`Scorer __init__` silently fails with empty dataframe due to special encoding in csv file

[matsuobasho]

I have the output from the inference step of running run_gpt_v1.5_gpt4_1106_cot.py saved in the appropriate directory.
When I get to the scoring step, fread returns an empty list for the data object:

class Scorer:
    def __init__(self, files, data_list_of_dicts=None):
        if not len(files):
            print('No files for evaluation')
            import sys
            sys.exit()

        from efficiency.log import fread
        data_list = []
        for file in sorted(files):
            data = fread(file)
            data_list += data
            print(file, len(data))

The fread intended behavior is to return an empty list when read_csv errors out. This is not good practice - very difficult to diagnose why the user gets an empty dataframe and eventual error here, especially since fread is programmed in another library:

    def truth_pred_scorer(self, df):
        df.drop(['prompt', 'question_id'], axis=1, inplace=True)

        df = self.apply_score_func(df)

The error occurs because the specified columns for dropping aren't in the dataframe, since it is empty.

I'm able to import my file by specifying a special encoding at line 258 of efficiency.logpy`:

data = pd.read_csv(path, encoding = 'cp1252').to_dict(orient="records")

But the fact that I have to modify another library highlights why this isn't ideal.