Alternative for TextList.from_df

[Anish_sri (anish srivathsav)]

Hello, I am looking for a solution for a multi label text classification where there are articles suggetsing there is a way to perform in fastai v1. can Anyone help me out how to build a datablock or alternative for the following code snippet.

[Archanjo (Marcelo Archanjo)]

Hello @Anish_sri,

I had the same problem with a language model, the solution was to use TextDataLoaders.from_df

The documentation is here:

Text data | fastai

For you case you can try this:

data_clas = TextDataLoaders.from_df(train, path=path, text_col=‘comment_text’,label_col=label_cols, text_vocab=data.vocab, bs=32)

[msivanes (Manikandan Sivanesan)]

AttributeError when running learn.loss_func

Hi, I have a private multilabel dataset for text classification. I ran into AttributeError at running learn.loss_func. Here my dataset looks like: dls_class_block = DataBlock( blocks=(TextBlock.from_df('proc_name', res_col_name='proc_name', vocab=dls_lm.vocab), MultiCategorize), get_x=ColReader('proc_name'), get_y=ColReader('cur_job_func', label_delim='|'), splitter=ColSplitter(col='is_valid') ) dls_class = dls_class_block.dataloaders(data, bs=5) dls_class.one_bat…

Note : Example provided should use MultiCategoryBlock instead of MultiCategorize

[sahil1234 (Sahil)]

This is extremely helpful!
Thank you so much
Would you suggest some rules to follow when converting from old notations (having TextList and DataBunches) to the new ones (DataBlocks and DataLoaders)