code_smells.Rmd

---
title: Code smells
---

```{r cs-include-setup, include=FALSE}
source(knitr::purl('setup.Rmd', output = tempfile()))
```

This section covers the analysis relating to the _code smells_ metric.

## Data preparation

```{r cs-prepare-data}
model_seed <- 20230302  # 2023 March 2

create_code_smells_data <- function(data) {
  list(
    code_smells_kloc_std=scale(data$code_smells_kloc),
        contributors_std=scale(data$contributors),
               stars_std=scale(data$stars),
                 age_std=scale(get_age(data$created_at))
  )
}

data <- create_code_smells_data(df)
data$category <- factor(df$category)
```

## Candidate models {.tabset}

### CS1

```{r cs-model-1}
fit_model_cs1 <- function(name, data) {
  fit_model(name,
            "code_smells_kloc_std ~ 0 + category",
            c(
              prior(normal(0, 1), class=b),
              prior(exponential(1), class=sigma)
            ),
            data,
            model_seed)
}

cs1 <- fit_model_cs1("cs1", data)
```

### CS2

```{r cs-model-2}
fit_model_cs2 <- function(name, data) {
  fit_model(name,
            "code_smells_kloc_std ~ 0 + category + contributors_std",
            c(
              prior(normal(0, 1), class=b),
              prior(normal(0, 1), class=b, coef="contributors_std"),
              prior(exponential(1), class=sigma)
            ),
            data,
            model_seed)
}

cs2 <- fit_model_cs2("cs2", data)
```

### CS3

```{r cs-model-3}
fit_model_cs3 <- function(name, data) {
  fit_model(name,
            "code_smells_kloc_std ~ 0 + category + stars_std",
            c(
              prior(normal(0, 1), class=b),
              prior(normal(0, 1), class=b, coef="stars_std"),
              prior(exponential(1), class=sigma)
            ),
            data,
            model_seed)
}

cs3 <- fit_model_cs3("cs3", data)
```

### CS4

```{r cs-model-4}
fit_model_cs4 <- function(name, data) {
  fit_model(name,
            "code_smells_kloc_std ~ 0 + category + age_std",
            c(
              prior(normal(0, 1), class=b),
              prior(normal(0, 1), class=b, coef="age_std"),
              prior(exponential(1), class=sigma)
            ),
            data,
            model_seed)
}

cs4 <- fit_model_cs4("cs4", data)
```

### CS5

```{r cs-model-5}
fit_model_cs5 <- function(name, data) {
  fit_model(name,
            "code_smells_kloc_std ~ 0 + category + contributors_std + stars_std",
            c(
              prior(normal(0, 1), class=b),
              prior(normal(0, 1), class=b, coef="contributors_std"),
              prior(normal(0, 1), class=b, coef="stars_std"),
              prior(exponential(1), class=sigma)
            ),
            data,
            model_seed)
}

cs5 <- fit_model_cs5("cs5", data)
```

### CS6

```{r cs-model-6}
fit_model_cs6 <- function(name, data) {
  fit_model(name,
            "code_smells_kloc_std ~ 0 + category + contributors_std + age_std",
            c(
              prior(normal(0, 1), class=b),
              prior(normal(0, 1), class=b, coef="contributors_std"),
              prior(normal(0, 1), class=b, coef="age_std"),
              prior(exponential(1), class=sigma)
            ),
            data,
            model_seed)
}

cs6 <- fit_model_cs6("cs6", data)
```

### CS7

```{r cs-model-7}
fit_model_cs7 <- function(name, data) {
  fit_model(name,
            "code_smells_kloc_std ~ 0 + category + contributors_std + stars_std + age_std",
            c(
              prior(normal(0, 1), class=b),
              prior(normal(0, 1), class=b, coef="contributors_std"),
              prior(normal(0, 1), class=b, coef="stars_std"),
              prior(normal(0, 1), class=b, coef="age_std"),
              prior(exponential(1), class=sigma)
            ),
            data,
            model_seed)
}

cs7 <- fit_model_cs7("cs7", data)
```

## Model comparison

```{r cs-compare-models}
cs1 <- add_criterion(cs1, criterion="loo", moment_match=TRUE)
cs2 <- add_criterion(cs2, criterion="loo", moment_match=TRUE)
cs3 <- add_criterion(cs3, criterion="loo", moment_match=TRUE)
cs4 <- add_criterion(cs4, criterion="loo", moment_match=TRUE)
cs5 <- add_criterion(cs5, criterion="loo", moment_match=TRUE)
cs6 <- add_criterion(cs6, criterion="loo", moment_match=TRUE)
cs7 <- add_criterion(cs7, criterion="loo", moment_match=TRUE)

print(loo_compare(cs1, cs2, cs3, cs4, cs5, cs6, cs7, criterion="loo"),
      simplify=FALSE)
```

## Diagnostics {.tabset}

### Prior predictive check

```{r cs-ppc}
plot_priors(cs6, "Code smells per KLOC (std)")
```

### MCMC chain trace plots

```{r b-caterpillars, message=FALSE}
plot_caterpillar(cs6, "b_category1", "N")
plot_caterpillar(cs6, "b_category2", "A")
plot_caterpillar(cs6, "b_category3", "AE")
plot_caterpillar(cs6, "b_category5", "AEP")
plot_caterpillar(cs6, "b_category7", "AEP+")
```

## Full population

This section covers the analysis using the full sample population.

### Estimates {.tabset}

#### Standardized

```{r cs-standardized-estimates}
summary(cs6)
```

#### Natural scale

```{r cs-natural-scale-estimates}
print(create_natural_scale_summary_df(cs6, df$code_smells_kloc))
```

### Plots {.tabset}

#### Major category intervals

```{r cs-plot-intervals-of-major-categories, message=FALSE, fig.height=2.8}
plot_intervals_of_major_categories(cs6) +
  labs(title="CS6", x="Code smells per KLOC (std)")
```

#### Binary warning usage

```{r cs-warnings-vs-no-warnings, message=FALSE, fig.height=2}
data_binary_factor  <- create_code_smells_data(df)
data_binary_factor$category <- factor(uses_warnings(df$category))

cs6_binary_factor <- fit_model_cs6("cs6_factor_uses_warnings",
                                     data_binary_factor)

plot_intervals_of_categories(cs6_binary_factor) +
  labs(title="CS6", x="Code smells per KLOC (std)") +
  scale_y_discrete(labels=c(
                     "b_category1"="No warnings",
                     "b_category2"="Any warnings"
                   ))
```

## Culled population

There are outliers in category AEP and AEP+, "clipp" and "hana", which bias the analysis results.
Here we remove these samples and fit CS2 to the trimmed data.

```{r cs-remove-outliers, message=FALSE}
df_culled <- df
df_culled <- df_culled[df_culled$name != "clipp",]
df_culled <- df_culled[df_culled$name != "hana",]

culled_data <- create_code_smells_data(df_culled)
culled_data$category <- factor(df_culled$category)

cs6_culled <- fit_model_cs6("cs6_no_outliers", culled_data)
```

### Estimates {.tabset}

#### Standardized

```{r cs-culled-standardized-estimates, message=FALSE}
summary(cs6_culled)
```

#### Natural scale

```{r cs-culled-natural-scale-estimates, message=FALSE}
print(create_natural_scale_summary_df(cs6_culled, df_culled$code_smells_kloc))
```

### Plots {.tabset}

#### Major category intervals

```{r cs-plot-culled-intervals, message=FALSE, fig.height=2.8}
plot_intervals_of_major_categories(cs6_culled) +
  labs(title="CS6 (culled)", x="Code smells per KLOC (std)")
```

#### Binary warning usage

```{r cs-plot-culled-binary-factor, message=FALSE, fig.height=2}
data_culled_binary_factor <- create_code_smells_data(df_culled)
data_culled_binary_factor$category <- factor(uses_warnings(df_culled$category))

cs6_culled_binary_factor <-
  fit_model_cs6("cs6_data_no_outliers_factor_uses_warnings",
                data_culled_binary_factor)

plot_intervals_of_categories(cs6_culled_binary_factor) +
  labs(title="CS6 (culled)", x="Code smells per KLOC (std)") +
  scale_y_discrete(labels=c(
                     "b_category1"="No warnings",
                     "b_category2"="Any warnings"
                   ))
```