app/tapas.py

from transformers import AutoTokenizer, AutoModelForTableQuestionAnswering
import pandas as pd
import re

p = re.compile('\d+(\.\d+)?')

def load_model_and_tokenizer():
  """
  Load
  """
  tokenizer = AutoTokenizer.from_pretrained("Meena/table-question-answering-tapas")
  model = AutoModelForTableQuestionAnswering.from_pretrained("Meena/table-question-answering-tapas")

  # Return tokenizer and model
  return tokenizer, model


def prepare_inputs(table, queries, tokenizer):
  """
  Convert dictionary into data frame and tokenize inputs given queries.
  """
  table = table.astype('str').head(100)
  inputs = tokenizer(table=table, queries=queries, padding='max_length', return_tensors="pt")
  return table, inputs


def generate_predictions(inputs, model, tokenizer):
  """
  Generate predictions for some tokenized input.
  """
  # Generate model results
  outputs = model(**inputs)

  # Convert logit outputs into predictions for table cells and aggregation operators
  predicted_table_cell_coords, predicted_aggregation_operators = tokenizer.convert_logits_to_predictions(
        inputs,
        outputs.logits.detach(),
        outputs.logits_aggregation.detach()
  )

  # Return values
  return predicted_table_cell_coords, predicted_aggregation_operators

def postprocess_predictions(predicted_aggregation_operators, predicted_table_cell_coords, table):
  """
  Compute the predicted operation and nicely structure the answers.
  """
  # Process predicted aggregation operators
  aggregation_operators = {0: "NONE", 1: "SUM", 2: "AVERAGE", 3:"COUNT"}
  aggregation_predictions_string = [aggregation_operators[x] for x in predicted_aggregation_operators]
  # Process predicted table cell coordinates
  answers = []
  for agg, coordinates in zip(predicted_aggregation_operators, predicted_table_cell_coords):
      if len(coordinates) == 1:
        # 1 cell
        answers.append(table.iat[coordinates[0]])
      else:
        # > 1 cell
        cell_values = []
        for coordinate in coordinates:
          cell_values.append(table.iat[coordinate])
        answers.append(", ".join(cell_values))
         
  # Return values
  return aggregation_predictions_string, answers


def show_answers(queries, answers, aggregation_predictions_string):
  """
    Visualize the postprocessed answers.
  """
  agg = {"NONE": lambda x: x, "SUM" : lambda x: sum(x), "AVERAGE": lambda x: (sum(x) / len(x)), "COUNT": lambda x: len(x)}
  results = []
  for query, answer, predicted_agg in zip(queries, answers, aggregation_predictions_string):
      print(query)
      if predicted_agg == "NONE":
          print("Predicted answer: " + answer)
      else:
          if all([not p.match(val) == None for val in answer.split(', ')]):
              # print("Predicted answer: " + predicted_agg + "(" + answer + ") = " + str(agg[predicted_agg](list(map(float, answer.split(','))))))
              result = str(agg[predicted_agg](list(map(float, answer.split(',')))))
          elif predicted_agg == "COUNT":
              # print("Predicted answer: " + predicted_agg + "(" + answer + ") = " + str(agg[predicted_agg](answer.split(','))))
              result = str(agg[predicted_agg](answer.split(',')))
          else:
              result = predicted_agg + " > " + answer
          results.append(result)
  return results

def execute_query(query, table):
  """
	  Invoke the TAPAS model.
	"""
  queries = [query]
  tokenizer, model = load_model_and_tokenizer()
  table, inputs = prepare_inputs(table, queries, tokenizer)
  predicted_table_cell_coords, predicted_aggregation_operators = generate_predictions(inputs, model, tokenizer)
  aggregation_predictions_string, answers = postprocess_predictions(predicted_aggregation_operators, predicted_table_cell_coords, table)
  return show_answers(queries, answers, aggregation_predictions_string)