Update pages/market_rent_estimation.py

pages/market_rent_estimation.py

@@ -21,14 +21,15 @@ def main():
     st.set_page_config(initial_sidebar_state="collapsed", layout="wide")
     tab1, tab2 = st.tabs(["Comps", "Combined estimation"])#, "Market distributions", "Rexy"])
 
-    distances_matrix = np.load('all_data_matrix.npy')
-    df_properties = pd.read_csv("all_data.csv", encoding='utf-8')
+    distances_matrix = np.load('all_data_with_market.npy')
+    df_properties = pd.read_csv("all_data_with_market.csv", encoding='utf-8')
 
 
     # update the values of date and price of the historical lease to today date and nan value for price
     row_index = (df_properties[df_properties['google_ola'] == st.session_state['user_select_value']].index)[0]
-    df_properties.loc[row_index, ["months_since", "rent_combined", "execution_date"]] = [0, -1, None]
-    
+    df_properties.loc[row_index, ["months_since", "rent_combined", "execution_date"]] = [0, 0, None]
+    first_row_coords = (df_properties.loc[row_index, 'lat'],
+                        df_properties.loc[row_index, 'long'])    
     
     # user's filter for property size is a range in percentage around picked property 
     min_size = (100 + st.session_state['min_property_size_perc']) / 100 * st.session_state['building_sf']
@@ -39,26 +40,17 @@ def main():
     lease_max_size = (100 + st.session_state['max_property_lease_size_perc']) / 100 * st.session_state['rented_sf']
 
     #apply all filters 
-    # 
-    mask = (df_properties['rented_sf'] >= lease_min_size) & (df_properties['rented_sf'] <= lease_max_size) & (df_properties['building_sf'] >= min_size) & (df_properties['building_sf'] <= max_size) & (df_properties['months_since'] <= st.session_state['months_back']) & (df_properties['market_costar'] <= st.session_state['market_val'])
+    mask = (df_properties['rented_sf'] >= lease_min_size) & (df_properties['rented_sf'] <= lease_max_size) & (df_properties['building_sf'] >= min_size) & (df_properties['building_sf'] <= max_size) & (df_properties['months_since'] <= st.session_state['months_back']) & (df_properties['market_costar'] == st.session_state['market_val'])
     df_properties_filtered = df_properties[mask]
     
     # create distances matrix for remaining rows
-    positions = np.nonzero(mask)[0]
-    distances_matrix_filtered = distances_matrix[np.ix_(positions, positions)]
+    filltered_rows_embeding = distances_matrix[mask]
+    selected_row_embeding = distances_matrix[row_index]
+    distances_to_properties = np.linalg.norm(filltered_rows_embeding - selected_row_embeding, axis=1)
+    df_properties_filtered["score_pred"] = distances_to_properties
+    reordered_df_properties_filtered = df_properties_filtered.sort_values("score_pred", ascending=False)
     
-    # row_index = df_properties_filtered.index.get_loc(st.session_state['user_select_value'])
     df_properties_filtered = df_properties_filtered.reset_index()
-    row_index = (df_properties_filtered[df_properties_filtered['google_ola'] == st.session_state['user_select_value']].index)[0]
-    distances = distances_matrix_filtered[row_index]
-    sorted_indices = np.argsort(distances)
-    sorted_distances = distances[sorted_indices]
-    sorted_indexes = df_properties_filtered.index[sorted_indices]
-    reordered_df_properties_filtered = df_properties_filtered.loc[sorted_indexes]
-
-    first_row_coords = (reordered_df_properties_filtered.loc[row_index, 'lat'],
-                        reordered_df_properties_filtered.loc[row_index, 'long'])
-
     reordered_df_properties_filtered['distance_from_first (km)'] = reordered_df_properties_filtered.apply(calculate_distance, axis=1)