2. Time series analysis
[1]:
import numpy as np
import pandas as pd
import scipy.stats as stats
from sklearn import metrics
from sklearn.model_selection import train_test_split
import os
import scipy.stats as stats
import matplotlib.pylab as plt
from collections import Counter
from datetime import datetime
import statsmodels as sm
import statsmodels.api as sma
from tqdm import tqdm
import catboost as catb
%matplotlib inline
[2]:
# Download dataset at https://www.kaggle.com/rohitsahoo/sales-forecasting
df = pd.read_csv('supersales_train.csv')
[3]:
df.head()
[3]:
| Row ID | Order ID | Order Date | Ship Date | Ship Mode | Customer ID | Customer Name | Segment | Country | City | State | Postal Code | Region | Product ID | Category | Sub-Category | Product Name | Sales | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 1 | CA-2017-152156 | 08/11/2017 | 11/11/2017 | Second Class | CG-12520 | Claire Gute | Consumer | United States | Henderson | Kentucky | 42420.0 | South | FUR-BO-10001798 | Furniture | Bookcases | Bush Somerset Collection Bookcase | 261.9600 |
| 1 | 2 | CA-2017-152156 | 08/11/2017 | 11/11/2017 | Second Class | CG-12520 | Claire Gute | Consumer | United States | Henderson | Kentucky | 42420.0 | South | FUR-CH-10000454 | Furniture | Chairs | Hon Deluxe Fabric Upholstered Stacking Chairs,... | 731.9400 |
| 2 | 3 | CA-2017-138688 | 12/06/2017 | 16/06/2017 | Second Class | DV-13045 | Darrin Van Huff | Corporate | United States | Los Angeles | California | 90036.0 | West | OFF-LA-10000240 | Office Supplies | Labels | Self-Adhesive Address Labels for Typewriters b... | 14.6200 |
| 3 | 4 | US-2016-108966 | 11/10/2016 | 18/10/2016 | Standard Class | SO-20335 | Sean O'Donnell | Consumer | United States | Fort Lauderdale | Florida | 33311.0 | South | FUR-TA-10000577 | Furniture | Tables | Bretford CR4500 Series Slim Rectangular Table | 957.5775 |
| 4 | 5 | US-2016-108966 | 11/10/2016 | 18/10/2016 | Standard Class | SO-20335 | Sean O'Donnell | Consumer | United States | Fort Lauderdale | Florida | 33311.0 | South | OFF-ST-10000760 | Office Supplies | Storage | Eldon Fold 'N Roll Cart System | 22.3680 |
[4]:
df.iloc[0]
[4]:
Row ID 1
Order ID CA-2017-152156
Order Date 08/11/2017
Ship Date 11/11/2017
Ship Mode Second Class
Customer ID CG-12520
Customer Name Claire Gute
Segment Consumer
Country United States
City Henderson
State Kentucky
Postal Code 42420.0
Region South
Product ID FUR-BO-10001798
Category Furniture
Sub-Category Bookcases
Product Name Bush Somerset Collection Bookcase
Sales 261.96
Name: 0, dtype: object
[10]:
# Generate daily sales and some columns specifing the amount some customers bought on each day
# This ammount could be used as features/exogenous data on time series
list_of_customers = Counter(df['Customer ID'])
list_of_customers = [k for k,v in list_of_customers.items() if v > 30]
def func(val):
#import pdb; pdb.set_trace()
res = [val.iloc[:,0].sum()]
for customer in list_of_customers:
res.append(sum(val['Customer ID']==customer))
return pd.DataFrame([res])
dfg = df.groupby("Order Date")[['Sales', 'Customer ID']].apply(func).reset_index()
del dfg['level_1']
dfg.rename(columns={'Order Date': "date", 0:'sales'}, inplace=True)
dfg
[10]:
| date | sales | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 01/01/2018 | 1481.8280 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 1 | 01/02/2015 | 468.9000 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 2 | 01/02/2017 | 161.9700 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 3 | 01/03/2015 | 2203.1510 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 4 | 01/03/2016 | 1642.1744 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 1225 | 31/10/2017 | 3750.4990 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 1226 | 31/10/2018 | 523.9280 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 1227 | 31/12/2015 | 5253.2700 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 1228 | 31/12/2016 | 1381.3440 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 1229 | 31/12/2017 | 731.7680 | 0 | 0 | 2 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
1230 rows × 12 columns
[11]:
# set daily frequency where days with no orders have 0 sales
assert not sum(dfg.sales.isna())
dfg['date'] = pd.to_datetime(dfg['date'], format='%d/%m/%Y')
dfg = dfg.set_index('date')
dfg = dfg.loc[sorted(dfg.index)]
dfg = dfg.asfreq('d')
dfg[dfg.isna()] = 0 # na generated because of the asfreq; implies no sales
assert not sum(dfg.sales.isna())
dfg
[11]:
| sales | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| date | |||||||||||
| 2015-01-03 | 16.4480 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| 2015-01-04 | 288.0600 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| 2015-01-05 | 19.5360 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| 2015-01-06 | 4407.1000 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| 2015-01-07 | 87.1580 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 2018-12-26 | 814.5940 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| 2018-12-27 | 177.6360 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| 2018-12-28 | 1657.3508 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| 2018-12-29 | 2915.5340 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| 2018-12-30 | 713.7900 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
1458 rows × 11 columns
[12]:
# Moving average, for sazonality verification
def smoother(series, N):
res = pd.DataFrame(series).rolling(N,center=True).mean()
res = res.iloc[N-1:,0]
return res
fig, ax = plt.subplots(2,3,figsize=(15,10))
ax = ax.flatten()
for ax, smooth_factor in zip(ax, [7, 15, 30, 60, 180, 360]):
ax.plot(smoother(dfg.sales, smooth_factor))
[13]:
# Resampling, for sazonality verification
fig, ax = plt.subplots(2,3,figsize=(15,10))
ax = ax.flatten()
for ax, smooth_factor in zip(ax, [7, 15, 30, 60, 180, 360]):
dfg.sales.resample(f"{smooth_factor}D", label='right').mean().plot(ax=ax)
[14]:
# Sarima decomposition
decomposition = sma.tsa.seasonal_decompose(dfg.sales, model='additive')
fig = decomposition.plot()
plt.show()
[15]:
# Function to store pair true_value, pred_value for each prediction horizon
def append_true_and_pred(true_vals, pred_vals, res):
assert len(true_vals) == len(pred_vals)
true_vals = np.array(true_vals)
pred_vals = np.array(pred_vals)
for i, (true_v, pred_v) in enumerate(zip(true_vals, pred_vals)):
res[i][0].append(true_v)
res[i][1].append(pred_v)
[17]:
# try a bunch of models
tv_pv_pairs=dict()
models = [
'arma_1_1', 'armax_1_1', 'arima_1_1', 'arimax_1_1',
'arma_2_2', 'armax_2_2', 'arima_2_2', 'arimax_2_2',
'sarma_1_1', 'sarmax_1_1', 'sarima_1_1', 'sarimax_1_1',
'sarma_2_2', 'sarmax_2_2', 'sarima_2_2', 'sarimax_2_2',
]
for model in models.copy():
tv_pv_pairs[model] = [[[], []] for _ in range(31)]
try:
for i in range(31, 0, -1):
model_type = sm.tsa.arima.model.ARIMA if model[0] == 'a' else sma.tsa.SARIMAX
model_order = [-1,-1,1] if model[:3] == 'ari' or model[:4] == 'sari' else [-1,-1,0]
model_order[0] = model_order[1] = int(model[-1])
with_exog = model.split("_")[0][-1] == 'x'
exog_fit = dfg.iloc[:-i,1:] if with_exog else None
exog_forecast = dfg.iloc[-i:,1:] if with_exog else None
if i == 31:
print('-'*50)
print(model)
print(model_type)
print(model_order)
print(with_exog)
sm_model = model_type(dfg.sales[:-i], exog=exog_fit, order=model_order)
if model[0] == 's':
fitted = sm_model.fit(disp=0)
else:
fitted = sm_model.fit()
append_true_and_pred(dfg.sales[-i:], fitted.forecast(i, exog=exog_forecast), tv_pv_pairs[model])
except Exception as e:
print(model, "failed")
print(e)
models.remove(model)
del tv_pv_pairs[model]
--------------------------------------------------
arma_1_1
<class 'statsmodels.tsa.arima.model.ARIMA'>
[1, 1, 0]
False
--------------------------------------------------
armax_1_1
<class 'statsmodels.tsa.arima.model.ARIMA'>
[1, 1, 0]
True
--------------------------------------------------
arima_1_1
<class 'statsmodels.tsa.arima.model.ARIMA'>
[1, 1, 1]
False
--------------------------------------------------
arimax_1_1
<class 'statsmodels.tsa.arima.model.ARIMA'>
[1, 1, 1]
True
--------------------------------------------------
arma_2_2
<class 'statsmodels.tsa.arima.model.ARIMA'>
[2, 2, 0]
False
--------------------------------------------------
armax_2_2
<class 'statsmodels.tsa.arima.model.ARIMA'>
[2, 2, 0]
True
--------------------------------------------------
arima_2_2
<class 'statsmodels.tsa.arima.model.ARIMA'>
[2, 2, 1]
False
--------------------------------------------------
arimax_2_2
<class 'statsmodels.tsa.arima.model.ARIMA'>
[2, 2, 1]
True
--------------------------------------------------
sarma_1_1
<class 'statsmodels.tsa.statespace.sarimax.SARIMAX'>
[1, 1, 0]
False
--------------------------------------------------
sarmax_1_1
<class 'statsmodels.tsa.statespace.sarimax.SARIMAX'>
[1, 1, 0]
True
--------------------------------------------------
sarima_1_1
<class 'statsmodels.tsa.statespace.sarimax.SARIMAX'>
[1, 1, 1]
False
--------------------------------------------------
sarimax_1_1
<class 'statsmodels.tsa.statespace.sarimax.SARIMAX'>
[1, 1, 1]
True
--------------------------------------------------
sarma_2_2
<class 'statsmodels.tsa.statespace.sarimax.SARIMAX'>
[2, 2, 0]
False
--------------------------------------------------
sarmax_2_2
<class 'statsmodels.tsa.statespace.sarimax.SARIMAX'>
[2, 2, 0]
True
--------------------------------------------------
sarima_2_2
<class 'statsmodels.tsa.statespace.sarimax.SARIMAX'>
[2, 2, 1]
False
--------------------------------------------------
sarimax_2_2
<class 'statsmodels.tsa.statespace.sarimax.SARIMAX'>
[2, 2, 1]
True
[18]:
# Now let's try catboosting
for model in ['catb_30', 'catb_15', 'catb_7', 'catb_3']:
tv_pv_pairs[model] = [[[], []] for _ in range(31)]
param = {
'verbose': 200,
'iterations': 1_000,
'early_stopping_rounds': 100,
'task_type': 'GPU',
'verbose': 0,
}
rawdata = dfg.sales.to_numpy().ravel()
window = int(model.split("_")[1])
for i in tqdm(range(31, 0, -1)):
x_train = np.array([rawdata[i:i+window] for i in range(len(dfg.sales)-window)])
y_train = np.array([rawdata[i+window] for i in range(len(dfg.sales)-window)])
x_pred = x_train[-i:]
x_train = x_train[:-i]
y_train = y_train[:-i]
x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=40)
estimator = catb.CatBoostRegressor(**param)
estimator.fit(x_train, y_train, eval_set=(x_val, y_val))
forecast = []
for j in range(len(x_pred)):
if j:
x_pred[j, -j:] = forecast[:window]
forecast.append(estimator.predict(x_pred[[j]]).item())
forecast = np.array(forecast)
append_true_and_pred(dfg.sales[-i:], forecast, tv_pv_pairs[model])
if not model in models:
models.append(model)
100%|███████████████████████████████████████████| 31/31 [00:49<00:00, 1.58s/it]
100%|███████████████████████████████████████████| 31/31 [00:43<00:00, 1.40s/it]
100%|███████████████████████████████████████████| 31/31 [00:29<00:00, 1.04it/s]
100%|███████████████████████████████████████████| 31/31 [00:27<00:00, 1.13it/s]
[19]:
# get models metrics for a set of forecast horizons
def get_smape(y_true, y_pred):
y_true = np.array(y_true)
y_pred = np.array(y_pred)
res = np.sum(2 * np.abs(y_pred - y_true) / (np.abs(y_true) + np.abs(y_pred)))
res = 1 / len(y_true) * res
return res
def get_rmsle(y_true, y_pred):
try:
return np.sqrt(metrics.mean_squared_log_error(y_true, y_pred))
except ValueError as e:
#print(e)
return np.inf
def get_metrics(res):
for res_h in res:
yield (
metrics.mean_absolute_percentage_error(res_h[0], res_h[1]),
get_smape(res_h[0], res_h[1]),
get_rmsle(res_h[0], res_h[1]),
)
mape = dict()
smape = dict()
rmsle = dict()
for i,model in enumerate(models):
mape[model], smape[model], rmsle[model] = zip(*get_metrics(tv_pv_pairs[model]))
[20]:
# list of with models with/without exogenous variables
fxmodels = [m for m in models if m.split("_")[0][-1] == 'x']
fmodels = [m for m in models if m.split("_")[0][-1] != 'x']
[44]:
# plot metrics for each forecast horizon for models with exogenous variables
fig, axes = plt.subplots(2,1,figsize=(15,15))
axes = axes.flatten()
colors = ['black', 'red', 'blue', 'green', 'yellow', 'purple', 'pink', 'orange']
plt.sca(axes[0])
for i, model in enumerate(fxmodels):
plt.plot(range(1,len(smape[model])+1), smape[model], label=model, color=colors[i])
axes[0].set_xlabel("Forecast horizon")
axes[0].set_xlabel("SMAPE")
plt.legend(bbox_to_anchor=(0, 1.07, 1, 0), mode="expand", ncol=3, borderaxespad=0.)
plt.sca(axes[1])
for i, model in enumerate(fxmodels):
plt.plot(range(1,len(rmsle[model])+1), rmsle[model], label=model, color=colors[i])
axes[1].set_xlabel("Forecast horizon")
axes[1].set_xlabel("RMSLE")
[44]:
Text(0.5, 0, 'RMSLE')
[45]:
# plot metrics for each forecast horizon for models without exogenous variables
fig, axes = plt.subplots(2,1,figsize=(15,15))
axes = axes.flatten()
colors = ['magenta', 'red', 'blue', 'green', 'yellow', 'purple', 'pink', 'orange', 'black',
'#050505', '#a0a0a0', '#a5a5a5']
plt.sca(axes[0])
for i, model in enumerate(fmodels):
plt.plot(range(1,len(smape[model])+1), smape[model], label=model, color=colors[i])
axes[0].set_xlabel("Forecast horizon")
axes[0].set_xlabel("RSMAPE")
plt.legend(bbox_to_anchor=(0, 1.07, 1, 0), mode="expand", ncol=3, borderaxespad=0.)
plt.sca(axes[1])
for i, model in enumerate(fmodels):
plt.plot(range(1,len(rmsle[model])+1), rmsle[model], label=model, color=colors[i])
axes[1].set_xlabel("Forecast horizon")
axes[1].set_xlabel("RMSLE")
[45]:
Text(0.5, 0, 'RMSLE')
[46]:
# Best model for each forecast horizon including models with exogenous variables
best_df = pd.DataFrame(dict(
MAPE=[models[np.argmin([mape[model][h] for model in models])] for h in range(30)],
SMAPE=[models[np.argmin([smape[model][h] for model in models])] for h in range(30)],
RMSLE=[models[np.argmin([rmsle[model][h] for model in models])] for h in range(30)],
))
best_df.index = range(1, len(best_df)+1)
best_df.index.name='Horizon'
best_df
[46]:
| MAPE | SMAPE | RMSLE | |
|---|---|---|---|
| Horizon | |||
| 1 | arma_2_2 | catb_30 | catb_3 |
| 2 | armax_2_2 | catb_7 | catb_15 |
| 3 | catb_3 | catb_15 | catb_3 |
| 4 | catb_30 | catb_30 | catb_30 |
| 5 | catb_30 | catb_30 | catb_30 |
| 6 | armax_1_1 | catb_15 | catb_3 |
| 7 | catb_3 | catb_7 | catb_15 |
| 8 | catb_3 | catb_30 | catb_3 |
| 9 | catb_15 | catb_7 | catb_15 |
| 10 | catb_3 | catb_15 | catb_3 |
| 11 | catb_30 | catb_15 | catb_3 |
| 12 | catb_3 | catb_30 | catb_3 |
| 13 | catb_3 | catb_7 | catb_3 |
| 14 | catb_3 | arma_1_1 | catb_3 |
| 15 | catb_3 | catb_3 | catb_3 |
| 16 | catb_3 | catb_30 | catb_3 |
| 17 | catb_3 | catb_7 | catb_7 |
| 18 | catb_3 | catb_7 | catb_3 |
| 19 | catb_3 | catb_30 | catb_30 |
| 20 | catb_3 | catb_7 | catb_3 |
| 21 | catb_3 | catb_3 | catb_3 |
| 22 | catb_3 | catb_3 | catb_3 |
| 23 | catb_3 | catb_15 | catb_3 |
| 24 | catb_7 | catb_7 | catb_7 |
| 25 | catb_3 | catb_15 | catb_3 |
| 26 | catb_3 | catb_3 | catb_3 |
| 27 | catb_3 | catb_3 | catb_3 |
| 28 | catb_3 | catb_3 | catb_3 |
| 29 | catb_3 | catb_3 | catb_3 |
| 30 | catb_3 | catb_7 | catb_3 |
[47]:
# Best model for each forecast horizon, limited to non-exogenous models
# To be fair, let us not use models with exogenous data from now on
# since it in practice we will not have those exogenous variables
best_df = pd.DataFrame(dict(
MAPE=[fmodels[np.argmin([mape[model][h] for model in fmodels])] for h in range(30)],
SMAPE=[fmodels[np.argmin([smape[model][h] for model in fmodels])] for h in range(30)],
RMSLE=[fmodels[np.argmin([rmsle[model][h] for model in fmodels])] for h in range(30)],
))
best_df.index = range(1, len(best_df)+1)
best_df.index.name='Horizon'
best_df
[47]:
| MAPE | SMAPE | RMSLE | |
|---|---|---|---|
| Horizon | |||
| 1 | arma_2_2 | catb_30 | catb_3 |
| 2 | catb_30 | catb_7 | catb_15 |
| 3 | catb_3 | catb_15 | catb_3 |
| 4 | catb_30 | catb_30 | catb_30 |
| 5 | catb_30 | catb_30 | catb_30 |
| 6 | arma_1_1 | catb_15 | catb_3 |
| 7 | catb_3 | catb_7 | catb_15 |
| 8 | catb_3 | catb_30 | catb_3 |
| 9 | catb_15 | catb_7 | catb_15 |
| 10 | catb_3 | catb_15 | catb_3 |
| 11 | catb_30 | catb_15 | catb_3 |
| 12 | catb_3 | catb_30 | catb_3 |
| 13 | catb_3 | catb_7 | catb_3 |
| 14 | catb_3 | arma_1_1 | catb_3 |
| 15 | catb_3 | catb_3 | catb_3 |
| 16 | catb_3 | catb_30 | catb_3 |
| 17 | catb_3 | catb_7 | catb_7 |
| 18 | catb_3 | catb_7 | catb_3 |
| 19 | catb_3 | catb_30 | catb_30 |
| 20 | catb_3 | catb_7 | catb_3 |
| 21 | catb_3 | catb_3 | catb_3 |
| 22 | catb_3 | catb_3 | catb_3 |
| 23 | catb_3 | catb_15 | catb_3 |
| 24 | catb_7 | catb_7 | catb_7 |
| 25 | catb_3 | catb_15 | catb_3 |
| 26 | catb_3 | catb_3 | catb_3 |
| 27 | catb_3 | catb_3 | catb_3 |
| 28 | catb_3 | catb_3 | catb_3 |
| 29 | catb_3 | catb_3 | catb_3 |
| 30 | catb_3 | catb_7 | catb_3 |
[48]:
# Amount of times each model won
Counter(best_df.to_numpy().flatten())
[48]:
Counter({'arma_2_2': 1,
'catb_30': 14,
'catb_3': 51,
'catb_7': 12,
'catb_15': 10,
'arma_1_1': 2})
[49]:
# Average model error
for model in np.unique(best_df.to_numpy().flatten()):
print(model.ljust(11),
str(np.round(np.mean(mape[model]), 3)).ljust(7),
str(np.round(np.mean(smape[model]), 3)).ljust(7),
str(np.round(np.mean(rmsle[model]), 3)).ljust(7),
)
arma_1_1 3.107131938611645e+17 0.932 1.784
arma_2_2 9.762554570372993e+17 1.735 inf
catb_15 1.6025748446850157e+17 0.771 1.53
catb_3 1.3457102263941246e+17 0.739 1.389
catb_30 1.4317837800743165e+17 0.787 1.518
catb_7 1.655668907470554e+17 0.741 1.472
[50]:
# Given that, we choose catb_3
chosen_model = 'catb_3'
[51]:
true_values = tv_pv_pairs[chosen_model][0][0]
dynamic_forecast = tv_pv_pairs[chosen_model][0][1]
static_forecast = [tv_pv_pairs[chosen_model][i][1][-1] for i in range(len(tv_pv_pairs[chosen_model]))]
assert len(static_forecast) == len(dynamic_forecast)
assert len(true_values) == len(dynamic_forecast)
[52]:
final_df = dfg.copy()
final_df = final_df.iloc[-31:, [0]]
all(final_df.iloc[:,0] == true_values)
[52]:
True
[53]:
final_df['dynamic forecast'] = dynamic_forecast
final_df['static forecast'] = static_forecast
final_df['MAPE'] = mape[chosen_model]
final_df['SMAPE'] = smape[chosen_model]
final_df['RMSLE'] = rmsle[chosen_model]
[55]:
final_df
[55]:
| sales | dynamic forecast | static forecast | MAPE | SMAPE | RMSLE | |
|---|---|---|---|---|---|---|
| date | ||||||
| 2018-11-30 | 6645.2820 | 1579.885138 | 1381.737779 | 3.106174e+17 | 0.773992 | 1.880974 |
| 2018-12-01 | 5331.1780 | 1661.055884 | 1322.080279 | 2.780335e+17 | 0.780637 | 1.903091 |
| 2018-12-02 | 9951.1820 | 2295.425096 | 1365.699152 | 2.477391e+17 | 0.776961 | 1.909545 |
| 2018-12-03 | 1403.8420 | 1613.165573 | 1129.972210 | 2.892465e+17 | 0.744544 | 1.908326 |
| 2018-12-04 | 2639.6380 | 3675.007987 | 1634.538122 | 2.969482e+17 | 0.785207 | 1.966193 |
| 2018-12-05 | 1453.1360 | 1636.171255 | 1777.265236 | 2.752879e+17 | 0.826421 | 1.994819 |
| 2018-12-06 | 10.6800 | 1620.562259 | 1569.417221 | 2.902782e+17 | 0.835272 | 2.037926 |
| 2018-12-07 | 2916.5140 | 1562.400791 | 1665.965092 | 2.921562e+17 | 0.813742 | 1.813769 |
| 2018-12-08 | 7643.0410 | 1890.732618 | 1680.781767 | 4.155220e+17 | 0.796329 | 1.887826 |
| 2018-12-09 | 5470.3900 | 1524.437425 | 2014.867056 | 3.279555e+17 | 0.788983 | 1.858189 |
| 2018-12-10 | 3873.5590 | 1717.154266 | 2167.933804 | 3.777282e+17 | 0.745360 | 1.889742 |
| 2018-12-11 | 2025.0350 | 1403.989314 | 1699.883499 | 3.836697e+17 | 0.796112 | 1.940526 |
| 2018-12-12 | 0.0000 | 2138.098352 | 1556.429934 | 3.865192e+17 | 0.814682 | 1.989378 |
| 2018-12-13 | 580.9360 | 2082.946266 | 1567.840671 | 1.377434e+00 | 0.727771 | 1.005086 |
| 2018-12-14 | 3897.7140 | 1453.233810 | 1486.190010 | 1.336495e+00 | 0.683824 | 0.995467 |
| 2018-12-15 | 306.8880 | 1473.546740 | 1892.628423 | 1.374497e+00 | 0.718033 | 1.020822 |
| 2018-12-16 | 858.7020 | 1021.082402 | 1550.243866 | 1.380527e+00 | 0.671705 | 1.000987 |
| 2018-12-17 | 2027.7580 | 1769.024422 | 1608.064481 | 1.192719e+00 | 0.660408 | 0.985577 |
| 2018-12-18 | 3645.9110 | 1592.930355 | 1433.254277 | 1.414952e+00 | 0.697012 | 1.018109 |
| 2018-12-19 | 1895.9260 | 2027.992818 | 1859.294676 | 1.500920e+00 | 0.716597 | 1.078780 |
| 2018-12-20 | 377.7360 | 1229.661154 | 1513.090017 | 1.457978e+00 | 0.721132 | 1.035881 |
| 2018-12-21 | 2140.9400 | 1931.339359 | 1798.317340 | 1.298428e+00 | 0.665431 | 1.025325 |
| 2018-12-22 | 7442.0210 | 1893.638955 | 1780.265133 | 1.485212e+00 | 0.776181 | 1.105266 |
| 2018-12-23 | 1926.7760 | 2149.547163 | 1589.280811 | 1.859388e+00 | 0.683736 | 1.087954 |
| 2018-12-24 | 6233.0540 | 3050.939678 | 1611.367737 | 1.755990e+00 | 0.797701 | 1.104065 |
| 2018-12-25 | 2698.9270 | 1730.174803 | 1556.921230 | 1.981585e+00 | 0.654126 | 1.060342 |
| 2018-12-26 | 814.5940 | 1626.333499 | 2122.081710 | 2.415952e+00 | 0.785068 | 1.193049 |
| 2018-12-27 | 177.6360 | 1616.229932 | 1602.056325 | 2.462046e+00 | 0.723857 | 1.204145 |
| 2018-12-28 | 1657.3508 | 1419.269849 | 1761.322919 | 6.330888e-01 | 0.462434 | 0.606322 |
| 2018-12-29 | 2915.5340 | 1448.019351 | 1703.835757 | 9.138620e-01 | 0.692136 | 0.739220 |
| 2018-12-30 | 713.7900 | 1381.737779 | 1630.665688 | 1.284517e+00 | 0.782165 | 0.825368 |