Skip to contents

Score calculation

Source: R/QR_functions.R
compute_score.Rd

To calculate a score for each series from a quality report

Usage

# S3 method for class 'QR_matrix'
compute_score(
  x,
  score_pond = c(qs_residual_sa_on_sa = 30, f_residual_sa_on_sa = 30, qs_residual_sa_on_i
    = 20, f_residual_sa_on_i = 20, f_residual_td_on_sa = 30, f_residual_td_on_i = 20,
    oos_mean = 15, oos_mse = 10, residuals_independency = 15, residuals_homoskedasticity
    = 5, residuals_skewness = 5, m7 = 5, q_m2 = 5),
  modalities = c("Good", "Uncertain", "", "Bad", "Severe"),
  normalize_score_value,
  na.rm = FALSE,
  n_contrib_score,
  conditional_indicator,
  ...
)

# S3 method for class 'mQR_matrix'
compute_score(x, ...)

Arguments

x

a QR_matrix or mQR_matrix object.

score_pond

the formula used to calculate the series score.

modalities

modalities ordered by importance in the score calculation (cf. details).

normalize_score_value

integer indicating the reference value for weights normalisation. If missing, weights will not be normalised.

na.rm

logical indicating whether missing values must be ignored when calculating the score.

n_contrib_score

integer indicating the number of variables to create in the quality report's values matrix to store the n_contrib_score greatest contributions to the score (cf. details). If not specified, no variable is created.

conditional_indicator

a list containing 3-elements sub-lists: "indicator", "conditions" and "condition_modalities". To reduce down to 1 the weight of chosen indicators depending on other variables' values (cf. details).

...

other unused parameters.

Value

a QR_matrix or mQR_matrix object.

Details

The function compute_score calculates a score from the modalities of a quality report: to each modality corresponds a weight that depends on the parameter modalities. The default parameter is c("Good", "Uncertain", "Bad","Severe"), and the associated weights are respectively 0, 1, 2 and 3.

The score calculation is based on the score_pond parameter, which is a named integer vector containing the weights to apply to the (modalities matrix) variables. For example, with score_pond = c(qs_residual_sa_on_sa = 10, f_residual_td_on_sa = 5), the score will be based on the variables qs_residual_sa_on_sa and f_residual_td_on_sa. The qs_residual_sa_on_sa grades will be multiplied by 10 and the f_residual_td_on_sa grades, by 5. To ignore the missing values when calculating a score, use the parameter na.rm = TRUE.

The parameter normalize_score_value can be used to normalise the scores. For example, to have all scores between 0 and 20, specify normalize_score_value = 20.

When using parameter n_contrib_score, n_contrib_score new variables are added to the quality report's values matrix. These new variables store the names of the variables that contribute the most to the series score. For example, n_contrib_score = 3 will add to the values matrix the three variables that contribute the most to the score. The new variables' names are i_highest_score, with i being the rank in terms of contribution to the score (1_highest_score contains the name of the greatest contributor, 2_highest_score the second greatest, etc). Only the variables that have a non-zero contribution to the score are taken into account: if a series score is 0, all i_highest_score variables will be empty. And if a series score is positive only because of the m7 statistic, 1_highest_score will have a value of "m7" for this series and the other i_highest_score will be empty.

Some indicators are only relevant under certain conditions. For example, the homoscedasticity test is only valid when the residuals are independant, and the normality tests, only when the residuals are both independant and homoscedastic. In these cases, the parameter conditional_indicator can be of use since it reduces the weight of some variables down to 1 when some conditions are met. conditional_indicator is a list of 3-elements sub-lists:

  • "indicator": the variable whose weight will be conditionally changed

  • "conditions": the variables used to define the conditions

  • "conditions_modalities": modalities that must be verified to induce the weight change For example, conditional_indicator = list(list(indicator = "residuals_skewness", conditions = c("residuals_independency", "residuals_homoskedasticity"), conditions_modalities = c("Bad","Severe"))), reduces down to 1 the weight of the variable "residuals_skewness" when the modalities of the independancy test ("residuals_independency") or the homoscedasticity test ("residuals_homoskedasticity") are "Bad" or "Severe".

See also

Traduction française

Examples

# Path of matrix demetra_m
demetra_path <- file.path(
    system.file("extdata", package = "JDCruncheR"),
    "WS/ws_ipi/Output/SAProcessing-1",
    "demetra_m.csv"
)

# Extract the quality report from the demetra_m file
QR <- extract_QR(demetra_path)

# Calculer le score
QR <- compute_score(QR, n_contrib_score = 2)
print(QR)
#> The quality report matrix has 13 observations
#> There are 19 indicators in the modalities matrix and 23 indicators in the values matrix
#> 
#> The quality report matrix contains the following variables:
#> series  qs_residual_sa_on_sa  f_residual_sa_on_sa  qs_residual_sa_on_i  f_residual_sa_on_i  f_residual_td_on_sa  f_residual_td_on_i  residuals_independency  residuals_normality  residuals_homoskedasticity  residuals_skewness  residuals_kurtosis  oos_mean  oos_mse  m7  q  q_m2  pct_outliers  frequency  arima_model  score  1_highest_contrib_score  2_highest_contrib_score
#> 
#> The variables exclusively found in the values matrix are:
#> frequency  arima_model  1_highest_contrib_score  2_highest_contrib_score
#> 
#> The smallest score is 75 and the greatest is 515
#> The average score is 318.462 and its standard deviation is 158.224
#> 
#> The following formula was used to calculate the score:
#> 30 * qs_residual_sa_on_sa + 30 * f_residual_sa_on_sa + 20 * qs_residual_sa_on_i + 20 * f_residual_sa_on_i + 30 * f_residual_td_on_sa + 20 * f_residual_td_on_i + 15 * oos_mean + 10 * oos_mse + 15 * residuals_independency + 5 * residuals_homoskedasticity + 5 * residuals_skewness + 5 * m7 + 5 * q_m2

# Extract the modalities matrix:
QR$modalities$score
#>  [1] 140 105 280 245  75 240 460 505 460 515 300 305 510