---
title: "Using Validators in optree"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Using Validators in optree}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
```

# Using Validators in optree

Validators are functions that check if option values are valid before they are set. optree provides a collection of built-in validators for common validation patterns, and also allows you to write custom validators for specialized validation logic.

## Why Use Validators?

Validators ensure that your options maintain internal consistency and meet your requirements. They:

- **Prevent invalid states** – Catch errors early before invalid options propagate through your code
- **Enforce constraints** – Ensure numeric values are in valid ranges, vectors have required lengths, etc.
- **Maintain relationships** – Ensure interdependent options stay synchronized (e.g., x and y vectors must have equal length)
- **Provide clear error messages** – Give users helpful feedback about what went wrong

## Built-in Validators

optree provides seven ready-to-use validators:

### Scalar Validators

#### `v_numeric_scalar()`
Validates that a value is a single numeric value (not a vector, not a string).

```{r, eval=FALSE}
validator <- v_numeric_scalar()
validator(42)           # ✓ OK
validator(c(1, 2))      # ✗ Error: must be a single numeric value
validator("text")       # ✗ Error: must be a single numeric value
```

#### `v_logical_scalar()`
Validates that a value is a single logical value (TRUE or FALSE).

```{r, eval=FALSE}
validator <- v_logical_scalar()
validator(TRUE)         # ✓ OK
validator(c(TRUE, FALSE))  # ✗ Error: must be a single logical value
validator(1)            # ✗ Error: must be a single logical value
```

#### `v_character_scalar()`
Validates that a value is a single character string. By default, rejects empty strings.


```{r, eval=FALSE}
validator <- v_character_scalar()
validator("hello")      # ✓ OK
validator("")           # ✗ Error: must not be empty
```

### Range Validators

#### `v_numeric_range(min = -Inf, max = Inf)`
Validates that a value is a single numeric value within a specified range (inclusive).

Parameters:
- `min`: Minimum allowed value (default: -Inf)
- `max`: Maximum allowed value (default: Inf)

```{r, eval=FALSE}
# Values between 0 and 1 (e.g., probability)
validator <- v_numeric_range(min = 0, max = 1)
validator(0.5)          # ✓ OK
validator(-0.1)         # ✗ Error: must be between 0 and 1

# Temperature range (Celsius)
validator <- v_numeric_range(min = -40, max = 50)
validator(-10)          # ✓ OK
```

### Vector Validators

#### `v_numeric_vector(min_len = 1, finite = TRUE)`
Validates that a value is a numeric vector with optional minimum length and finiteness requirements.

Parameters:
- `min_len`: Minimum required vector length (default: 1)
- `finite`: If TRUE (default), rejects Inf/-Inf/NaN values. If FALSE, they are allowed.

```{r, eval=FALSE}
# Any numeric vector
validator <- v_numeric_vector()
validator(1)            # ✓ OK
validator(c(1, 2, 3))   # ✓ OK

# Vector of at least 3 elements
validator <- v_numeric_vector(min_len = 3)
validator(c(1, 2, 3, 4)) # ✓ OK
validator(c(1, 2))      # ✗ Error: length must be >= 3

# Finite values only (no Inf, -Inf, NaN)
validator <- v_numeric_vector(finite = TRUE)
validator(c(1, 2, 3))   # ✓ OK
validator(c(1, Inf, 3)) # ✗ Error: must be finite
```

### Choice Validators

#### `v_enum(choices)`
Validates that a value is one of a predefined set of allowed choices.

Parameters:
- `choices`: A character vector of allowed values

```{r, eval=FALSE}
validator <- v_enum(c("train", "test", "validation"))
validator("train")      # ✓ OK
validator("predict")    # ✗ Error: must be one of: train, test, validation

# Backend choices
validator <- v_enum(c("cpu", "gpu", "tpu"))
validator("gpu")        # ✓ OK
```

### Structured Data Validators

#### `v_xypair(min_len = 1)`
Validates that a value is a list with paired `x` and `y` components. Useful for time-series data, phenology parameters, or any paired numeric sequences.

Parameters:
- `min_len`: Minimum required length for both x and y vectors (default: 1)

The validator checks that:
- The value is a list
- It has both `x` and `y` named elements
- Neither `x` nor `y` is NULL
- Both `x` and `y` are atomic vectors
- Neither contains NA values
- Both have the same length
- Both are at least `min_len` long

```{r, eval=FALSE}
validator <- v_xypair()
validator(list(x = c(1, 2, 3), y = c(10, 20, 30)))  # ✓ OK

validator <- v_xypair(min_len = 2)
validator(list(x = c(1), y = c(10)))  # ✗ Error: xypair length must be >= 2

validator(list(x = c(1, 2), y = c(10, 20, 30)))  # ✗ Error: different lengths
validator(list(x = c(1, 2)))  # ✗ Error: missing y
```

## Using Validators in Options Managers

Validators are specified when creating an options manager:

```{r, eval=FALSE}
library(optree)

# Create a machine learning config manager
ml_config <- create_options_manager(
  defaults = list(
    model_name = "my_model",
    learning_rate = 0.01,
    batch_size = 32,
    backend = "cpu",
    epochs = 10,
    coefficients = c(0.1, 0.2, 0.3)
  ),
  validators = list(
    "model_name" = v_character_scalar(),
    "learning_rate" = v_numeric_range(min = 0, max = 1),
    "batch_size" = v_numeric_scalar(),
    "backend" = v_enum(c("cpu", "gpu", "tpu")),
    "epochs" = v_numeric_range(min = 1, max = 1000),
    "coefficients" = v_numeric_vector(min_len = 2)
  )
)

# Valid updates work
ml_config$set(learning_rate = 0.05)  # ✓ OK
ml_config$set(backend = "gpu")       # ✓ OK

# Invalid updates fail
ml_config$set(learning_rate = 1.5)   # ✗ Error: must be between 0 and 1
ml_config$set(backend = "quantum")   # ✗ Error: must be one of: cpu, gpu, tpu
```

## Validators with Hierarchical Options

Validators can be applied to any level of your hierarchical options:

```{r, eval=FALSE}
crop_model <- create_options_manager(
  defaults = list(
    phenology = list(
      thermaltime = list(
        x = c(2, 30, 35),
        y = c(0, 28, 0)
      ),
      photoperiod = list(
        x = c(1, 2, 3),
        y = c(0, 1, 0)
      )
    ),
    soil = list(
      ph = 7,
      moisture = 0.5
    )
  ),
  validators = list(
    # Validate paired data structures
    "phenology.thermaltime" = v_xypair(min_len = 3),
    "phenology.photoperiod" = v_xypair(min_len = 3),
    
    # Validate scalar values
    "soil.ph" = v_numeric_range(min = 4, max = 9),
    "soil.moisture" = v_numeric_range(min = 0, max = 1)
  )
)
```

## Writing Custom Validators

For specialized validation logic, you can write custom validators. A validator is simply a function that:
1. Takes a single argument (the value being validated)
2. Silently returns if the value is valid
3. Calls `stop()` with an error message if the value is invalid

```{r, eval=FALSE}
# Custom validator: must be a positive multiple of 5
positive_multiple_of_5 <- function(x) {
  if (!is.numeric(x) || length(x) != 1) {
    stop("must be a single numeric value", call. = FALSE)
  }
  if (x <= 0) {
    stop("must be positive", call. = FALSE)
  }
  if (x %% 5 != 0) {
    stop("must be a multiple of 5", call. = FALSE)
  }
}

my_config <- create_options_manager(
  defaults = list(
    step_size = 5
  ),
  validators = list(
    "step_size" = positive_multiple_of_5
  )
)

my_config$set(step_size = 10)  # ✓ OK
my_config$set(step_size = 7)   # ✗ Error: must be a multiple of 5
```

## Combining Multiple Validators

You can create higher-level validators by combining simpler ones and custom logic:

```{r, eval=FALSE}
# Validator factory: numeric value that must be a multiple of a given number
v_multiple_of <- function(divisor) {
  function(x) {
    # First check it's a valid numeric scalar
    v_numeric_scalar()(x)
    
    # Then check it's a multiple
    if (x %% divisor != 0) {
      stop(sprintf("must be a multiple of %d", divisor), call. = FALSE)
    }
  }
}

validator <- v_multiple_of(10)
validator(50)   # ✓ OK
validator(45)   # ✗ Error: must be a multiple of 10
```

## Error Handling and Transactional Safety

When validation fails, optree automatically rolls back any changes. This ensures your options manager never enters an invalid state:

```{r, eval=FALSE}
config <- create_options_manager(
  defaults = list(
    x = 1,
    y = 2
  ),
  validators = list(
    "x" = v_numeric_range(min = 0, max = 10)
  )
)

# Valid update works
config$set(x = 5)
config$get("x")  # Returns 5

# Invalid update fails and rolls back
tryCatch(
  config$set(x = 15),  # Out of range
  error = function(e) cat("Error caught:", e$message, "\n")
)

config$get("x")  # Still returns 5 (rolled back)
```

## Summary

optree provides a comprehensive set of built-in validators for common validation patterns:

| Validator | Purpose |
|-----------|---------|
| `v_numeric_scalar()` | Single numeric value |
| `v_logical_scalar()` | Single logical value |
| `v_character_scalar()` | Single character string |
| `v_numeric_range()` | Bounded numeric value |
| `v_numeric_vector()` | Numeric vector with constraints |
| `v_enum()` | One of predefined choices |
| `v_xypair()` | Paired x/y data structures |

For more information, see the [function documentation](https://optree.bangyou.me/reference/) or create custom validators for your specific needs.