Package 'countfitteR'

countfitteR - a framework for fitting count distributions in R

Description

The countfitteR package is a toolbox for the analysis of count data.

Acknowledgements

countfitteR is a wrapper around existing count models in R. To standardize error messages and ease up the integration, we slightly modified the zeroinfl function by Achim Zeileis.

Author(s)

Jaroslaw Chilimoniuk, Stefan Roediger, Michal Burdukiewcz

Examples

set.seed(15390)
library(countfitteR)
df <- data.frame(pois = rpois(25, 0.3), 
                 binom = rbinom(25, 1, 0.8))

cmp <- compare_fit(df, fitlist = fit_counts(df, model = "all"))

---

Short version of the case_study_FITC

Description

shorter version of the case_study_FITC. Used as an example in shiny app, when the user will not load his own count data.

Usage

case_study

---

Case study with two fluorescent dyes

Description

example data extracted from Aklides system and merged into one file. Counts in this file will not fit properly, due to the fact that we integrated into the file counts with two different fluorescent dyes used.

Usage

case_study_all

---

Case study for APC dye

Description

example data extracted from Aklides system. Counts with only APC fluorescent dye were merged.

Usage

case_study_APC

---

Case study for FITC dye

Description

example data extracted from Aklides system. Counts with only FITC fluorescent dye were merged.

Usage

case_study_FITC

---

Compare fits

Description

Compare empirical distribution of counts with the distribution defined by the model fitted to counts.

Usage

compare_fit(count_list, fitlist = fit_counts(count_list, model = "all"))

Arguments

count_list	A list of counts. Each count should be in separate column, rows should represent values of these counts.
fitlist	a list of fits, as created by fit_counts.

Value

A data.frame with distribution values for each unique count. Count is the name of the original count, model is the name of distribution model, x is unique count value, n is the frequency of unique counts, value is result of calculations made by chosen distribution model.

Examples

df <- data.frame(poisson = rpois(25, 0.3), binomial = rbinom(25, 1, 0.8))
compare_fit(df, fitlist = fit_counts(df, model = "all"))

---

countfitteR Graphical User Interface

Description

Launches graphical user interface that analyses given count data and chooses the best performing distribution model.

Usage

countfitteR_gui()

Warning

Any ad-blocking software may cause malfunctions.

Author(s)

Jaroslaw Chilimoniuk, Stefan Roediger, Michal Burdukiewcz

Examples

if(interactive()) { 
  countfitteR_gui()
}

---

Make a decision based on the BIC value

Description

Select the most appropriate distribution for the count data in the html-friendly format.

Usage

decide(summary_fit, separate)

Arguments

summary_fit	a result of the summary_fitlist function.
separate	logical. If TRUE, each count is separately fitted to the model. If FALSE, all counts are fitted to the same models having the count name as the independent variable.

See Also

fit_counts

Examples

df <- data.frame(poisson = rpois(25, 0.3), binomial = rbinom(25, 1, 0.8))
fc <- fit_counts(df, model = "all") 
summ <- summary_fitlist(fc) 
decide(summ, separate = FALSE)

---

Fit counts to distributions

Description

Fit counts to distributions

Usage

fit_counts(counts_list, separate = TRUE, model, level = 0.95, ...)

Arguments

counts_list	A list of count data. Each count should be in separate column, rows should represent values of that counts.
separate	logical. If TRUE, each count is separately fitted to the model. If FALSE, all counts are fitted to the same models having the count name as the independent variable.
model	single character: "pois", "nb", "zinb", "zip", "all". If "all", all possible model are fitted.
level	Confidence level, default is 0.95.
...	Dots parameters are ignored.

Value

The list of fitted models. Names are names of original counts, an underline and a name of model used. confint is a matrix with the number of rows equal to the number of parameters. Rownames are names of parameters. The columns contain respectively lower and upper confidence intervals.

Examples

df <- data.frame(poisson = rpois(25, 0.3), binomial = rbinom(25, 1, 0.8))
fit_counts(df, model = "pois")

---

plot_fitcmp

Description

Compare empirical distribution of counts with the distribution defined by the model fitted to counts. The bar charts represent theoretical counts depending on the chosen distribution. Red dots describe the real number of counts.

Usage

plot_fitcmp(fitcmp)

Arguments

fitcmp

You need to input data frame that is created by compare_fit function.

Examples

df <- data.frame(poisson = rpois(25, 0.3), binomial = rbinom(25, 1, 0.8))
fitcmp <- compare_fit(df, fitlist = fit_counts(df, model = "all"))
plot_fitcmp(fitcmp)

---

Process counts

Description

Converts data in a table-like formats into lists of counts.

Usage

process_counts(x)

Arguments

x	data.frame or matrix.

Details

case_study does not consider NAs and NaNs effectively omitting them (as per the is.na function).

Value

A list of counts.

Examples

data(case_study)
process_counts(case_study)

---

Select the most appropriate model

Description

Select the most appropriate model

Usage

select_model(fitlist)

Arguments

fitlist

a list of fits, as created by fit_counts.

Value

a data.frame with two columns: count representing the name of the count and chosen model with the model with the lowest BIC.

Examples

set.seed(1)
df <- data.frame(poisson1 = rpois(50, 2), 
                 poisson2 = rpois(50, 5),
                 zip1 = rZIP(50, 2, 0.7),
                 zip2 = rZIP(50, 5, 0.7))
fitlist_separate <- fit_counts(df, model = c("pois", "zip")) 
select_model(fitlist_separate)

---

Data created from simulation of NB Poiss

Description

Data created from simulation of NB Poiss

Usage

sim_dat

Examples

# code used to generate the data
# be warned: the simulations will take some time
## Not run: 
library(dplyr)
set.seed(15390)
sim_dat <- do.call(rbind, lapply(10^(-3L:2), function(single_theta)
  do.call(rbind, lapply(1L:10/2, function(single_lambda) 
    do.call(rbind, lapply(1L:100, function(single_rep) {
      
      foci <- lapply(1L:10, function(dummy) rnbinom(600, size = single_theta, mu = single_lambda))
      names(foci) <- paste0("C", 1L:10)
      
      fit_counts(foci, separate = TRUE, model = "all") %>%
        summary_fitlist %>% 
        mutate(between = single_lambda < upper & single_lambda > lower) %>%
        group_by(model) %>% 
        summarize(prop = mean(between)) %>%
        mutate(replicate = single_rep, lambda = single_lambda, theta = single_theta)
    }))
  ))
))

## End(Not run)

---

Summary of estimates

Description

Counts are fitted to model(s) using the count name as the explanatory variable. Estimates are presented in the table below along with the BIC values of their models. Estimated coefficients of models (lambda for all distributions, theta for NB and ZINB, r for ZIP and ZINB).

Usage

summary_fitlist(fitlist)

Arguments

fitlist

a list of fits, as created by fit_counts.

Value

Data frame with summarised results of all distribution models.

• Count: the name of the original count.

• lambda: $\lambda$ - Poisson mean, lower and upper confidence intervals.

• BIC: Bayesian information criterion

• theta: $\theta$ - dispersion parameter

• r: probability of excess zeros.

See Also

fit_counts

Examples

df <- data.frame(poisson = rpois(25, 0.3), binomial = rbinom(25, 1, 0.8))
fc <- fit_counts(df, model = "all") 
summary_fitlist(fc)

---

Validate data

Description

Validates count data.

Usage

validate_counts(x)

Arguments

x	data.frame or matrix.

Details

Errors if x has negative values or non-numeric values, otherwise TRUE.

Value

An input object.

Examples

data(case_study)
process_counts(case_study)

---

Zero-inflated negative binomial distrbution

Description

Density and random generation for the zero-inflated negative binomial distribution.

Usage

rZINB(n, size, mu, r)

dZINB(x, size, mu, r)

Arguments

n	number of random values to return.
size	target for number of successful trials, or dispersion parameter (the shape parameter of the gamma mixing distribution). Must be strictly positive, need not be integer..
mu	mean.
r	probability of excess zeros.
x	vector of (non-negative integer) quantiles.

See Also

Negative binomial distribution: NegBinomial.

Examples

rZINB(15, 1.9, 0.9, 0.8)

---

Zero-inflated Poisson distrbution

Description

Density and random generation for the zero inflated Poisson distribution.

Usage

dZIP(x, lambda, r)

rZIP(n, lambda, r)

Arguments

x	vector of (non-negative integer) quantiles.
lambda	vector of (non-negative) means.
r	probability of excess zeros.
n	number of random values to return.

See Also

Poisson distribution: Poisson.

Examples

rZIP(15, 1.9, 0.9)

---