README.md

In ntyndall/footballstats: Classify and predict football results

Code status

Installing and running this package

# Run this as a script

# Install football stats from github
cat(' ## Installing footballstats \n ## ')

library(devtools)

devtools::install_github(
  repo = "ntyndall/footballstats"
)

library(footballstats)

cat(' ## Install Complete. \n\n')

# Set up log directory
footballstats::create_log_dir()

Deployment

• Download and install redis then from the command line start it by typing

cd redis-stable/utils
sudo ./install_server.sh

To start and stop the redis server type sudo service redis_6379 start / stop - Edit the /usr/lib/R/etc/Rprofile.site file and include the following 3 tokens

Sys.setenv(FS_HOST = "___")
Sys.setenv(FS_APIKEY = "Authorization=___")
Sys.setenv(FS_SLACK = "___")
Sys.setenv(FS_DEPLOYLOC = "___")

where the last token is the root path of where the above script is run from, e.g. /root/. - Set up a cron job by typing crontab -e with the following...

# Collect football information - such as match / team / commentary information THEN predict matches
0 22 * * 0 Rscript -e 'library(footballstats); footballstats::analyse_and_predict(deployed = TRUE)'

# After information has been gathered - update player status etc (COSTLY operation!)
0 22 * * 1 Rscript -e 'library(footballstats); footballstats::analyse_players(deployed = TRUE)'

# Generate report
# ...

Unfortunately, the size of the XGBoost model being stored in the R package leads to memory issues when lazy loading - to save some cash on large memory boxes, I can just scp mymodels/xgModel.rda root@IP:/root the file across and load it while performing the analysis so I don't have to lazy load it.

Tagging a release

• Update DESCRIPTION file to required version (must be > than most recent GIT version).
• Run Rscript -e "organisR::tag()" inside root directory /footballstats/.
• Update Github with the newest tag and binary that is produced from step 2.

Scripts

There are a number of useful scripts for testing and building models. Make sure that redis is running on port 6379.

• In order to run the tests on a local machine, run from the root directory;
  • Rscript demo/run_tests.R
• To rebuild one of the data models with updated information in redis, run from the root directory
  • Rscript demo/create_nn.R

Accuracy

Below will be information on the accuracy so far that the app is achieving. To add more results, run through the data sets and append to the data/accuracyFrame data set in a similar fashion and from the root directory create the plot by running Rscript demo/create_acc_plot.R and push the new image to the repo.

ntyndall/footballstats documentation built on Aug. 13, 2019, 1:36 p.m.