Bayesian Paired Comparison Analysis with Stan


[Up] [Top]

Documentation for package ‘bpcs’ version 1.0.0

Help Pages

bpcs-package bpcs - A package for Bayesian Paired Comparison analysis with Stan
bpc Bayesian Paired comparison regression models in Stan
brasil_soccer_league This is a dataset with the results matches fromo the first league of the Brazilian soccer championship from 2017-2019. It was reduced and translatedfrom the adaduque/Brasileirao_Dataset repository
check_if_there_are_na Check for NA in the specfic columns and returns T or F is there is at least 1 NA in those columns
check_if_there_are_ties Check if a data frame column contains ties
check_numeric_predictor_matrix Check if all values in the predictor matrix are numeric and not NA. Note that TRUE will be cast to 1 and FALSE will be cast to 0
check_predictors_df_contains_all_players Check if the predictor df contains all players and only those
check_result_column Check if a data frame column contains only the values 1 0 and 2. Used to check the format of the results
check_z_column Check if a data frame column contains only the values 1 or 0. For the z column
compute_scores Giving a player0 an player1 scores, this functions adds one column to the data frame containing who won (0= player0 1=player1 2=tie) and another if it was a tie. The ties column superseeds the y column. If it was tie the y column does not matter y column: (0= player0 1=player1 2=tie) ties column (0=not tie, 1=tie)
compute_ties Giving a result column we create a new column with ties (0 and 1 if it has)
create_array_of_par_names Create an array with the parameter name and to what player/cluster it refers to in the order stan presents
create_bpc_object Defines the class bpc and creates the bpc object. To create we need to receive some defined parameters (the arguments from the bpc function), a lookup table and a the stanfit object generated from the rstan sampling procedure
create_cluster_index Create two columns with the indexes for the names of the players Here we create a new lookup table. Should be used when sampling the parameters
create_cluster_index_with_existing_lookup_table Create two columns with the indexes for the names Here we use an existing lookup table. Should be used in predicting
create_index Create two columns with the indexes for the names of the players Here we create a new lookup table. Should be used when sampling the parameters
create_index_cluster_lookuptable Create a lookup table of names and indexes Note that the indexes will be created in the order they appear. For string this does not make much difference but for numbers the index might be different than the actual number that appears in names
create_index_lookuptable Create a lookup table of names and indexes Note that the indexes will be created in the order they appear. For string this doesnt make much difference but for numbers the index might be different than the actual number that appears in names
create_index_predictors_with_lookup_table Receives one column with player names and returns a data frame with the relevant index columns based on a given lookup table To be used with the predictors data frame
create_index_with_existing_lookup_table Create two columns with the indexes for the names Here we use an existing lookup table. Should be used in predicting
create_predictors_lookup_table Receives a vector with predictors strings (the column names) and returns a predictor_lookup_table
create_predictor_matrix_with_player_lookup_table Receives a predictor dataframe, a string with the column of the player, a vector of strings with the columns for the predictors and a lookup table and returns an ordered matrix for Stan To be used with the predictors data frame
expand_aggregated_data Expand aggregated data Several datasets for the Bradley-Terry Model aggregate the number of wins for each player in a different column. The models we provide are intended to be used in a long format. A single result for each contest. This function expands datasets that have aggregated data into this long format.
get_hpdi_parameters Return the mean and the HPDI of the parameters of the model
get_loo Tiny wrapper for the PSIS-LOO-CV method from the loo package.
get_model_parameters Return all the name of parameters in a model from a bpc_object. Here we exclude the log_lik and the lp__ since they are not parameters of the model
get_probabilities Get the empirical win/draw probabilities based on the ability/strength parameters. Instead of calculating from the probability formula given from the model we create a predictive posterior distribution for all pair combinations and calculate the posterior wins/loose/draw The function returns the mean value of win/loose/draw for the player i. To calculate for player j the probability is 1-p_i
get_rank_of_players Generate a ranking of the ability based on sampling the posterior distribution of the ranks.
get_sample_posterior Get the posterior samples for a parameter of the model.
get_stanfit Retrieve the stanfit object generated by rstan.
get_stanfit_summary Get stanfit summary table of all parameters excluding log_lik.
get_waic Tiny wrapper for the WAIC method from the loo package.
HPDI_from_stanfit Calculate HPDI for all parameters from a stanfit object Here we use the coda package
HPD_higher_from_column Returns the higher value of the HPD interval for a data frame column
HPD_lower_from_column Returns the lower value of the HPD interval for a data frame column
inv_logit Inverse logit function
launch_shinystan Tiny wrapper to launch a shinystan app to investigate the MCMC.
logit Logit function
match_cluster_names_to_cluster_lookup_table Receives a column with cluster names and returns a data frame with the relevant index column based on a given cluster lookup table
match_player_names_to_lookup_table Receives two columns with player names and returns a data frame with the relevant index columns based on a given lookup table
optimization_algorithms Dataset containing an example of the performance of different optimization algorithms against different benchmark functions. This is a reduced version of the dataset presented at the paper: "Statistical Models for the Analysis of Optimization Algorithms with Benchmark Functions.". For details on how the data was collected we refer to the paper.
predict.bpc Predict results for new data.
print.bpc Print method for the bpc object.
replace_parameter_index_with_names Replace the name of the parameter from index to name using a lookup_table Receives a data frame and returns a dataframe.
sample_stanfit Return a data frame by resampling the posterior from a stanfit Here we select a parameter, retrieve the all the posterior from the stanfit and then we resample this posterior n times
summary.bpc Summary of the model bpc model.
tennis_agresti This is the expansion of the tennis data from Agresti (2003) p.449 This data refers to matches for several women tennis players during 1989 and 1990