Put VRAVEN inside your workflow.

Use the API when you need repeatable analyses, notebooks, dataset loops or your own report pipeline.

Explain one prediction

import vraven

result = vraven.explain(
    model=model,
    input_tensor=x,
    target=target_class,
    mode="causal",
)

result.report("explanation.html")

Use fast, gradient or causal mode according to the evidence required and the compute available.

Reuse one explainer

from vraven import Explainer

explainer = Explainer(model, mode="gradient")
for sample in dataset:
    trace = explainer.explain(sample)
    trace.export("json", output_dir="runs")

Keep model configuration together

vraven bundle-init ./my_model
vraven explain --model-dir ./my_model --report html

A model bundle can carry weights, architecture loader, example input and VRAVEN defaults. See the model-folder walkthrough.

Choose one output

# One report per invocation
vraven explain --model-dir ./my_model --report html
vraven explain --model-dir ./my_model --report pdf
vraven explain --model-dir ./my_model --report json