Different results of code with local account and technical account

[dreyz64]

You really need to provide more details. If your question is about Python code, we need to see the code that causes the problem. (If that code more than about ten lines, I highly recommend you come up with example code which is minimized to reproduce your problem, hard-code as much as possible.) As for the most recent results you posted, I'm not especially familiar with MS SQL but you'll need to provide your queries if you want people to comment on the results.

hi Micseydel,

here is the code:

from datetime import datetime, timedelta
import pandas as pd
import numpy as np
import ast
from warnings import catch_warnings
from warnings import filterwarnings
from statsmodels.tsa.holtwinters import ExponentialSmoothing
import time
import sqlite3
import pyodbc as msql
from dateutil.relativedelta import relativedelta

def lit_eval(strg):
    if type(strg) == str:
        return ast.literal_eval(strg)
    else:
        return strg

def pd_dict2list(row):
    return [value for key, value in row.items()][0:-1]

def frcst_cot(y, model, setup, fcst_hrz):
    result = None
    try:
        # never show warnings when grid searching, too noisy
        with catch_warnings():
            filterwarnings("ignore")

            t, d, s, p, b, r = setup
            # fit model
            model = ExponentialSmoothing(y, seasonal_periods=p, trend=t, seasonal=s, damped=d).fit(use_boxcox=b,
                                                                                                   remove_bias=r)
            yhat_v = model.predict(len(y) - fcst_hrz)
            yhat = model.forecast(fcst_hrz)
            result = pd.concat([yhat_v, yhat], axis=0)
    except:
        error = None
        result = None

    # if no forecasting was done applying generic model    
    if result is None:
        print('result is none, trying generic model...')
        try:
            # never show warnings when grid searching, too noisy
            with catch_warnings():
                filterwarnings("ignore")

                t, d, s, p, b, r = setup
                # fit model
                model = ExponentialSmoothing(y).fit()
                yhat_v = model.predict(len(y) - fcst_hrz)
                yhat = model.forecast(fcst_hrz)
                result = pd.concat([yhat_v, yhat], axis=0)
        except:
            error = None
            result = None

    # if results are still None using Naive model prediction 
    if result is None:
        print('result is still none, applying Naive model...')
        try:
            # never show warnings when grid searching, too noisy
            with catch_warnings():
                filterwarnings("ignore")

                forecst = y.copy()
                idx = pd.date_range(start=y.tail(1).index.min(), periods=fcst_hrz, freq='MS')
                forecst = forecst.reindex(idx, fill_value=0)
                forecst['Volume'] = y.tail(1).values[0][0]
                forecst = forecst.squeeze()
        except:
            error = None
    return result

# TODO: Add path to pickle file BEST_MODELS FILE
best_models = pd.read_pickle(r'C:\Users\XXX\Desktop\..\best_models.pkl')

starttime = time.time()

for itm in range(len(best_models)):
    row = best_models[itm:itm + 1]

    fcst_horizon = 12

    cols = (row.T != '(All)')
    cols.columns = ['chek']
    cols = cols[cols.chek == True].T

    needed_cols = list(set(cols.columns) - {'DATE', 'MODEL', 'SETUP', 'VERTEX'})

    qry_str = '\tand '.join([(c + ' == "' + row[c].values[0] + '" ') for c in needed_cols])

    # TODO: ADD PICKLE PATH COT_VERTEX
    CoT_vertex = pd.read_pickle(r'C:\Users\XXX\Desktop\..CoT_vertex.pkl')
    CoT_vertex = CoT_vertex.copy()

    CoT_vertex = CoT_vertex.pivot_table(index='Date', values=['Volume'], aggfunc=np.sum).reset_index()

    if CoT_vertex.Date.min() != CoT_vertex.Date.min():
        print('no COT data for:' + qry_str)
    else:
        if datetime.strptime(CoT_vertex.Date.max(), '%Y-%m-%d').date() >= (
                datetime.now().replace(day=1) + relativedelta(months=-2)).date():  # Added
            idx = pd.date_range(start=CoT_vertex.Date.min(), end=CoT_vertex.Date.max(), freq='MS')
            CoT_vertex.set_index('Date', inplace=True)
            CoT_vertex = CoT_vertex.reindex(idx, fill_value=0)
            CoT_vertex = CoT_vertex.squeeze()

            frcst = frcst_cot(CoT_vertex, row['MODEL'].values[0], row['SETUP'].values[0], fcst_horizon)

            print(frcst)

        else:
            print('Last COT older than 3m')

I tried to hardcode as much as I can but I am very new at Python so still struggling.
I am now looking to see how to send you the 2 pickle files so that you can run the code...