Exception: TypeError: a float is required

[lrq3000]

catboost version: 0.16.4
Operating System: Windows 10 64-bits
CPU: Intel Core i7 8th Gen
Python 2.7
Problem:

When I try to fit a simple CatBoostClassifier on my DataFrame containing categorical variables, I get the very cryptic error below.

Here is what I did to try to troubleshoot down the issue:

• tested a clean notebook on a tutorial (https://github.com/catboost/tutorials/blob/master/python_tutorial.ipynb) --> OK
• fit CatBoostClassifier only on my non-categorical variables --> OK
• try to encode with a LabelEncoder my categorical variables and then fit CatBoostClassifier --> OK
• try to use any single categorical variable without encoding like in the titanic tutorial above --> KO

I can't understand what's happening here and have no idea on how to debug this further. If anyone has any idea, please let me know, I can provide more info.

Here is the code failing (it's not a minimal example, I'm not sure how I can reproduce this):

import catboost
model = catboost.CatBoostClassifier(custom_loss=['Accuracy'], random_seed=0, eval_metric='AUC', logging_level='Silent')
model.fit(X_train.loc[:, ['categoricalvariable']], y_train_enc,
    cat_features=[0],
    plot=True
)

And the error traceback:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-100-7e62575dc820> in <module>()
----> 5     plot=True
      6 )

C:\Users\AI\Anaconda2\lib\site-packages\catboost\core.pyc in fit(self, X, y, cat_features, sample_weight, baseline, use_best_model, eval_set, verbose, logging_level, plot, column_description, verbose_eval, metric_period, silent, early_stopping_rounds, save_snapshot, snapshot_file, snapshot_interval, init_model)
   3455         self._fit(X, y, cat_features, None, sample_weight, None, None, None, None, baseline, use_best_model,
   3456                   eval_set, verbose, logging_level, plot, column_description, verbose_eval, metric_period,
-> 3457                   silent, early_stopping_rounds, save_snapshot, snapshot_file, snapshot_interval, init_model)
   3458         return self
   3459 

C:\Users\AI\Anaconda2\lib\site-packages\catboost\core.pyc in _fit(self, X, y, cat_features, pairs, sample_weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, use_best_model, eval_set, verbose, logging_level, plot, column_description, verbose_eval, metric_period, silent, early_stopping_rounds, save_snapshot, snapshot_file, snapshot_interval, init_model)
   1386             use_best_model, eval_set, verbose, logging_level, plot,
   1387             column_description, verbose_eval, metric_period, silent, early_stopping_rounds,
-> 1388             save_snapshot, snapshot_file, snapshot_interval, init_model
   1389         )
   1390         params = train_params["params"]

C:\Users\AI\Anaconda2\lib\site-packages\catboost\core.pyc in _prepare_train_params(self, X, y, cat_features, pairs, sample_weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, use_best_model, eval_set, verbose, logging_level, plot, column_description, verbose_eval, metric_period, silent, early_stopping_rounds, save_snapshot, snapshot_file, snapshot_interval, init_model)
   1281             del params['cat_features']
   1282 
-> 1283         train_pool = _build_train_pool(X, y, cat_features, pairs, sample_weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, column_description)
   1284         if train_pool.is_empty_:
   1285             raise CatBoostError("X is empty.")

C:\Users\AI\Anaconda2\lib\site-packages\catboost\core.pyc in _build_train_pool(X, y, cat_features, pairs, sample_weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, column_description)
    697             raise CatBoostError("y has not initialized in fit(): X is not catboost.Pool object, y must be not None in fit().")
    698         train_pool = Pool(X, y, cat_features=cat_features, pairs=pairs, weight=sample_weight, group_id=group_id,
--> 699                           group_weight=group_weight, subgroup_id=subgroup_id, pairs_weight=pairs_weight, baseline=baseline)
    700     return train_pool
    701 

C:\Users\AI\Anaconda2\lib\site-packages\catboost\core.pyc in __init__(self, data, label, cat_features, column_description, pairs, delimiter, has_header, weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, feature_names, thread_count)
    326                         )
    327 
--> 328                 self._init(data, label, cat_features, pairs, weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, feature_names)
    329         super(Pool, self).__init__()
    330 

C:\Users\AI\Anaconda2\lib\site-packages\catboost\core.pyc in _init(self, data, label, cat_features, pairs, weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, feature_names)
    678             baseline = np.reshape(baseline, (samples_count, -1))
    679             self._check_baseline_shape(baseline, samples_count)
--> 680         self._init_pool(data, label, cat_features, pairs, weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, feature_names)
    681 
    682 

_catboost.pyx in _catboost._PoolBase._init_pool()

_catboost.pyx in _catboost._PoolBase._init_pool()

_catboost.pyx in _catboost._PoolBase._init_features_order_layout_pool()

_catboost.pyx in _catboost._set_features_order_data_pd_data_frame()

_catboost.pyx in _catboost.get_cat_factor_bytes_representation()

_catboost.pyx in _catboost.get_id_object_bytes_string_representation()

TypeError: a float is required

[lrq3000]

After further debugging, and updating my scripts and notebook to Python 3.7, I can confirm this bug happens only with Python 2. So this is a Python version specific bug. On Python 3.7, CatBoost can handle categorical features without any prior encoding, but for Python 2.7 there should at least be a warning that categorical features should be encoded, in case this bug cannot be fixed.

[lrq3000]

What is surprising is that with the titanic dataset provided in catboost, there is no issue even without any prior encoding in Python 2.7. Ah also I found that my columns here had some None values. In Python 3.7 the error was more explicit and thus I could track down and convert the None values to strings, maybe this is the culprit of this error? String or object columns having None values?

[annaveronika]

Can you please try with the latest version? We've made changes to this part of the code and the bug should be fixed.

[annaveronika]

I hope it has helped! Let us know if you still have problems, I'll reopen the issue!

[AnnaMosen]

I keep getting the same issue. I made sure I am using the Python 3.7 and also that the Catboost package is updated and I am still getting the error: CatBoostError: Bad value for num_feature[non_default_doc_idx=0,feature_idx=11]="['5']": Cannot convert 'b"['5']"' to float
is it make any difference that the value is a list? the data type is a string but the string is a list of numbers (I don't know if it makes any difference)
type(X['myfeature'][0])
str

I also made sure I'm passing the right index for the categorical feature:
categorical_features_indices = [11]

X.info()
Data columns (total 76 columns):
'#' Column Non-Null Count Dtype
...
11 myfeature 91607 non-null object
...