Multiclass classifier error during model training (random search)

[apokrif333]

Problem: Multiclass classifier error during model training (random search)
catboost version: '0.17.5'
Operating System: Linux 3.10.0-957.21.3.el7.x86_64
CPU:
GPU:

I get error - 'catboost/private/libs/lapack/linear_system.cpp:31: System of linear equations is not positive definite'.

train_set = np.array([
    [
        0.        , 0.        , 0.        , 0.        , 0.        ,
        0.        , 0.        , 0.        , 0.        , 0.        
    ],
    [
        0.00473934, 0.05      , 0.        , 0.        , 0.        ,
        0.        , 0.        , 0.        , 0.        , 0.        
    ],
    [
        0.04739336, 0.1       , 0.        , 0.        , 0.        ,
        0.03191489, 0.        , 0.        , 0.        , 0.        
    ],
    [
        0.        , 0.        , 0.        , 0.        , 0.00298507,
        0.09574468, 0.0195122 , 0.01492537, 0.00787402, 0.        
    ],
    [
        0.0521327 , 0.15      , 0.00480769, 0.07692308, 0.        ,
        0.        , 0.        , 0.        , 0.        , 0.        
    ]
])

y_train = np.array([1, 2, 4, 2, 1])

----------------------------------

from catboost import Pool, CatBoostClassifier, cv

cat_params = { 
    'iterations': 500,
    'random_state': 42,
    'loss_function': 'MultiClass',
    'eval_metric': 'TotalF1',
    'early_stopping_rounds': 30,
    'thread_count': -1
}
cat_model = CatBoostClassifier(**cat_params)

param_test ={
    'learning_rate': [0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.3, 0.5],
    'l2_leaf_reg': [1e-20, 1e-3, 1e-2, 1e-1, 1, 2, 5, 7, 10, 50, 100],
    'depth': [2, 3, 4, 5, 6, 7],
    'border_count':[5, 10, 20, 32, 50],
}
randomized_search_result = cat_model.randomized_search(
    param_test,
    X=np.array(train)[:5, :10],
    y=np.array(y_train)[:5],
    n_iter=500,
    cv=3,
    plot=True,
    refit=True,
    verbose=0
)

-----------------------------------

@annaveronika , I here from - #1022

@Evgueni-Petrov-aka-espetrov , thank you for your hint. If I'm removing '0' from l2_leaf_reg - it's working. But, if I set '1e-20', I still get error.
@Evgueni-Petrov-aka-espetrov, you second question. If I add cat feature:

['male', 'female', 'male', 'female', 'male']
train_pool = Pool(data=small_train, label=y_train[:5], cat_features=[5])

I still get error.

[annaveronika]

Thank you very much for the report! We have fixed it! The code for the fix is out on github and will be on pypi in the next release. Until than please increase the value of l2_leaf_reg to make it work.