Description
I'm submitting a
- bug report.
- improvement.
- [ x] feature request.
Current Behaviour:
SVC and SVR models skip the parameters during serialization. What is the correct way to deserialize a model in order to perform inference?
let lr_json = serde_json::to_string(&my_model);
produces the following json file:
{ "classes": [-1, 1], "instances": [ [4.9, 2.4, 3.3, 1.0], [4.4, 2.9, 1.4, 0.2], [7.0, 3.2, 4.7, 1.4], [5.5, 2.3, 4.0, 1.3], [6.9, 3.1, 4.9, 1.5], [5.4, 3.9, 1.7, 0.4], [5.7, 2.8, 4.5, 1.3], [6.3, 3.3, 4.7, 1.6] ], "w": [ 0.7330568313232404, -0.9886291197226762, 0.4983642540817094, 0.1369267334839056, 0.12914908421939392, -0.9760276163690054, 0.3306627164164916, 0.13649711656694088 ], "b": 0.18156339068191985, "phantomdata": null }
The "parameters" field is missing due a serde skip command in svc.rs:128:7.
A deserialized model cannot perform inference using SVC.predict() due to the missing parameters field.
let y_hat: &Vec<f64> = &model.predict(&x).unwrap();
results in an error:
thread 'main' panicked at 'called Option::unwrap() on a None value', /Users/.../.cargo/registry/src/index.crates.io-6f17d22bba15001f/smartcore-0.3.2/src/svm/svc.rs:349:26
Expected Behaviour:
If serde skip is removed and I am able to serialize the parameters, I would expect a deserialized model to perform inference.