Pool class using cat_features could not convert string to float error about catboost HOT 15 CLOSED

Categorical feature indices should be correct. If the 0 and 2 features are categorical and others numeric, then the value should be cat_index=[0,2]
In case of this error some categorical feature was not included to this list, so alrogithm was trying to parse it as numeric.

from catboost.

No, if you provide the date and say that it's a categorical feature, then it will be treated like a string with this value.

from catboost.

@shivamsaboo17, for now if you use python package you should convert the labels to categories or try catboost command line version with option --class-names. Possibility to use class-names in python version will be added soon.

Also automatic setting class_names will be added into both versions.

from catboost.

Does it convert date into numerical value ?

from catboost.

@abaspinar, could you provide your data?
*And check that all features have right indices that you give into cat_features (indices start from 0 (zero))

from catboost.

Thanks! @Donskov7
It was because of wrong indices.

from catboost.

I did'nt understand @abaspinar how did you correct the wrong indicies

from catboost.

I am getting same error but it's with the target labels which are in string format. Should I convert the labels to categories as string labels are not supported?

from catboost.

Same issue
What should i fix if:
I trained catboost and defined cat features as this:
clf.fit(train, labels, cat_dims)
And when i look at indexes of cat and float features in the trained model they are the same as in cat_dims variable before.
I got this error when i load test set and try to make class prediction:
_catboost.CatboostError: Bad value for num_feature[0,3]="1.281.501.0": Cannot convert 'b'1.281.501.0'' to float

My cat indexes in trained model:
[0, 1, 2, 9, 10, 11, 12, 13, 15, 16, 19, 21, 24, 28, 29, 31, 32, 34, 35, 36, 38, 39]
So, Bad value for num_feature[0,3] error - means, that my third feature was float (and it is true; it is not presented in the cat indexes list above), and it was OK while training, and now, when i try to predict on test sample it suddenly became unable to convert it to float, because third feature in test sample is now categorical (according to the value "1.281.501.0")? And how to know the name of this feature? The "third" feature does not make any sense to me. What should i do to load test samples and make a prediction successfully?

PS.: i open test sample in pandas with the same data types as train. And the number of features is the same

from catboost.

@goshulina Opened #633

from catboost.

for the sake of future visitors - this error is most likely thrown when you have a categorical feature which is not casted to string, prior to fitting

from catboost.

I'm having the same problem on a new dataset. In the coursera example all the column dtypes are int64 and it works fine. Nothing was converted. My new dataset is throwing the error with a float64 that I converted to string (object) and am having the same problem... @annaveronika what am I missing?

from catboost.

do we need to label encode the string containing categorical features which contain dtypes as object to numeric ? for data to be pooled ? because I am also getting same error

TypeError: Cannot convert 'b'W'' to float

from catboost.

I'm having the same problem on a new dataset. In the coursera example all the column dtypes are int64 and it works fine. Nothing was converted. My new dataset is throwing the error with a float64 that I converted to string (object) and am having the same problem... @annaveronika what am I missing?

It is not allowed to use floating point columns for categorical features. The best thing is to use Categorical type, you can also use integer or object types.
Here is the explanation why you cannot use float numbers for categorical features:
https://catboost.ai/docs/concepts/faq.html#specify-weights-or-baseline-for-eval-set

from catboost.

do we need to label encode the string containing categorical features which contain dtypes as object to numeric ? for data to be pooled ? because I am also getting same error

TypeError: Cannot convert 'b'W'' to float

This error means that you have not listed your categorical feature in cat_features parameter. By default all features are considered numeric. You must explicitly say that the feature is categorical.

from catboost.