Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -5,19 +5,17 @@ import matplotlib.pyplot as plt
|
|
| 5 |
import yfinance as yf
|
| 6 |
from sklearn.metrics import mean_squared_error, mean_absolute_percentage_error
|
| 7 |
import math, warnings
|
|
|
|
| 8 |
|
| 9 |
-
from
|
| 10 |
-
from prophet import Prophet
|
| 11 |
-
from sklearn.preprocessing import MinMaxScaler
|
| 12 |
-
from tensorflow.keras.models import Sequential
|
| 13 |
-
from tensorflow.keras.layers import LSTM, Dense, Dropout
|
| 14 |
|
| 15 |
warnings.filterwarnings("ignore")
|
| 16 |
|
| 17 |
-
# Utility
|
| 18 |
def rmse(a, b):
|
| 19 |
return math.sqrt(mean_squared_error(a, b))
|
| 20 |
|
|
|
|
| 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()
|
|
@@ -25,6 +23,7 @@ def download_data(ticker, start="2015-01-01", end=None):
|
|
| 25 |
df.rename(columns={'Close': 'close'}, inplace=True)
|
| 26 |
return df
|
| 27 |
|
|
|
|
| 28 |
def forecast_models(ticker="AAPL", test_len=180):
|
| 29 |
df = download_data(ticker)
|
| 30 |
series = df['close']
|
|
@@ -32,10 +31,9 @@ def forecast_models(ticker="AAPL", test_len=180):
|
|
| 32 |
|
| 33 |
results = {}
|
| 34 |
|
| 35 |
-
# =====
|
| 36 |
try:
|
| 37 |
-
|
| 38 |
-
arima_fit = arima_model.fit()
|
| 39 |
arima_forecast = arima_fit.forecast(steps=len(test))
|
| 40 |
arima_forecast = pd.Series(arima_forecast, index=test.index)
|
| 41 |
results['ARIMA'] = {
|
|
@@ -46,14 +44,11 @@ def forecast_models(ticker="AAPL", test_len=180):
|
|
| 46 |
except Exception as e:
|
| 47 |
results['ARIMA'] = {"error": str(e)}
|
| 48 |
|
| 49 |
-
# =====
|
| 50 |
try:
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
m.fit(prophet_df)
|
| 55 |
-
future = m.make_future_dataframe(periods=len(test), freq='D')
|
| 56 |
-
forecast = m.predict(future)
|
| 57 |
prophet_forecast = forecast.set_index('ds').reindex(test.index)['yhat'].fillna(method='ffill')
|
| 58 |
results['Prophet'] = {
|
| 59 |
"forecast": prophet_forecast,
|
|
@@ -63,37 +58,29 @@ def forecast_models(ticker="AAPL", test_len=180):
|
|
| 63 |
except Exception as e:
|
| 64 |
results['Prophet'] = {"error": str(e)}
|
| 65 |
|
| 66 |
-
# =====
|
| 67 |
try:
|
| 68 |
-
|
| 69 |
-
|
|
|
|
| 70 |
|
| 71 |
-
|
|
|
|
|
|
|
| 72 |
X, y = [], []
|
| 73 |
for i in range(window_size, len(data)):
|
| 74 |
-
X.append(data[i-window_size:i,
|
| 75 |
y.append(data[i,0])
|
| 76 |
X, y = np.array(X), np.array(y)
|
| 77 |
-
X = X.reshape((X.shape[0], X.shape[1],
|
| 78 |
return X, y
|
| 79 |
|
| 80 |
-
|
| 81 |
-
X_all, y_all = create_windows(scaled_series, window_size=window_size)
|
| 82 |
split_index = len(train) - window_size
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
LSTM(64, return_sequences=True, input_shape=(window_size,1)),
|
| 88 |
-
Dropout(0.2),
|
| 89 |
-
LSTM(32),
|
| 90 |
-
Dropout(0.2),
|
| 91 |
-
Dense(1)
|
| 92 |
-
])
|
| 93 |
-
model.compile(optimizer='adam', loss='mse')
|
| 94 |
-
model.fit(X_train, y_train, epochs=5, batch_size=32, verbose=0) # smaller epochs for demo
|
| 95 |
-
|
| 96 |
-
pred_scaled = model.predict(X_test, verbose=0)
|
| 97 |
pred = scaler.inverse_transform(pred_scaled)
|
| 98 |
lstm_forecast = pd.Series(pred.flatten(), index=test.index[:len(pred)])
|
| 99 |
|
|
@@ -132,8 +119,7 @@ def forecast_models(ticker="AAPL", test_len=180):
|
|
| 132 |
|
| 133 |
return plot_path, metrics_df
|
| 134 |
|
| 135 |
-
|
| 136 |
-
# Gradio UI
|
| 137 |
def run_app(ticker):
|
| 138 |
plot_path, metrics_df = forecast_models(ticker)
|
| 139 |
return plot_path, metrics_df
|
|
|
|
| 5 |
import yfinance as yf
|
| 6 |
from sklearn.metrics import mean_squared_error, mean_absolute_percentage_error
|
| 7 |
import math, warnings
|
| 8 |
+
import joblib
|
| 9 |
|
| 10 |
+
from tensorflow.keras.models import load_model
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
warnings.filterwarnings("ignore")
|
| 13 |
|
| 14 |
+
# Utility function
|
| 15 |
def rmse(a, b):
|
| 16 |
return math.sqrt(mean_squared_error(a, b))
|
| 17 |
|
| 18 |
+
# Download stock data
|
| 19 |
def download_data(ticker, start="2015-01-01", end=None):
|
| 20 |
df = yf.download(ticker, start=start, end=end, progress=False)
|
| 21 |
df = df[['Close']].dropna()
|
|
|
|
| 23 |
df.rename(columns={'Close': 'close'}, inplace=True)
|
| 24 |
return df
|
| 25 |
|
| 26 |
+
# Forecast using pre-saved models
|
| 27 |
def forecast_models(ticker="AAPL", test_len=180):
|
| 28 |
df = download_data(ticker)
|
| 29 |
series = df['close']
|
|
|
|
| 31 |
|
| 32 |
results = {}
|
| 33 |
|
| 34 |
+
# ===== Load ARIMA =====
|
| 35 |
try:
|
| 36 |
+
arima_fit = joblib.load("arima_model.pkl")
|
|
|
|
| 37 |
arima_forecast = arima_fit.forecast(steps=len(test))
|
| 38 |
arima_forecast = pd.Series(arima_forecast, index=test.index)
|
| 39 |
results['ARIMA'] = {
|
|
|
|
| 44 |
except Exception as e:
|
| 45 |
results['ARIMA'] = {"error": str(e)}
|
| 46 |
|
| 47 |
+
# ===== Load Prophet =====
|
| 48 |
try:
|
| 49 |
+
prophet_model = joblib.load("prophet_model.pkl")
|
| 50 |
+
future = prophet_model.make_future_dataframe(periods=len(test), freq='D')
|
| 51 |
+
forecast = prophet_model.predict(future)
|
|
|
|
|
|
|
|
|
|
| 52 |
prophet_forecast = forecast.set_index('ds').reindex(test.index)['yhat'].fillna(method='ffill')
|
| 53 |
results['Prophet'] = {
|
| 54 |
"forecast": prophet_forecast,
|
|
|
|
| 58 |
except Exception as e:
|
| 59 |
results['Prophet'] = {"error": str(e)}
|
| 60 |
|
| 61 |
+
# ===== Load LSTM =====
|
| 62 |
try:
|
| 63 |
+
lstm_model = load_model("lstm_model.h5")
|
| 64 |
+
scaler = joblib.load("scaler.pkl")
|
| 65 |
+
scaled_series = scaler.transform(series.values.reshape(-1,1))
|
| 66 |
|
| 67 |
+
# Create sliding windows
|
| 68 |
+
window_size = 60
|
| 69 |
+
def create_windows(data):
|
| 70 |
X, y = [], []
|
| 71 |
for i in range(window_size, len(data)):
|
| 72 |
+
X.append(data[i-window_size:i,0])
|
| 73 |
y.append(data[i,0])
|
| 74 |
X, y = np.array(X), np.array(y)
|
| 75 |
+
X = X.reshape((X.shape[0], X.shape[1],1))
|
| 76 |
return X, y
|
| 77 |
|
| 78 |
+
X_all, y_all = create_windows(scaled_series)
|
|
|
|
| 79 |
split_index = len(train) - window_size
|
| 80 |
+
X_test = X_all[split_index:]
|
| 81 |
+
y_test = y_all[split_index:]
|
| 82 |
+
|
| 83 |
+
pred_scaled = lstm_model.predict(X_test, verbose=0)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
pred = scaler.inverse_transform(pred_scaled)
|
| 85 |
lstm_forecast = pd.Series(pred.flatten(), index=test.index[:len(pred)])
|
| 86 |
|
|
|
|
| 119 |
|
| 120 |
return plot_path, metrics_df
|
| 121 |
|
| 122 |
+
# ===== Gradio UI =====
|
|
|
|
| 123 |
def run_app(ticker):
|
| 124 |
plot_path, metrics_df = forecast_models(ticker)
|
| 125 |
return plot_path, metrics_df
|