wandb -tracking hyperparameter tuning

svm.py

# preprocessing


for gamma in tqdm(['scale', 'auto'], 'gamma', leave=False):
    for kernel in tqdm(['rbf', 'poly'], 'kernel', leave=False):
        for c in tqdm([0.1, 1, 10, 100, 1000], 'c', leave=False):
            wandb.init(project="ails-challenge-1", entity="keckelt", tags=["svm"])

            params = {
                "gamma": gamma,
                "kernel": kernel,
                "C": c,
                "probability":True,
                "cache_size":2048,  # 2GB
                "random_state":  korok_seed
                }
            models, y_hats_proba, y_hats_class = train_svm(X_train, y_train, X_test, params)

            auc_per_task = calc_masked_AUC_per_task(y_hats_proba, y_test)
            auc = np.mean(auc_per_task)

            config = {
                "AUC": auc,
                "TASK_AUC": auc_per_task,
                **params
            }

            if (auc > best_auc):
                best_auc = auc
                tqdm.write(str(config))

            wandb.log(config)