# Load RICs
ek.set_app_key('DEFAULT_CODE_BOOK_APP_KEY')
sd = datetime.date(day = 1, month = 1, year = 2018).strftime('%d-%m-%Y')
ed = datetime.date.today().strftime('%d-%m-%Y')
a_interval = 'daily'
 
ric_names = ['SH-CCN-IRNOR62','SH-WUH-3HRCOIL']
rics = {}
 
def get_timeseries(ric):
    print(f'Getting timeseries data of {ric}')
    df = ek.get_timeseries(ric, start_date = sd, interval = a_interval)
    df.columns.values[0] = 'Value'
    df['Date'] = pd.to_datetime(df.index)    
    df['Price'] = df['Value']
    rics[ric] = df
    
    
for ric in ric_names:
    get_timeseries(ric)

        
        Copy
    

    	
            
# Prep Data
ric_prep = {}
def prep_data(ric):
    this_ric = rics[ric]
    this_ric['Price'] = this_ric['Price'].apply(lambda x:float(x)) # Price values as float
    this_ric = this_ric.resample('D').mean() # resample for daily
    this_ric['Date_Day'] = this_ric.index
    this_ric['Date_Day'] = this_ric['Date_Day'].dt.strftime('%Y-%m-%d').copy() # changing Date format
 
    holidays = this_ric[this_ric['Price'].isin([np.nan])] # saving holidays and weekends for further predictions in prophet
    holidays = holidays.Date_Day.tolist()
    holidays_df = pd.DataFrame({'holiday' : 'weekend', 'ds' : pd.to_datetime(holidays)})
 
    this_ric['Price'] = this_ric['Price'].fillna(method='ffill') # replacing gaps in values
    this_ric = this_ric.reset_index(drop=True)
    
    ric_prep[ric] = this_ric
    return holidays_df
 
 
for ric in rics:
    holidays_df = prep_data(ric)
print('Done Loading RICs...')

        
        Copy
    

    	
            
# Augmented Dickey-Fuller Test
def adf_test(ric):
    ric_adf = ric_prep[ric].copy()
    vals = ric_adf.Price
    diff = vals.diff(periods=1).dropna()
    test = sm.tsa.adfuller(diff)
    print('adf:',test[0])
    print('====> p-value:',test[1])
    print('Critical values:',test[4])
    if test[0] > test[4]['5%']:
        print(f'{ric} non stationary')
    else:
        print(f'{ric} stationary! The row of first differences is stationary and the initial row is an integrated row of the first order')
        
        
for ric in ric_prep:
    adf_test(ric)
print('done ADF Test...')

        
        Copy
    

    	
            adf: -21.404872748901784

====> p-value: 0.0

Critical values: {'1%': -3.433926456232207, '5%': -2.863119805756803, '10%': -2.5676110344808043}

SH-CCN-IRNOR62 stationary! The row of first differences is stationary and the initial row is an integrated row of the first order

adf: -9.09112259902564

====> p-value: 3.821856158152293e-15

Critical values: {'1%': -3.4339580289794225, '5%': -2.8631337437669275, '10%': -2.567618456010139}

SH-WUH-3HRCOIL stationary! The row of first differences is stationary and the initial row is an integrated row of the first order

done ADF Test...
        
        Copy
    

    	
            ric_train = ric[:-period]
        
        Copy
    

    	
            
# Test Prediction
ric_plot = {}
for ric_name in ric_prep:
    print(f'run Test {ric_name} Predict...')
    period = 30
    ric = ric_prep[ric_name]
    ric = pd.DataFrame({'ds' : ric.Date_Day, 'y' : ric.Price})
    ric['y'] = np.log(ric['y'])
    ric = ric.reset_index(drop=True)
    
    ric_train = ric[:-period]
    pr_model = Prophet(interval_width=0.95, # The width of the uncertainty intervals (by default 80%) can be set using the parameter interval_width
                       n_changepoints=40,
                       changepoint_prior_scale=0.08,
                       changepoint_range=0.09,   
                       yearly_seasonality=True,
                       weekly_seasonality=False,
                       daily_seasonality=False,
                       holidays=holidays_df)
    pr_model.fit(ric_train)
    
    future = pr_model.make_future_dataframe(periods = period, freq = 'D') # df for date +30 forecast days
    forecast = pr_model.predict(future)
 
    forecast['ds'] = forecast['ds'].dt.strftime('%Y-%m-%d').copy()
    cmp_df = forecast.set_index('ds')[['yhat_lower','yhat_upper','yhat']].join(ric.set_index('ds')).copy() #df compare
    cmp_df['MAE'] = cmp_df['y'] - cmp_df['yhat'] # mean absolute error (on average the models error)
    print('MAE', np.mean(abs(cmp_df[-period:]['MAE'])))
 
    # Prediction
    print(f'run {ric_name} Predict...')
    
    ric = ric_prep[ric_name]
    ric = pd.DataFrame({'ds':ric.Date_Day,'y':ric.Price})
    pr_model = Prophet(interval_width=0.95, # The width of the uncertainty intervals (by default 80%) can be set using the parameter interval_width
                       n_changepoints=40, # 25 by default
                       changepoint_prior_scale=0.08, # 0.05 by default
                       changepoint_range=0.09,   # 0.8 by default
                       yearly_seasonality=True,
                       weekly_seasonality=False,
                       daily_seasonality=False,
                       holidays=holidays_df)
 
    ric['y'] = np.log(ric['y']) 
    ric = ric.reset_index(drop=True)
    pr_model.fit(ric)
    future = pr_model.make_future_dataframe(periods = period, freq = 'D')
    forecast = pr_model.predict(future)
 
    ric['y'] = np.exp(ric['y']) #converting back into numerical values 
    forecast['yhat_lower'] = np.exp(forecast['yhat_lower'])
    forecast['yhat_upper'] = np.exp(forecast['yhat_upper'])
    forecast['yhat'] = np.exp(forecast['yhat']) 
 
    forecast['ds'] = forecast['ds'].dt.strftime('%Y-%m-%d').copy() #creating a table with actual data and predictions 
    prophet_ric = pd.DataFrame({'Date':forecast.ds,'Price':forecast.yhat})
    prophet_ric = prophet_ric.tail(period).copy()
 
    ric = ric.rename(columns={'ds':'Date','y':'Price'}) #renaming back our columns as date and price
    ric = pd.concat([ric, prophet_ric])
    ric = ric.reset_index(drop=True)
    ric_plot[ric_name] = ric
    print(f'done {ric_name} Predict...')

        
        Copy
    

    	
            
# Plot
for ric_name in ric_plot:
    if ric_name == 'SH-CCN-IRNOR62':
        color = 'cyan'
        title_text = 'SteelHome 62% CFR China & forecast'
        yaxis_title = 'Price USD'
    elif ric_name == 'SH-WUH-3HRCOIL':
        color = 'crimson'
        title_text = 'SteelHome HRC Wuhan & forecast'
        yaxis_title = 'Price CNY'
    else:
        color = ''
        title_text = ''
        yaxis_title = ''
    
    ric = ric_plot[ric_name]
    fig = go.Figure()
    fig.add_trace(go.Scatter(x=ric.Date.values, y=ric.Price.values, name=ric_name, line=dict(color=color)))
    start_pred = ric.head(1).reset_index(drop = True).loc[0, 'Date']
    fig.update_layout(title_text=title_text, yaxis_title=yaxis_title)
    fig.show()
    
print('Finish')

        
        Copy