Struct GradientFlowAnimator 

pub struct GradientFlowAnimator { /* private fields */ }

Collects per-step gradient data and provides export / analysis utilities.

Implementations

impl GradientFlowAnimator

pub fn new(max_frames: usize) -> Self

Create a new animator.

• max_frames — rolling window size (0 means unlimited).

pub fn with_vanishing_threshold(self, threshold: f64) -> Self

Set a custom vanishing-gradient threshold (default: 1e-7).

pub fn with_exploding_threshold(self, threshold: f64) -> Self

Set a custom exploding-gradient threshold (default: 1e3).

pub fn record_step(&mut self, step: u64, gradients: &HashMap<String, Vec<f64>>)

Record gradient tensors for one training step.

gradients maps a layer name to the flat gradient vector for that layer.

pub fn frames(&self) -> &[GradientFrame]

All retained frames (possibly a rolling subset).

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

Export all frames to a JSON file.

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

Export a CSV timeline: step,layer,mean_abs_grad,max_abs_grad,grad_norm.

pub fn to_ascii_animation(&self) -> String

Render an ASCII “heatmap” animation string suitable for terminal display.

Each row represents one layer; each column one recorded step. Intensity is encoded with the characters ' ', '░', '▒', '▓', '█'.

pub fn summary_report(&self) -> GradientFlowSummary

Generate a summary report for the recorded gradient history.