Struct MarkovModel 

pub struct MarkovModel { /* private fields */ }

N-gram Markov model for shell command prediction

Implementations

impl MarkovModel

pub fn new(n: usize) -> Self

Create a new model with given n-gram size

pub fn train(&mut self, commands: &[String])

Train on a list of commands

pub fn train_incremental(&mut self, commands: &[String])

Incrementally train on new commands (appends to existing model)

pub fn last_trained_position(&self) -> usize

Get the last trained position in history

pub fn total_commands(&self) -> usize

Get total commands trained on

pub fn suggest(&self, prefix: &str, count: usize) -> Vec<(String, f32)>

Suggest completions for a prefix

Optimized for minimal allocations (Issue #93):

• Pre-allocated vectors with capacity
• HashSet for O(1) duplicate detection
• Reused string buffers where possible

pub fn save(&self, path: &Path) -> Result<()>

Save model to .apr file

Uses ModelType::NgramLm (0x10) for proper classification (QA report fix).

pub fn load(path: &Path) -> Result<Self>

Load model from .apr file using memory-mapped I/O

Uses mmap for zero-copy loading, reducing syscalls from ~970 to <50 (see bundle-mmap-spec.md Section 8).

Supports both NgramLm (new) and Custom (legacy) model types for backward compatibility.

pub fn save_encrypted(&self, path: &Path, password: &str) -> Result<()>

Save model with AES-256-GCM encryption (spec §4.1.2)

Uses Argon2id for key derivation from password. The model can only be loaded with the correct password.

pub fn load_encrypted(path: &Path, password: &str) -> Result<Self>

Load encrypted model from .apr file (spec §4.1.2)

Requires the same password used during encryption. Returns an error if the password is incorrect.

pub fn is_encrypted(path: &Path) -> Result<bool>

Check if a model file is encrypted

pub fn save_compressed(&self, path: &Path) -> Result<()>

Save model with zstd compression (Tier 2)

Achieves ~14x size reduction with minimal decompression overhead (~10-20ms). Actually faster in practice due to reduced I/O.

pub fn save_compressed_encrypted( &self, path: &Path, password: &str, ) -> Result<()>

Save model with both compression and encryption (Tier 2+3)

Best of both worlds: small size and protection.

pub fn is_compressed(path: &Path) -> Result<bool>

Check if a model file is compressed

Returns true if payload_size < uncompressed_size (compression was applied)

impl MarkovModel

pub fn ngram_count(&self) -> usize

Number of unique n-grams

pub fn vocab_size(&self) -> usize

Vocabulary size

pub fn ngram_size(&self) -> usize

N-gram size

pub fn size_bytes(&self) -> usize

Approximate model size in bytes

pub fn top_commands(&self, count: usize) -> Vec<(String, u32)>

Top commands by frequency

Optimized to reduce allocations:

• Pre-allocated result vector
• Uses sort_unstable for better cache locality

pub fn validate( commands: &[String], ngram_size: usize, train_ratio: f32, ) -> ValidationResult

Validate model using holdout evaluation with aprender’s ranking metrics.

Uses aprender::metrics::ranking for Hit@K and MRR with prefix matching (appropriate for command completion where partial matches count).

Trait Implementations

impl<'de> Deserialize<'de> for MarkovModel

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for MarkovModel

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.