def fbp(ts, stock, pers=90):
"""
Plots the forecast of the FBProphet model
"""
plt.style.use('fivethirtyeight')
# Instantiating the FBP
mod = proph(interval_width=.95, daily_seasonality=True)
# Fitting the model
mod.fit(ts)
# Making future Dataframes
future = mod.make_future_dataframe(periods=pers, freq='D')
# Forecasting
forecast = mod.predict(future)
# Plotting the forecast
mod.plot(forecast, uncertainty=True)
plt.xlabel("Dates")
plt.ylabel("Price")
plt.xlim(['2018-10', None])
st.pyplot()
def get_ROI_FBprophet(ts,
periods,
plot=False,
plot_components=False,
plot_title=None):
'''
Get forecasted ROI based on number of periods from end of ts.
'''
df = pd.DataFrame(ts)
df.reset_index(inplace=True)
df.columns = ['ds', 'y']
model = proph(interval_width=0.95)
model.fit(df)
future_dates = model.make_future_dataframe(periods=periods, freq='MS')
forecast = model.predict(future_dates)
pred_end_val = forecast[-1:]['yhat'].values[0]
start_val = df[-1:]['y'].values[0]
roi = (pred_end_val - start_val) / start_val
if plot:
model.plot(forecast, uncertainty=True)
plt.title(plot_title)
plt.show()
if plot_components:
model.plot_components(forecast)
plt.title(plot_title)
plt.show()
return roi
def forecast(date=None, showPlot = False, histPeriodDays=30, periodsAhead=100):
date = date.strftime('%Y-%m-%d %H:%M:%S')
start = time.time()
# Get data from binance
ts = getBinanceData.getData('BTCUSDT', interval=getBinanceData.Client.KLINE_INTERVAL_4HOUR, minutes= histPeriodDays * 24 * 60)
# print(ts.tail())
# Resetting the index back so Dates are no longer indexed
ts.reset_index(inplace=True)
# Renaming the columns for use in FB prophet
ts.rename(columns={'DateTime': 'ds', 'close': 'y'}, inplace=True)
# Fitting and training
mod = proph(interval_width=0.95,daily_seasonality=True,)
#region:add daily
# m = Prophet(weekly_seasonality=False)
mod.fit(ts)
#endregion
# Setting up predictions to be made
future = mod.make_future_dataframe(periods=periodsAhead, freq='4H')
future.tail()
# Making predictions
forecast = mod.predict(future)
forecast.to_csv('forecast.csv')
end = time.time()
print('Time taken: %.2f s' % (end - start))
# Plotting the model
if showPlot:
mod.plot(forecast, uncertainty=True)
plt.title('Facebook Prophet Forecast and Fitting')
plt.show()
selected = forecast.ds == date
selected = forecast[selected]
upper = selected.yhat_upper.values[0]
lower = selected.yhat_lower.values[0]
yhat = selected.yhat.values[0]
return yhat, upper, lower
def ProphetETH(self):
conn = self._getConnection()
df = pd.read_sql('select * from eth1', conn)
#df.to_csv("/home/student/Pied Piper/ethfinal101.csv")
conn.close()
lst = ['Open', 'High', 'Low', 'Close']
#df.drop(df.index[0],inplace=True)
df.insert(0, 'index', np.arange(1, 1607), True)
df.columns = [
'index', 'Date', 'Open', 'High', 'Low', 'Close', 'Volume'
]
result_ETH = []
for i in lst:
df_ETH = pd.DataFrame(df)
df_ETH.rename(columns={'Date': 'ds', i: 'y'}, inplace=True)
mod_ETH = proph(interval_width=0.95)
mod_ETH.fit(df_ETH)
future_ETH = mod_ETH.make_future_dataframe(periods=360, freq='D')
forecast_ETH = mod_ETH.predict(future_ETH)
result_ETH.append(forecast_ETH)
for jB in result_ETH:
df_ETH.predict = pd.DataFrame(jB)
df_ETH.predict.to_csv(str(i) + 'ETH' + '.csv')
mod_ETH.plot(forecast_ETH,
uncertainty=True,
xlabel='Date',
ylabel=i).savefig('/home/student/Pied Piper/static/' +
str(i) + 'ETH' + '.png')
mod_ETH.plot_components(
forecast_ETH,
uncertainty=True).savefig('/home/student/Pied Piper/static/' +
str(i) + 'ETHComponent' + '.png')
plt.title('Facebook Prophet Forecast and Fitting' +
(str(i) + 'ETH'))
plt.suptitle(str(i) + ' ' + 'ETH', fontsize=20)
def prophet_forecast(df, zipcodes_list):
""" Function that when inputed a dataframe and a list of zipcodes, retrieves a dictionary containing each
zipcode as a key and the forecasted values from the Prophet model associated with that zipcode as values.
"""
forecasts = {}
for zipcode in zipcodes_list:
returns = df.loc[(df['RegionName'] == zipcode)][['time', 'value']]
returns = returns.rename(columns={'time': 'ds','value': 'y'})
Model = proph(interval_width=0.95)
Model.fit(returns)
future_dates = Model.make_future_dataframe(periods=36, freq='MS')
forecast = Model.predict(future_dates)
forecasts[zipcode] = forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']]
return forecasts
