def model(self, column_name, df, apply_smoothing, smoothing_level=None):
"""
performs predictions using the simple exponential smoothing model approach
:input column_name : str, name of column to hold the predicted values
:input df : dataframe, weekly-level data
:input apply_smoothing : bool, indicates whether to factor-in smoothing parameters in the Holt model
:input smoothing_level : int, default=None, l parameter in Simple Exponential Smoothing model
:returns df : dataframe, weekly-level, with predictions
"""
m = self.prediction_period
if apply_smoothing == True:
fit1 = SimpleExpSmoothing(df["train"][:-m]).fit(
smoothing_level=smoothing_level, optimized=True)
params = None
elif apply_smoothing == False:
fit1 = SimpleExpSmoothing(df["train"][:-m]).fit(optimized=True)
params = fit1.params
y_fit = fit1.fittedvalues
y_fore = fit1.forecast(m)
df[column_name] = np.nan
#df[column_name][:-1] = y_fit
df[column_name][:-m] = df['train'].iloc[:-m]
df[column_name][-m:] = y_fore
#df[column_name].iloc[-1:] = list(y_pred)[-1]
return df
def sem(obs, p_mean:bool, boxcox:bool, n_forecast:int):
'''
Implement a Simple Exponential Smoothin with Grid Search to find the model with the best
combination of parameters based on an SSE minimization.
Input:
:param obs: sequential data for forecasting
:param p_mean: wether or not to apply penalized_mean
:param boxcox: wether or not to apply boxcox transformation
:param n_forecast: number of observations to forecast
Output:
forecast: Forecast for the next n_forecast observations
aic_min: AIC of model
bic: BIC of model
mse: MSE of model
sse_min: SSE of model
'''
assert type(obs) == pd.core.series.Series, "Data must be of pandas Series type"
if p_mean == True:
y_prc = pp_transforms().penalized_mean(obs)
else:
y_prc = obs
untransform = False
if boxcox == True:
y_prc, lmbd = pp_transforms().boxcox(y_prc)
untransform = True
# Perform a Grid Search to find the best model
sse_min, sl = 1000000000, 0
print(obs)
print(y_prc)
try:
for i in [.95,.9,.85,.8,.75,.7,.65,.6,.55,.5,.45,.4,.35,.3,.25,.2,.15,.1,.05]:
mdl = SimpleExpSmoothing(y_prc).fit(smoothing_level = i, optimized = False)
sse = np.sum((y_prc.astype(float) - mdl.fittedvalues)**2)
if sse <= sse_min:
sse_min, sl = sse, i
except Exception as e:
print("Error while fitting the model","\n",e)
# Best model
mdl = SimpleExpSmoothing(y_prc).fit(smoothing_level = sl,optimized=False)
aic, bic = mdl.aic, mdl.bic
# Untransform Box-Cox Data
if untransform:
pred = pp_transforms().boxcox_untransform(mdl.fittedvalues, lmbd)
forecast = pp_transforms().boxcox_untransform(mdl.forecast(n_forecast), lmbd)
mse = ((obs.astype(float).values - pred)**2).mean()
else:
pred = mdl.fittedvalues
forecast = mdl.forecast(n_forecast)
mse = ((obs.astype(float).values - pred)**2).mean()
return (forecast, aic, bic, mse, sse_min)
def fit_SES(params, df):
data = df.carbon_monoxide[:-1]
if 'a' in params:
a = params['a']
fit = SimpleExpSmoothing(
data,
initialization_method="estimated").fit(smoothing_level=float(a))
else:
fit = SimpleExpSmoothing(data, initialization_method="estimated").fit()
fcast = fit.forecast(1)
return fcast[0], fit.model.params['smoothing_level']
def simple_exponetial_smoothing_forX(arr, alpha=0.3):
# initialization
sample_size = int(arr.shape[0])
time_size = int(arr.shape[1])
feature_size = int(arr.shape[2])
# create empty array
smoothing_arr = np.zeros((sample_size, time_size, feature_size - 1))
for idx, temp_arr in enumerate(arr):
for col in range(1, feature_size): # open col is 1 index
if col < 5:
temp_series = temp_arr[:, col].reshape(-1)
smoother = SimpleExpSmoothing(temp_series, initialization_method="heuristic").fit(smoothing_level=0.3,optimized=False)
temp_smoothing_series = smoother.fittedvalues
smoothing_arr[idx, :, col-1] = temp_smoothing_series
else:
pass_series = temp_arr[:, col].reshape(-1)
smoothing_arr[idx, :, col-1] = pass_series
return smoothing_arr
def forecast_ses(og_df):
if len(og_df) <= 1:
result = [0, 0]
else:
df = og_df.copy()
train = aggregate_by_day(df)
test = train.copy()
#print('train df before create split')
#print(train)
test = test.reindex(create_split(train))
y_hat_avg = test.copy()
fit2 = SimpleExpSmoothing(np.asarray(train['Count'])).fit(
smoothing_level=0.6, optimized=False)
y_hat_avg['SES'] = fit2.forecast(len(test))
plt.figure(figsize=(16, 8))
plt.plot(train['Count'], label='Train')
plt.plot(y_hat_avg['SES'], label='SES')
plt.legend(loc='best')
# plt.show()
# print(y_hat_avg['SES'].iloc[0])
# get max y value and index (x)
date_projected = str(y_hat_avg['SES'].idxmax())
qty_projected = str(y_hat_avg.loc[y_hat_avg['SES'].idxmax(), 'SES'])
result = [date_projected, qty_projected]
return result
def simpleExpSmoothing(x, y, save_fn):
pred = []
for data in x:
fit = SimpleExpSmoothing(data).fit(smoothing_level=0.6,
optimized=False)
pred.append(fit.forecast(1))
save_fn('rate_simpleExpSmoothing.txt', np.array(pred), y)
def simple_exponential_smoothing(input_df, kunag, matnr, alpha=0.6):
df = input_df.copy()
df = remove_negative_rows(df)
df_series = individual_series(df, kunag, matnr)
df_series = data_transformation.get_weekly_aggregate(df_series)
df_series["date"] = df_series["dt_week"].map(str)
df_series["date"] = df_series["date"].apply(lambda x: x.replace("-", ""))
df_series["prediction"] = df_series["quantity"]
df_series_train, df_series_test = splitter(df_series)
k = 0
for index, row in df_series_test.iterrows():
fit2 = SimpleExpSmoothing(np.asarray(df_series_train["quantity"])).fit(
smoothing_level=alpha, optimized=False)
row["prediction"] = fit2.forecast(1)
df_series_train = pd.concat([df_series_train,
pd.DataFrame(row).T
]).reset_index(drop=True)
if k == 0:
test_index = df_series_train.shape[0] - 1
k = 1
output_df = df_series_train
test_df = df_series_train.iloc[test_index:]
# print("mean squared error is :",mean_squared_error(output_df["quantity"], output_df["prediction"]))
return output_df, mean_squared_error(test_df["quantity"],
test_df["prediction"])
def produce(self,
*,
inputs: Inputs,
timeout: float = None,
iterations: int = None) -> base.CallResult[Outputs]:
self.logger.info('Simple Exponential Smoothing Primitive called')
outputs = inputs
self._training_inputs, self._training_indices = self._get_columns_to_fit(
inputs, self.hyperparams)
try:
columns_to_calculate_simple_exponential_smoothing = List[str]
if (self.hyperparams['use_columns'] == ()):
columns_to_calculate_simple_exponential_smoothing = list(
set(inputs.columns) -
set(['d3mIndex', 'timestamp', 'ground_truth']))
else:
columns_to_calculate_simple_exponential_smoothing = self.hyperparams[
'use_columns']
for column in self._training_indices:
outputs[inputs.columns[column] +
"_simple_exponential_smoothing"] = SimpleExpSmoothing(
inputs.iloc[:,
column]).fit(
smoothing_level=0.2,
optimized=False).fittedvalues
except Exception as e:
self.logger.error(
"Error in Calculating simple exponential smoothing", e)
self._update_metadata(outputs)
#print(inputs)
#print("-------------")
print(outputs)
return base.CallResult(outputs)
def SES_find_smoothng_level(df):
train = df[0:20]
test = df[20:]
num = 0
rms = 0.0
for id_idx in range(1, 25, 1):
if train['id' + str(id_idx)].sum() > 25:
fit = SimpleExpSmoothing(train['id' + str(id_idx)]).fit(
smoothing_level=0.1, optimized=False)
fcast = fit.forecast(len(test))
plt.plot(train['id' + str(id_idx)],
marker='o',
label='train_id' + str(id_idx))
plt.plot(test['id' + str(id_idx)],
marker='o',
label='test_id' + str(id_idx))
plt.plot(fcast, marker='o', label='SES' + str(id_idx))
plt.legend(loc='best')
rms = rms + math.sqrt(
mean_squared_error(test['id' + str(id_idx)], fcast))
num = num + 1
plt.show()
mean_error = rms / num
#print(mean_error)
return 0.1
def ses(input_df, kunag, matnr, n, alpha):
i = 0
lst = []
test1 = train_test_split(df, kunag, matnr, n)[1]
y_hat_avg = test1.copy()
for i in range(n, 0, -1):
train, test = train_test_split(df, kunag, matnr, i)
dd = np.asarray(train["quantity"])
fit2 = SimpleExpSmoothing(np.asarray(train['quantity'])).fit(
smoothing_level=alpha, optimized=False)
y_hat_avg['SES'] = fit2.forecast(len(test1))
pred = y_hat_avg['SES']
lst.append(pred.iloc[-1])
pd.DataFrame(lst)
y_hat_avg['pred_column'] = lst
plt.figure(figsize=(12, 8))
plt.plot(train.set_index("date")['quantity'], label='Train', marker='.')
plt.plot(test1.set_index("date")['quantity'], label='Test', marker='.')
plt.plot(y_hat_avg.set_index("date")['pred_column'],
label='SES',
marker='.')
plt.legend(loc='best')
plt.title("SES")
plt.show()
rms = sqrt(mean_squared_error(test1.quantity, y_hat_avg.pred_column))
mae = mean_absolute_error(test1.quantity, y_hat_avg.pred_column)
del y_hat_avg['SES']
return y_hat_avg, rms, mae
def get_gravity_from_acc(acc):
"""
Get gravity from acc data by using low pass filter
Parameters
----------
acc : numpy array
[time, x, y, z]
Returns
--------
gravity : 1-D array
The gravity component in 3 axis
"""
if debug:
print('get gravity component...')
gravity_component = [0.0] * 3
# use low pass filter (exponential)
# https://developer.android.com/guide/topics/sensors/sensors_motion
# https://medium.com/datadriveninvestor/how-to-build-exponential-smoothing-models-using-python-simple-exponential-smoothing-holt-and-da371189e1a1
for i in range(3):
# TODO: we don't need to use all rows to get the component
top_rows = min(len(acc), 1000)
acc_x = acc[:top_rows, i + 1]
fit_x = SimpleExpSmoothing(acc_x).fit()
fcast_x = fit_x.forecast(1)
gravity_component[i] = fcast_x[0]
if debug:
print_floats(*gravity_component, description="Gravity component:")
return gravity_component
def run(self): # Default called function with mythread.start()
print("{} started!".format(self.getName()))
data = pd.read_csv('cpuData.txt',
dtype={
'date': str,
'cpu': np.float16
},
sep=';',
names=['date', 'cpu'])
data = format_values(data)
# Simple Exponential Smoothing
fit = SimpleExpSmoothing(data['cpu']).fit(smoothing_level=0.2,
optimized=False)
data['cpu'] = fit.fittedvalues
# END
data = filter_day(data)
if self.small:
data = data.tail(50)
plot_data([data.to_numpy()], ['0.2'], is_cpu=True)
print("{} finished!".format(self.getName()))
def ses1(input_df, kunag, matnr, n, l, alpha):
index = str(kunag) + "-" + str(matnr)
dfw = n_series(df, kunag, matnr)
test1 = test(df, kunag, matnr)
mae = []
for i in range(0, l):
k = n - i
lst = []
train1, cv1 = train_cv_split(df, kunag, matnr, k, l)
y_hat_avg = cv1.copy()
for j in range(k, l, -1):
train, cv = train_cv_split(df, kunag, matnr, j, l)
dd = np.asarray(train["quantity"])
fit2 = SimpleExpSmoothing(np.asarray(train['quantity'])).fit(
smoothing_level=alpha, optimized=False)
y_hat_avg['SES'] = fit2.forecast(len(cv1))
pred = y_hat_avg['SES']
lst.append(pred.iloc[-1])
pd.DataFrame(lst)
y_hat_avg['pred_column'] = lst
rms = sqrt(mean_squared_error(cv1.quantity, y_hat_avg.pred_column))
mae1 = mean_absolute_error(cv1.quantity, y_hat_avg.pred_column)
mae.append(mae1)
del y_hat_avg['SES']
l = l - 1
return mae
def gen_plot_forecast():
es_conn = fetchData.elasticSearch(
url="https://*****:*****@kf6-stage.ikit.org/es/_search")
df = es_conn.get_nginx_reliability(interval='1h')
df = df.sort_values('buckets', ascending=True)
data_in_window = df.tail(1000)
rel_data = data_in_window["reliability"].to_numpy()
date_data = data_in_window["buckets"].to_numpy()
last_bucket = parse(data_in_window["buckets"].iloc[-1])
if np.isnan(np.sum(rel_data)):
print("NaN in data")
exit(1)
simple_exp_model = SimpleExpSmoothing(rel_data).fit(smoothing_level=.5)
predicted_data = np.average(simple_exp_model.forecast(5))
fitted_values = simple_exp_model.fittedvalues
for idx, val in enumerate(fitted_values):
if val < 0.5:
fitted_values[idx] = 0.5
fig = go.Figure()
fig.add_trace(
go.Scatter(x=date_data,
y=rel_data,
mode='lines',
name="Observed Reliability"))
fig.add_trace(
go.Scatter(x=date_data,
y=fitted_values,
mode='lines',
name="Predicted Reliability"))
print(datetime.datetime.now())
sys.stdout.flush()
tm = datetime.datetime.now()
return fig, predicted_data, last_bucket, tm
def simple_exponential_smoothing_statsmodels():
N, t, alpha, x0 = 200, 160, 0.5, 20
realisations = pd.Series(sample_gaussian_process(20, 5, N), range(N))
mod = SimpleExpSmoothing(realisations[:t+1]).fit(smoothing_level=alpha, initial_level=x0, optimized=False)
forecasts = mod.forecast(N-(t+1)).rename(r'$\alpha=0.5$')
plot(realisations, pd.Series(np.nan, range(t+1)).append(forecasts), alpha)
py.show()
def evaluate_simp_avg_model(X):
"""
Evaluate a Simple Expontential Smoothing Model
:param X: list or series containing all historical data
:return: mse (error metric) and the fitted model
"""
# Prepare training dataset
train_size = int(len(X) * 0.75)
train, test = X[0:train_size], X[train_size:]
history = [x for x in train]
# Make predictions
predictions = list()
for t in range(len(test)):
# Fit model
model = SimpleExpSmoothing(history)
model_fit = model.fit(smoothing_level=0.6, optimized=False)
# Forecast
yhat = model_fit.forecast()[0]
# Store prediction and move forward one time step
predictions.append(yhat)
history.append(test[t])
# calculate out of sample error
mse = mean_squared_error(test, predictions)
return mse, model_fit
def simple_smoothing(data,company):
fit1 = SimpleExpSmoothing(data).fit(smoothing_level=0.1, optimized=False)
fcast1 = fit1.forecast(1).rename(r'$\alpha=0.1$')
print(fit1['fittedvalues'])
fcast1.plot(marker='o', color='blue', legend=True)
fit1.fittedvalues.plot(marker='o', color='blue')
pyplot.title("Trade War Impact - "+company)
pyplot.show()
def exponential_smoothing(x, alpha):
"""
Apply an exponential smoothing to the input data
:param x: input time-series
:param alpha: smoothness (between 0 and 1) coefficient
:return: smoothed time-series
"""
return SimpleExpSmoothing(x).fit(alpha).fittedvalues
def simpleExponentialSmoothing(self, train, test, trend):
if trend == 'no trend':
sm = SimpleExpSmoothing(train).fit()
sm_pred = sm.forecast(len(test))
rmse_sm = rootMeanSquaredError(test, sm_pred)
if rmse_sm < self.rmse:
self.rmse = rmse_sm
self.__model__ = 'simpleExponentialSmoothing'
def estimate_SES(dataframe, name, alpha, sizeestimate):
array = np.asarray(dataframe[name])
model = SimpleExpSmoothing(array)
fit = model.fit(smoothing_level=alpha,optimized=False)
forecast = fit.forecast(sizeestimate)
for index in range ( len(forecast) ):
forecast[index] = round(forecast[index], 4)
return forecast
def ES(data=None, horizon=24, alpha=0.3):
"""
Build ES model => Train model => Get forecasts.
:param data: history data.
:param horizon: Length of forecasts.
:return: list, Forecasts in next h time steps.
"""
model = SimpleExpSmoothing(data).fit(smoothing_level=alpha)
fcasts = model.predict(start=len(data), end=len(data) + horizon)
return fcasts
def sess(i, fc_periods, df):
df = df
saledata = np.asarray(df.iloc[0:, 0])
fit1 = SimpleExpSmoothing(saledata).fit(smoothing_level=0.2,
optimized=True)
suav = fit1.fittedvalues
df['pronostico'] = suav
nombre = list(df.columns.values.tolist())
fcast1 = fit1.forecast(fc_periods)
return (fcast1)
def fit_model(self, n_predict):
fit = SimpleExpSmoothing(self.train).fit()
forecast = fit.forecast(n_predict)
ds = self.ds_test
self.forecast = pd.DataFrame({"ds": ds, "yhat": forecast})
return self.forecast
def plot_all_graphs(data,company):
i=0
for df in data:
fit1 = SimpleExpSmoothing(df['sentiment']).fit(smoothing_level=1, optimized=False)
fcast1 = fit1.forecast(1).rename(r'$\alpha=0.1$')
fcast1.plot(marker='o', color='blue', legend=True)
fit1.fittedvalues.plot(marker='o', color='blue')
pyplot.title("Trade War Impact - " + company[i])
i=i+1
pyplot.show()
def twocolorball_ses_forecast(df):
l = []
for i in range(1, 8):
column = "红球%d" % i if i < 7 else "蓝球"
fit_model = SimpleExpSmoothing(np.asarray(df[column])).fit(
smoothing_level=random.randint(1, 10) / 10, optimized=False)
predict = fit_model.predict()
l.append(int(predict[0]))
print(l)
return l
def SES(paramsList=['pollution.csv', '0.93','pm', 'humidity', 'date'],specialParams=['0.6']):
'''
1.时间序列比较平稳时，选择较小的α值，0.05-0.20。
2.时间序列有波动，但长期趋势没大的变化，可选稍大的α值，0.10-0.40。
3.时间序列波动很大，长期趋势变化大有明显的上升或下降趋势时，宜选较大的α值，0.60-0.80。
4.当时间序列是上升或下降序列，满足加性模型，α取较大值，0.60-1。
'''
path = paramsList[0]
trainRows = float(paramsList[1])
saveto = 'result.csv'
df = pd.read_csv(path, usecols=paramsList[2:])
allRows = df.shape[0]
es = specialParams[0]
train = df[0:int(allRows*trainRows)]
test = df[int(allRows*trainRows)+1:]
df['Timestamp'] = pd.to_datetime(df[paramsList[-1]], format='%Y/%m/%d %H:%M')
df.index = df['Timestamp']
df = df.resample('D').mean()
train['Timestamp'] = pd.to_datetime(train[paramsList[-1]], format='%Y/%m/%d %H:%M')
train.index = train['Timestamp']
train = train.resample('D').mean()
test['Timestamp'] = pd.to_datetime(test[paramsList[-1]], format='%Y/%m/%d %H:%M')
test.index = test['Timestamp']
test = test.resample('D').mean()
y_hat = test.copy()
nullArray = train.copy()
nullArray['time'] = train.index
# 以上可通用----------------------------
for i in range(2,len(paramsList)-1):
fit = SimpleExpSmoothing(np.asarray(train[paramsList[i]])).fit(smoothing_level=float(es), optimized=False)
y_hat[paramsList[i]] = fit.forecast(len(test))
y_hat[paramsList[i]] = round(y_hat[paramsList[i]],2)
rms = sqrt(mean_squared_error(test[paramsList[i]], y_hat[paramsList[i]]))
print(rms)
y_hat['time'] = test.index
yhat_naive = np.array(y_hat)
nArray = np.array(nullArray)
newArray = np.concatenate((nArray,yhat_naive),axis=0)
s = pd.DataFrame(newArray, columns=paramsList[2:])
for i in range(2,len(paramsList)-1):
s[paramsList[i]][0:int(len(s)*trainRows)] = ""
s.to_csv(saveto,index=False,header=True,float_format='%.2f')
def estimate_SES(dataframe, name, alpha, sizeestimate):
# SES requires an array to work with, so we convert the column into an array
array = np.asarray(dataframe[name])
model = SimpleExpSmoothing(array)
fit = model.fit(smoothing_level=alpha,optimized=False)
# because this model assumes no trend or seasonality
# all forecasts can be the same, i.e. a straight line
forecast = fit.forecast(sizeestimate)
for index in range ( len(forecast) ):
forecast[index] = round(forecast[index], 4)
return forecast
def sesm(i):
df = i
train = np.asarray(df.iloc[:(round(len(df) * .85)), 0])
hell = df.iloc[(round(len(df) * .85)):, 0]
fit1 = SimpleExpSmoothing(train).fit(smoothing_level=0.2, optimized=True)
suav = fit1.fittedvalues
fcast1 = fit1.forecast(len(hell))
sreal = (sum(hell))
spred = (sum(fcast1))
mape = calculomape(sreal, spred)
return (mape)
def SES_f(self, df, a):
try:
simpleexp = SimpleExpSmoothing(np.array(np.array(df['Actual'])))
fit_simpleexp = simpleexp.fit(smoothing_level=a,optimized=False)
forecast = fit_simpleexp.forecast()[0]
Cluster, Warehouse, WF, YF = generate_attrib(df)
self.df_forecast.append({'Cluster':Cluster, 'Warehouse':Warehouse, 'Year':YF, "Week": WF, "Forecast":forecast})
return print(f'DEBUG:Forecast:{Cluster}:{Warehouse}:{YF}:{WF}:{forecast}')
except:
return print("ERROR:FORECAST-SES")
def ewma(data, col, train, test, frequency):
y_hat_avg = test.copy()
fit2 = SimpleExpSmoothing(np.asarray(train[col])).fit(smoothing_level=0.6,optimized=False)
y_hat_avg['SES'] = fit2.forecast(len(test))
print('Rmse= ', rmse(test[col], y_hat_avg['SES']))
plt.figure(figsize=(16,8))
plt.plot(train[col], label='Train')
plt.plot(test[col], label='Test')
plt.plot(y_hat_avg['SES'], label='Exponential Smoothing')
plt.legend(loc='best')
plt.savefig(frequency+'ses.png')
