Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,6 +4,7 @@ import numpy as np
|
|
| 4 |
import matplotlib.pyplot as plt
|
| 5 |
import yfinance as yf
|
| 6 |
import joblib
|
|
|
|
| 7 |
from tensorflow.keras.models import Sequential
|
| 8 |
from tensorflow.keras.layers import LSTM, Dense, Dropout
|
| 9 |
from sklearn.metrics import mean_squared_error, mean_absolute_percentage_error
|
|
@@ -12,11 +13,11 @@ import warnings
|
|
| 12 |
|
| 13 |
warnings.filterwarnings("ignore")
|
| 14 |
|
| 15 |
-
# Utility functions
|
| 16 |
def rmse(a, b):
|
| 17 |
return math.sqrt(mean_squared_error(a, b))
|
| 18 |
|
| 19 |
-
# Download stock data
|
| 20 |
def download_data(ticker, start="2015-01-01", end=None):
|
| 21 |
df = yf.download(ticker, start=start, end=end, progress=False)
|
| 22 |
df = df[['Close']].dropna()
|
|
@@ -24,7 +25,7 @@ def download_data(ticker, start="2015-01-01", end=None):
|
|
| 24 |
df.rename(columns={'Close': 'close'}, inplace=True)
|
| 25 |
return df
|
| 26 |
|
| 27 |
-
#
|
| 28 |
def build_lstm(window_size=60):
|
| 29 |
model = Sequential([
|
| 30 |
LSTM(64, return_sequences=True, input_shape=(window_size,1)),
|
|
@@ -36,131 +37,48 @@ def build_lstm(window_size=60):
|
|
| 36 |
model.compile(optimizer="adam", loss="mse")
|
| 37 |
return model
|
| 38 |
|
| 39 |
-
# Forecast
|
| 40 |
-
def forecast_models(ticker="AAPL", test_len=180):
|
| 41 |
-
df = download_data(ticker)
|
| 42 |
-
series = df['close']
|
| 43 |
-
train, test = series[:-test_len], series[-test_len:]
|
| 44 |
-
|
| 45 |
-
results = {}
|
| 46 |
-
|
| 47 |
-
# ====== ARIMA ======
|
| 48 |
-
try:
|
| 49 |
-
arima_model = joblib.load("arima_model.pkl")
|
| 50 |
-
arima_forecast = arima_model.forecast(steps=len(test))
|
| 51 |
-
arima_forecast = pd.Series(arima_forecast, index=test.index)
|
| 52 |
-
results['ARIMA'] = {
|
| 53 |
-
"forecast": arima_forecast,
|
| 54 |
-
"RMSE": rmse(test, arima_forecast),
|
| 55 |
-
"MAPE": mean_absolute_percentage_error(test, arima_forecast)
|
| 56 |
-
}
|
| 57 |
-
except Exception as e:
|
| 58 |
-
results['ARIMA'] = {"error": str(e)}
|
| 59 |
-
|
| 60 |
-
# ====== Prophet ======
|
| 61 |
-
try:
|
| 62 |
-
prophet_model = joblib.load("prophet_model.pkl")
|
| 63 |
-
future = prophet_model.make_future_dataframe(periods=len(test), freq='D')
|
| 64 |
-
forecast = prophet_model.predict(future)
|
| 65 |
-
prophet_forecast = forecast.set_index('ds').reindex(test.index)['yhat'].fillna(method='ffill')
|
| 66 |
-
results['Prophet'] = {
|
| 67 |
-
"forecast": prophet_forecast,
|
| 68 |
-
"RMSE": rmse(test, prophet_forecast),
|
| 69 |
-
"MAPE": mean_absolute_percentage_error(test, prophet_forecast)
|
| 70 |
-
}
|
| 71 |
-
except Exception as e:
|
| 72 |
-
results['Prophet'] = {"error": str(e)}
|
| 73 |
-
|
| 74 |
-
# ====== LSTM (weights only) ======
|
| 75 |
-
# try:
|
| 76 |
-
# scaler = joblib.load("scaler.pkl")
|
| 77 |
-
# lstm_model = build_lstm(window_size=60)
|
| 78 |
-
# lstm_model.load_weights("lstm_model.weights.h5") # ✅ weights-only loading
|
| 79 |
-
|
| 80 |
-
# scaled_series = scaler.transform(series.values.reshape(-1,1))
|
| 81 |
-
|
| 82 |
-
# window_size = 60
|
| 83 |
-
# X = []
|
| 84 |
-
# for i in range(window_size, len(scaled_series)):
|
| 85 |
-
# X.append(scaled_series[i-window_size:i,0])
|
| 86 |
-
# X = np.array(X).reshape((-1, window_size, 1))
|
| 87 |
-
|
| 88 |
-
# split_index = len(train) - window_size
|
| 89 |
-
# X_test = X[split_index:]
|
| 90 |
-
|
| 91 |
-
# pred_scaled = lstm_model.predict(X_test, verbose=0)
|
| 92 |
-
# pred = scaler.inverse_transform(pred_scaled)
|
| 93 |
-
# lstm_forecast = pd.Series(pred.flatten(), index=test.index[:len(pred)])
|
| 94 |
-
|
| 95 |
-
# results['LSTM'] = {
|
| 96 |
-
# "forecast": lstm_forecast,
|
| 97 |
-
# "RMSE": rmse(test[:len(pred)], lstm_forecast),
|
| 98 |
-
# "MAPE": mean_absolute_percentage_error(test[:len(pred)], lstm_forecast)
|
| 99 |
-
# }
|
| 100 |
-
# except Exception as e:
|
| 101 |
-
# results['LSTM'] = {"error": str(e)}
|
| 102 |
-
|
| 103 |
-
# ===== Plot comparison =====
|
| 104 |
-
plt.figure(figsize=(12,5))
|
| 105 |
-
plt.plot(train.index[-400:], train[-400:], label="Train")
|
| 106 |
-
plt.plot(test.index, test, label="Test")
|
| 107 |
-
for model_name, res in results.items():
|
| 108 |
-
if "forecast" in res:
|
| 109 |
-
plt.plot(res["forecast"].index, res["forecast"], label=f"{model_name} forecast")
|
| 110 |
-
plt.legend()
|
| 111 |
-
plt.title(f"{ticker} Stock Forecasts")
|
| 112 |
-
plt.xlabel("Date"); plt.ylabel("Price")
|
| 113 |
-
plt.tight_layout()
|
| 114 |
-
plot_path = "forecast_plot.png"
|
| 115 |
-
plt.savefig(plot_path)
|
| 116 |
-
plt.close()
|
| 117 |
-
|
| 118 |
-
# ===== Metrics table =====
|
| 119 |
-
metrics = {}
|
| 120 |
-
for model_name, res in results.items():
|
| 121 |
-
if "RMSE" in res:
|
| 122 |
-
metrics[model_name] = {"RMSE": round(res["RMSE"],2), "MAPE": round(res["MAPE"],4)}
|
| 123 |
-
else:
|
| 124 |
-
metrics[model_name] = {"Error": res["error"]}
|
| 125 |
-
|
| 126 |
-
metrics_df = pd.DataFrame(metrics).T
|
| 127 |
-
|
| 128 |
-
return plot_path, metrics_df
|
| 129 |
-
|
| 130 |
-
# ===== Gradio UI =====
|
| 131 |
def run_app(ticker):
|
| 132 |
stock_data = yf.download(ticker, start="2015-01-01", end="2023-01-01")
|
| 133 |
stock_data = stock_data[['Close']].dropna()
|
| 134 |
|
| 135 |
results = []
|
| 136 |
|
| 137 |
-
# ARIMA
|
| 138 |
-
|
| 139 |
-
arima_model
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
|
|
|
|
|
|
|
|
|
| 144 |
|
| 145 |
-
# Prophet
|
| 146 |
-
|
| 147 |
-
prophet_model
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
|
| 153 |
-
|
|
|
|
|
|
|
|
|
|
| 154 |
|
| 155 |
-
#
|
| 156 |
df_results = pd.DataFrame(results, columns=["Model", "RMSE", "MAPE"])
|
| 157 |
-
df_results = df_results.dropna() # remove any
|
| 158 |
|
| 159 |
-
# Plot
|
| 160 |
plt.figure(figsize=(10, 5))
|
| 161 |
plt.plot(stock_data.index, stock_data['Close'], label="Actual")
|
| 162 |
-
|
| 163 |
-
|
|
|
|
|
|
|
| 164 |
plt.legend()
|
| 165 |
plt.title(f"{ticker} - Stock Price Forecasting")
|
| 166 |
plt.xlabel("Date")
|
|
@@ -171,5 +89,17 @@ def run_app(ticker):
|
|
| 171 |
return "forecast_plot.png", df_results
|
| 172 |
|
| 173 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 174 |
if __name__ == "__main__":
|
| 175 |
demo.launch()
|
|
|
|
| 4 |
import matplotlib.pyplot as plt
|
| 5 |
import yfinance as yf
|
| 6 |
import joblib
|
| 7 |
+
import pickle
|
| 8 |
from tensorflow.keras.models import Sequential
|
| 9 |
from tensorflow.keras.layers import LSTM, Dense, Dropout
|
| 10 |
from sklearn.metrics import mean_squared_error, mean_absolute_percentage_error
|
|
|
|
| 13 |
|
| 14 |
warnings.filterwarnings("ignore")
|
| 15 |
|
| 16 |
+
# ===== Utility functions =====
|
| 17 |
def rmse(a, b):
|
| 18 |
return math.sqrt(mean_squared_error(a, b))
|
| 19 |
|
| 20 |
+
# ===== Download stock data =====
|
| 21 |
def download_data(ticker, start="2015-01-01", end=None):
|
| 22 |
df = yf.download(ticker, start=start, end=end, progress=False)
|
| 23 |
df = df[['Close']].dropna()
|
|
|
|
| 25 |
df.rename(columns={'Close': 'close'}, inplace=True)
|
| 26 |
return df
|
| 27 |
|
| 28 |
+
# ===== LSTM architecture =====
|
| 29 |
def build_lstm(window_size=60):
|
| 30 |
model = Sequential([
|
| 31 |
LSTM(64, return_sequences=True, input_shape=(window_size,1)),
|
|
|
|
| 37 |
model.compile(optimizer="adam", loss="mse")
|
| 38 |
return model
|
| 39 |
|
| 40 |
+
# ===== Forecast function =====
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
def run_app(ticker):
|
| 42 |
stock_data = yf.download(ticker, start="2015-01-01", end="2023-01-01")
|
| 43 |
stock_data = stock_data[['Close']].dropna()
|
| 44 |
|
| 45 |
results = []
|
| 46 |
|
| 47 |
+
# ===== ARIMA =====
|
| 48 |
+
try:
|
| 49 |
+
with open("arima_model.pkl", "rb") as f:
|
| 50 |
+
arima_model = pickle.load(f)
|
| 51 |
+
arima_forecast = arima_model.forecast(steps=30)
|
| 52 |
+
arima_rmse = np.sqrt(mean_squared_error(stock_data['Close'][-30:], arima_forecast))
|
| 53 |
+
arima_mape = mean_absolute_percentage_error(stock_data['Close'][-30:], arima_forecast)
|
| 54 |
+
results.append(["ARIMA", arima_rmse, arima_mape])
|
| 55 |
+
except Exception as e:
|
| 56 |
+
results.append(["ARIMA", None, f"Error: {str(e)}"])
|
| 57 |
|
| 58 |
+
# ===== Prophet =====
|
| 59 |
+
try:
|
| 60 |
+
with open("prophet_model.pkl", "rb") as f:
|
| 61 |
+
prophet_model = pickle.load(f)
|
| 62 |
+
prophet_future = prophet_model.make_future_dataframe(periods=30)
|
| 63 |
+
prophet_forecast = prophet_model.predict(prophet_future)
|
| 64 |
+
prophet_forecast = prophet_forecast.set_index("ds").loc[stock_data.index[-30:]]
|
| 65 |
+
prophet_rmse = np.sqrt(mean_squared_error(stock_data['Close'][-30:], prophet_forecast['yhat']))
|
| 66 |
+
prophet_mape = mean_absolute_percentage_error(stock_data['Close'][-30:], prophet_forecast['yhat'])
|
| 67 |
+
results.append(["Prophet", prophet_rmse, prophet_mape])
|
| 68 |
+
except Exception as e:
|
| 69 |
+
results.append(["Prophet", None, f"Error: {str(e)}"])
|
| 70 |
|
| 71 |
+
# ===== Results Table =====
|
| 72 |
df_results = pd.DataFrame(results, columns=["Model", "RMSE", "MAPE"])
|
| 73 |
+
df_results = df_results.dropna(how="all") # remove any empty rows
|
| 74 |
|
| 75 |
+
# ===== Plot =====
|
| 76 |
plt.figure(figsize=(10, 5))
|
| 77 |
plt.plot(stock_data.index, stock_data['Close'], label="Actual")
|
| 78 |
+
if "arima_forecast" in locals():
|
| 79 |
+
plt.plot(stock_data.index[-30:], arima_forecast, label="ARIMA Forecast")
|
| 80 |
+
if "prophet_forecast" in locals():
|
| 81 |
+
plt.plot(prophet_forecast.index, prophet_forecast['yhat'], label="Prophet Forecast")
|
| 82 |
plt.legend()
|
| 83 |
plt.title(f"{ticker} - Stock Price Forecasting")
|
| 84 |
plt.xlabel("Date")
|
|
|
|
| 89 |
return "forecast_plot.png", df_results
|
| 90 |
|
| 91 |
|
| 92 |
+
# ===== Gradio Interface =====
|
| 93 |
+
demo = gr.Interface(
|
| 94 |
+
fn=run_app,
|
| 95 |
+
inputs=gr.Textbox(label="Stock Ticker", placeholder="Enter ticker (e.g. AAPL)"),
|
| 96 |
+
outputs=[
|
| 97 |
+
gr.Image(type="filepath", label="Forecast Plot"),
|
| 98 |
+
gr.Dataframe(label="Performance Metrics")
|
| 99 |
+
],
|
| 100 |
+
title="📈 Stock Price Forecasting (ARIMA & Prophet)",
|
| 101 |
+
description="This app compares ARIMA and Prophet models on stock closing prices using RMSE & MAPE."
|
| 102 |
+
)
|
| 103 |
+
|
| 104 |
if __name__ == "__main__":
|
| 105 |
demo.launch()
|