def forecast_markov(trainingdata, evaldata):
n = len(evaldata)
mod = mc.MarkovChain().from_data(trainingdata['count'])
ids, states = mod.simulate(n, start=10)
plot(f'markov_{n}.png', states, 'Forecast based on Markov Chains',
evaldata['count'].values)
print('Accuracy Scores for Markov Chains:')
accuracy.eval_model(states, 1, evaldata['count'].values)
def forecast_sarimax(trainingdata, evaldata):
n = len(evaldata)
periods_in_season = 24
mod = sm.tsa.statespace.SARIMAX(trainingdata['count'].astype(float), order=(0, 0, 0),
seasonal_order=(1, 0, 1, periods_in_season),
enforce_stationarity=False)
res = mod.fit(disp=False)
print(res.summary())
predictions = res.forecast(n)
print(predictions)
accuracy.eval_model(predictions.values, 1, evaldata['count'].values)
plot(f'sarimax/sarimax_s{periods_in_season}_{n}.png', predictions.values,
'Forecast based on the SARIMAX Model', evaldata['count'].values)
def forecast_linear_reg(trainingdata, evaldata):
n = len(evaldata)
to_timestamp_converter = lambda t: (t - np.datetime64('1970-01-01T00:00:00Z')) / np.timedelta64(1, 's')
x = np.array([to_timestamp_converter(t) for t in trainingdata.index.values]).reshape((-1, 1))
print(x)
y = np.array(trainingdata['count'])
model = LinearRegression().fit(x, y)
predictions = model.predict(x[:n])
plot(f'linear_reg_{n}.png', predictions,
'Forecast based on Linear Regression (OLS)',
evaldata['count'].values)
print('Summary')
print(f'Coefficient of Determination: {model.score(x, y)}')
print(f'Function: {model.coef_}t + {model.intercept_}')
print('Accuracy Scores for OLS:')
accuracy.eval_model(predictions, 1, evaldata['count'].values)
def forecast_neural(trainingdata, evaldata):
scaler = MinMaxScaler(feature_range=(0, 1))
dataset = scaler.fit_transform(np.array(trainingdata['count'].astype(float)).reshape(-1, 1))
validationset = scaler.fit_transform(np.array(evaldata['count'].astype(float)).reshape(-1, 1))
look_back = 20
trainX, trainY = create_dataset(dataset, look_back)
testX, testY = create_dataset(validationset, look_back)
trainX = np.reshape(trainX, (trainX.shape[0], 1, trainX.shape[1]))
try:
model = keras.models.load_model(f"lstm_network_{look_back}")
print('Successfully loaded model.')
except IOError:
print('Creating new model.')
model = Sequential()
model.add(LSTM(4, input_shape=(1, look_back)))
model.add(Dense(1))
model.compile(loss='mean_squared_error', optimizer='adam')
model.fit(trainX, trainY, epochs=100, batch_size=1, verbose=2)
model.save(f'lstm_network_{look_back}')
# make predictions
# trainPredict = model.predict(trainX)
# testPredict = model.predict(testX)
prediction_list = trainX[-1]
for _ in range(len(evaldata)):
x = prediction_list[-look_back:]
x = x.reshape((1, 1, look_back))
out = model.predict(x)[0][0]
prediction_list = np.append(prediction_list, out)
prediction_list = prediction_list[look_back:]
# invert predictions
print(np.shape(testX))
print(testX)
testPredict = scaler.inverse_transform(prediction_list.reshape(-1, 1))
testX = scaler.inverse_transform(testX)
accuracy.eval_model(testPredict[:,0], 1, evaldata['count'].values)
plot(f'lstm_l{look_back}_{len(testPredict[:,0])}.png', testPredict[:,0],
'Forecast based on a LSTM Neural Network', evaldata['count'].values)
def forecast_neural_timestamp(trainingdata, evaldata):
scaler = MinMaxScaler(feature_range=(0, 1))
trainY = scaler.fit_transform(np.array(trainingdata['count'].astype(float)).reshape(-1, 1))
trainX = scaler.fit_transform(np.array(trainingdata.index.values.astype(float)).reshape(-1, 1))
testY = scaler.fit_transform(np.array(evaldata['count'].astype(float)).reshape(-1, 1))
testX = scaler.fit_transform(np.array(trainingdata.index.values.astype(float)).reshape(-1, 1))
try:
model = keras.models.load_model(f"lstm_network_timestamp")
print('Successfully loaded model.')
except IOError:
print('Creating new model.')
model = Sequential()
model.add(LSTM(4, input_shape=(1, 1)))
model.add(Dense(1))
model.compile(loss='mean_squared_error', optimizer='adam')
model.fit(trainX, trainY, epochs=100, batch_size=1, verbose=2)
model.save(f'lstm_network_timestamp')
# make predictions
# trainPredict = model.predict(trainX)
# testPredict = model.predict(testX)
prediction_list = np.array()
for x in testX:
out = model.predict(x)[0][0]
print(out)
prediction_list = np.append(prediction_list, out)
# invert predictions
print(np.shape(testX))
print(testX)
testPredict = scaler.inverse_transform(prediction_list.reshape(-1, 1))
testX = scaler.inverse_transform(testX)
accuracy.eval_model(testPredict[:,0], 1, evaldata['count'].values)
plot(f'lstm_timestamp_{len(testPredict[:,0])}.png', testPredict[:,0],
'Forecast based on a LSTM Neural Network', evaldata['count'].values)