Meena's Tapas Table Understanding (Base)

Description

This is a Zero-shot Table Understanding Model which allows you to carry out Question Answering on Spark Dataframes. If you have a file stored in any table format, as csv, load it before using Spark.

Size of this model: Base Has aggregation operations?: True

Predicted Entities

Download Copy S3 URI

How to use

json_data = """
{
  "header": ["name", "money", "age"],
  "rows": [
    ["Donald Trump", "$100,000,000", "75"],
    ["Elon Musk", "$20,000,000,000,000", "55"]
  ]
}
"""

queries = [
    "Who earns less than 200,000,000?",
    "Who earns 100,000,000?", 
    "How much money has Donald Trump?",
    "How old are they?",
]

data = spark.createDataFrame([
        [json_data, " ".join(queries)]
    ]).toDF("table_json", "questions")
    
document_assembler = MultiDocumentAssembler() \
    .setInputCols("table_json", "questions") \
    .setOutputCols("document_table", "document_questions")

sentence_detector = SentenceDetector() \
    .setInputCols(["document_questions"]) \
    .setOutputCol("questions")

table_assembler = TableAssembler()\
    .setInputCols(["document_table"])\
    .setOutputCol("table")

tapas = TapasForQuestionAnswering\
    .pretrained("table_qa_table_question_answering_tapas","en")\
    .setInputCols(["questions", "table"])\
    .setOutputCol("answers")

pipeline = nlp.Pipeline(stages=[
    document_assembler,
    sentence_detector,
    table_assembler,
    tapas
])

model = pipeline.fit(data)
model\
    .transform(data)\
    .selectExpr("explode(answers) AS answer")\
    .select("answer")\
    .show(truncate=False)

import nlu
nlu.load("en.answer_question.tapas").predict("""
{
  "header": ["name", "money", "age"],
  "rows": [
    ["Donald Trump", "$100,000,000", "75"],
    ["Elon Musk", "$20,000,000,000,000", "55"]
  ]
}
""")

Results

+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
|answer                                                                                                                                                                |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
|{chunk, 0, 12, Donald Trump, {question -> Who earns less than 200,000,000?, aggregation -> NONE, cell_positions -> [0, 0], cell_scores -> 0.9999999}, []}             |
|{chunk, 0, 12, Donald Trump, {question -> Who earns 100,000,000?, aggregation -> NONE, cell_positions -> [0, 0], cell_scores -> 0.9999999}, []}                       |
|{chunk, 0, 12, $100,000,000, {question -> How much money has Donald Trump?, aggregation -> NONE, cell_positions -> [1, 0], cell_scores -> 0.9999998}, []}             |
|{chunk, 0, 6, AVERAGE > 75, 55, {question -> How old are they?, aggregation -> AVERAGE, cell_positions -> [2, 0], [2, 1], cell_scores -> 0.99999976, 0.9999995}, []}  |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+

Model Information

Model Name: table_qa_table_question_answering_tapas
Compatibility: Spark NLP 4.2.0+
License: Open Source
Edition: Official
Language: en
Size: 413.9 MB
Case sensitive: false

References

https://huggingface.co/models?pipeline_tag=table-question-answering