microsoft

/

tapex-large-finetuned-wtq

like 51

Thanks for sharing your model. I am trying to load some data from csv. The rows have string values and float values for some columns. Idea is to query on these float values and get results.

from transformers import TapexTokenizer, BartForConditionalGeneration
import pandas as pd

tokenizer = TapexTokenizer.from_pretrained("microsoft/tapex-large-sql-execution")
model = BartForConditionalGeneration.from_pretrained("microsoft/tapex-large-sql-execution")

csv_file_path = "smallData.csv"
data = pd.read_csv(csv_file_path)
 #Convert all values to string to resolve error sequence item 4: expected str instance, float found
for key, values in data.items():
    data[key] = [str(value) for value in values]

table = pd.DataFrame.from_dict(data)

# tapex accepts uncased input since it is pre-trained on the uncased corpus
query = "Give me asins with protein_value below 111"
encoding = tokenizer(table=data, query=query, return_tensors="pt")

outputs = model.generate(**encoding,max_length=1024)

print(tokenizer.batch_decode(outputs, skip_special_tokens=True))

I convert the float values to string to resolve an error, but lose the ability to query on numeric values. Such as Select products where float_value_YY < 75.0 -> This does not understand 75.0 as its stored in string fashion.

In the examples, I see I am able to query over float values as well. What can I do to fix this?