Support for custom column names in harness for csv

[Prikshit7766]

Checklist:

• I've added Google style docstrings to my code.
• I've used pydantic for typing when/where necessary.
• I have linted my code
• I have added tests to cover my changes.

allowing users to specify custom column names for various tasks.

Notebook

text-classification

• with custom column

harness = Harness(task="text-classification",
                  model={"model":"lvwerra/distilbert-imdb", "hub":"huggingface"},
                  data={"data_source": "data_file.csv",
                        "feature_column": "text",    
                        "target_column": 'label',     
                       })

• Directly providing the data path

harness = Harness(
                  task="text-classification", 
                 model={"model":"lvwerra/distilbert-imdb", "hub":"huggingface"},
                data={"data_source": "data_file.csv"}

question-answering

• with custom column

harness = Harness(task="question-answering", 
                  model={"model":"gpt-3.5-turbo","hub":"openai"},
                  data={"data_source":"data_file.csv",
                  "feature_column":{"passage": "context", "question": "question"},
                  "target_column":'answer_start',
                  })

• Directly providing the data path

harness = Harness(task="question-answering", 
                 model={"model":"gpt-3.5-turbo","hub":"openai"},
                  data={"data_source":"data_file.csv"},

                  )

summarization

• with custom column

harness = Harness(task="summarization",
                  model={"model":"gpt-3.5-turbo","hub":"openai"},
                  data={"data_source":"data_file.csv",
                  "feature_column":"headlines",
                  "target_column":'text',
                  })

• Directly providing the data path

harness = Harness(task="summarization",
                   model={"model":"gpt-3.5-turbo","hub":"openai"},
                  data={"data_source":"data_file.csv"}
                  )

ner

• with custom column

harness = Harness( 
                  task="ner",
                  model={"model":"dslim/bert-base-NER", "hub":"huggingface"},
                  data={"data_source": "data_file.csv",
                        "feature_column": "tokens",    
                        "target_column": 'ner_tags',     
                       })

• Directly providing the data path

harness = Harness(
        task="ner",
        model={"model":"dslim/bert-base-NER", "hub":"huggingface"},
        data={"data_source":"data_file.csv"},
       
)

[JulesBelveze]

1. Please merge the latest changes from release/1.2.0
2. I am not getting why we are not extending the already existing CSVDataset we have?
3. Let's put this on hold for a moment as we are still debating with Arshaan on the parameters the Harness class should take. We'll think about it on Monday and get back to you but for now we're thinking of something like this:

Harness(
    task="ner",
    model={
        [{"model": "bert-base-cased", "hub": "huggingface"},
        {"model": "path/to/local/model", "hub": "johnsnowlabs"}]
    },
    data={
        "data_source" : "mydataset"
        "subset": "sst2",
        "feature_column": "sentence",
        "target_column": "label",
        "split": "train"
    }
)

[JulesBelveze]

How is that different from the CSVDataset we already have?

[Prikshit7766]

I think CSVDataset is dynamically called based on the .csv extenstion that we have in the file path
you can check the load method of DataFactory
for custom column names we are using dictionary format in data so it will not match .

What do you think @JulesBelveze?

[JulesBelveze]

I would rather have only one class that handles CSV files @Prikshit7766

[Prikshit7766]

we can try that

[JulesBelveze]

Hmmm not sure where we going here and we need to have a standard way to load data across all tasks and hubs.
I think that by looking at the current codebase each XXXDataset object should have a load_data method to which we pass feature_column, target_column.

[JulesBelveze]

Don't get what is this _custom_label attribute

[Prikshit7766]

_custom_label conatin

{"name": r"data\imdb.csv",
                        "feature_column": "text",    
                        "target_column": 'label',     
                       }```
            

[JulesBelveze]

Well then file_path is not a string anymore and doesn't refer to the file location

[Prikshit7766]

file_path contains only the path i think ,
also based on the path extension we are calling the load_data method of respective class

[JulesBelveze]

I am sorry guys but the approach is, to me, not user friendly and having this file_path parameter actually being a dictionary is to me a bit weird..
I'll write down and share with you what I have in mind and we can discuss it.