Caching of LSTM model

app.py

@@ -68,7 +68,6 @@ if st.session_state.json_content:
     # Generating synthetic data
     rng = np.random.RandomState(42)
 
-    print(algorithm_box)
     selected_algorithm_index = options.index(algorithm_box)
     if selected_algorithm_index == 0:
         plotted_result = apply_isolation_forest(rng, 

lstm.py

@@ -1,11 +1,12 @@
-# Long Short-Term Memory system
+# Long Short-Term Memory
 
-from sklearn import svm
+import streamlit as st
 import numpy as np
-from sklearn.preprocessing import StandardScaler
 import tensorflow as tf
+from sklearn.preprocessing import StandardScaler
 
-def apply_lstm(all_data: np.ndarray) -> np.ndarray:
+@st.cache_resource
+def train_lstm_model(all_data: np.ndarray, epochs: int = 50, batch_size: int = 1):
     scaled_features = StandardScaler().fit_transform(all_data)
 
     time_steps = 1
@@ -13,28 +14,32 @@ def apply_lstm(all_data: np.ndarray) -> np.ndarray:
     features = scaled_features.shape[1]
     lstm_input = scaled_features.reshape(samples, time_steps, features)
 
-    # Define the model
+    # Define and compile the model
     model = tf.keras.Sequential([
         tf.keras.layers.LSTM(50, input_shape=(time_steps, features)),
         tf.keras.layers.Dropout(0.2),
         tf.keras.layers.Dense(features)
     ])
-
-    # Compile the model
     model.compile(optimizer='adam', loss='mae')
 
     # Fit model
-    history = model.fit(lstm_input, scaled_features, epochs=50, batch_size=1, verbose=1)
+    model.fit(lstm_input, scaled_features, epochs=epochs, batch_size=batch_size, verbose=1)
+    return model, scaled_features, lstm_input
+
+
+def apply_lstm(all_data: np.ndarray) -> np.ndarray:
+    # Train or retrieve cached model and scaled features
+    model, scaled_features, lstm_input = train_lstm_model(all_data)
 
     # Prediction and error calculation
     predictions = model.predict(lstm_input)
     mse = np.mean(np.power(scaled_features - predictions, 2), axis=1)
 
-    # Define a threshold for what you consider an outlier
+    # Threshold for outliers
     threshold = np.quantile(mse, 0.8)
 
-    # Outliers where mse is greater than the threshold
+    # Determine outliers
     outliers = mse > threshold
 
-    mapped_array = np.where(outliers, -1, 1)
-    return mapped_array
+    # Map to usable -1/1 format
+    return np.where(outliers, -1, 1)