cvhist.py

import pandas as pd
import numpy as np
from scipy.stats.mstats import gmean, hmean

def cv_store(desc, kaggle, tcv, kcv, valid, filename='cvhist.csv'):
    """
    Takes the latest CV and LB results and stores them into a csv file.
    
    Args:
     desc: Submission description
     kaggle: kaggle public LB score
     tcv: time-series CV
     kcv: K-Fold CV
     valid: validation score
     
    Returns:
     latest historic dataframe
    """
    tcv_col = [f'tcv{i}' for i in range(0, len(tcv))]
    kcv_col = [f'kcv{i}' for i in range(0, len(kcv))]
    columns = ['ts', 'description', 'kaggle', 'valid', 'tcv_mean', 'tcv_std', 'gmean', *tcv_col, *kcv_col]
    new_row =  [
        pd.datetime.utcnow(),
        desc, kaggle, valid,
        np.mean(tcv), np.std(tcv),
        gmean([np.mean(tcv_), np.mean(kcv_)]),
        *tcv, *kcv,
    ]
    try:
        df = pd.read_csv(filename)
        df = df.append(pd.DataFrame([new_row], columns=columns))
        df.reset_index(drop=True, inplace=True)
    except:
        df = pd.DataFrame(
            [new_row],
            columns=columns,
        )
    df['ts'] = pd.to_datetime(df['ts'])
    df.sort_values('ts', ascending=False, inplace=True)
    df = df.drop_duplicates(subset=['description', 'kaggle'], keep='last')
    df.to_csv(filename, index=False)
    return df

def cv_hist(filename='cvhist.csv'):
    try:
        df = pd.read_csv(filename)
        df['ts'] = pd.to_datetime(df['ts'])
        df.sort_values('ts', ascending=False, inplace=True)
    except:
        print('History not available.')
    return df