Update app.py
Browse files
app.py
CHANGED
@@ -86,38 +86,93 @@ def lstm_gru_forecast(data, model_type, steps):
|
|
86 |
def ensemble_forecast(predictions_list):
|
87 |
return pd.DataFrame(predictions_list).mean(axis=0)
|
88 |
|
89 |
-
#
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
107 |
|
108 |
-
#
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
st.
|
120 |
-
st.
|
121 |
-
st.
|
122 |
-
st.
|
123 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
def ensemble_forecast(predictions_list):
|
87 |
return pd.DataFrame(predictions_list).mean(axis=0)
|
88 |
|
89 |
+
# Function to fit ARIMA model and make predictions
|
90 |
+
def arima_forecast(data, p, d, q, steps):
|
91 |
+
# Differencing
|
92 |
+
for i in range(d):
|
93 |
+
data_diff = np.diff(data)
|
94 |
+
data = data_diff
|
95 |
+
|
96 |
+
# Autoregressive (AR) and Moving Average (MA) components
|
97 |
+
ar_coef = np.zeros(p) if p > 0 else []
|
98 |
+
ma_coef = np.zeros(q) if q > 0 else []
|
99 |
+
|
100 |
+
# Initial prediction
|
101 |
+
predictions = list(data[:p])
|
102 |
+
|
103 |
+
# ARIMA forecasting
|
104 |
+
for i in range(len(data) - p):
|
105 |
+
ar_term = sum(ar_coef[j] * data[i + p - j - 1] for j in range(p))
|
106 |
+
ma_term = sum(ma_coef[j] * (data[i + p - j - 1] - predictions[-1]) for j in range(q))
|
107 |
+
next_prediction = data[i + p] + ar_term + ma_term
|
108 |
+
predictions.append(next_prediction)
|
109 |
+
|
110 |
+
# Update coefficients using online learning (optional)
|
111 |
+
if i + p + 1 < len(data):
|
112 |
+
ar_coef = ar_coef + (2.0 / (i + p + 2)) * (data[i + p + 1] - next_prediction) * np.flip(data[i:i + p])
|
113 |
+
ma_coef = ma_coef + (2.0 / (i + p + 2)) * (data[i + p + 1] - next_prediction) * np.flip(predictions[i - q + 1:i + 1])
|
114 |
+
|
115 |
+
# Inverse differencing
|
116 |
+
for i in range(d):
|
117 |
+
predictions = np.cumsum([data[p - 1]] + predictions)
|
118 |
+
|
119 |
+
return predictions[-steps:]
|
120 |
|
121 |
+
# Streamlit App
|
122 |
+
def main():
|
123 |
+
st.title("Stock Price Forecasting App")
|
124 |
+
|
125 |
+
# Load stock data using Streamlit sidebar
|
126 |
+
symbol = st.sidebar.text_input("Enter Stock Symbol", value='AAPL')
|
127 |
+
start_date = st.sidebar.text_input("Enter Start Date", value='2021-01-01')
|
128 |
+
end_date = st.sidebar.text_input("Enter End Date", value='2022-01-01')
|
129 |
+
stock_prices = get_stock_data(symbol, start_date, end_date)
|
130 |
+
|
131 |
+
# ARIMA parameters using Streamlit sliders
|
132 |
+
p = st.sidebar.slider("AR Component (p)", min_value=0, max_value=10, value=3)
|
133 |
+
d = st.sidebar.slider("Differencing (d)", min_value=0, max_value=5, value=0)
|
134 |
+
q = st.sidebar.slider("MA Component (q)", min_value=0, max_value=10, value=0)
|
135 |
+
arima_forecast_steps = st.sidebar.slider("ARIMA Forecast Steps", min_value=1, max_value=100, value=30)
|
136 |
+
|
137 |
+
# LSTM and GRU parameters using Streamlit sliders
|
138 |
+
lstm_gru_forecast_steps = st.sidebar.slider("LSTM/GRU Forecast Steps", min_value=1, max_value=100, value=30)
|
139 |
+
|
140 |
+
# Custom ARIMA Forecast using Streamlit button
|
141 |
+
if st.sidebar.button("Run Custom ARIMA Forecast"):
|
142 |
+
arima_predictions_custom = arima_forecast(stock_prices.values, p, d, q, arima_forecast_steps)
|
143 |
+
arima_predictions_custom = pd.Series(arima_predictions_custom, index=pd.date_range(start=stock_prices.index[-1], periods=arima_forecast_steps + 1, freq=stock_prices.index.freq))
|
144 |
+
|
145 |
+
# Display ARIMA Forecast Plot
|
146 |
+
st.subheader("Custom ARIMA Forecast")
|
147 |
+
st.line_chart(pd.concat([stock_prices, arima_predictions_custom], axis=1).rename(columns={0: "ARIMA Forecast"}))
|
148 |
+
|
149 |
+
# LSTM Forecast using Streamlit button
|
150 |
+
if st.sidebar.button("Run LSTM Forecast"):
|
151 |
+
lstm_predictions = lstm_gru_forecast(stock_prices, 'lstm', lstm_gru_forecast_steps)
|
152 |
+
|
153 |
+
# Display LSTM Forecast Plot
|
154 |
+
st.subheader("LSTM Forecast")
|
155 |
+
st.line_chart(pd.concat([stock_prices, pd.Series(lstm_predictions, index=pd.date_range(start=stock_prices.index[-1], periods=lstm_gru_forecast_steps + 1, freq=stock_prices.index.freq))], axis=1).rename(columns={0: "LSTM Forecast"}))
|
156 |
+
|
157 |
+
# GRU Forecast using Streamlit button
|
158 |
+
if st.sidebar.button("Run GRU Forecast"):
|
159 |
+
gru_predictions = lstm_gru_forecast(stock_prices, 'gru', lstm_gru_forecast_steps)
|
160 |
+
|
161 |
+
# Display GRU Forecast Plot
|
162 |
+
st.subheader("GRU Forecast")
|
163 |
+
st.line_chart(pd.concat([stock_prices, pd.Series(gru_predictions, index=pd.date_range(start=stock_prices.index[-1], periods=lstm_gru_forecast_steps + 1, freq=stock_prices.index.freq))], axis=1).rename(columns={0: "GRU Forecast"}))
|
164 |
+
|
165 |
+
# Ensemble Forecast using Streamlit button
|
166 |
+
if st.sidebar.button("Run Ensemble Forecast"):
|
167 |
+
ensemble_predictions = ensemble_forecast([arima_predictions_custom, lstm_predictions, gru_predictions])
|
168 |
+
|
169 |
+
# Display Ensemble Forecast Plot
|
170 |
+
st.subheader("Ensemble Forecast")
|
171 |
+
st.line_chart(pd.concat([stock_prices, ensemble_predictions], axis=1).rename(columns={0: "Ensemble Forecast"}))
|
172 |
+
|
173 |
+
# Plotting Historical Stock Prices
|
174 |
+
st.subheader("Historical Stock Prices")
|
175 |
+
st.line_chart(stock_prices)
|
176 |
+
|
177 |
+
if __name__ == "__main__":
|
178 |
+
main()
|