
               num_cols_used = 0)


# performing Deep Learning Regression using the custom function
aml_make_model(symbol = 'USDJPY',
               timeframe = 60,
               path_model = path_model,
               path_data = path_data,
               force_update=FALSE,
               objective_test = FALSE,
               num_nn_options = 6,
               num_epoch = 10,
               min_perf = 0,
               num_bars_test = 600,
               num_bars_ahead = 34,
               num_cols_used = 16)

# performing Deep Learning Regression, fixed mode
aml_make_model(symbol = 'USDJPY',
               timeframe = 60,
               path_model = path_model,
               path_data = path_data,
               force_update=TRUE,
               num_nn_options = 0,
               fixed_nn_struct = c(100, 100),
               num_epoch = 10,
               min_perf = 0)

# stop h2o engine
h2o.shutdown(prompt = FALSE)

#set delay to insure h2o unit closes properly before the next test
Sys.sleep(5)