Keep training

recommender.py

@@ -5,6 +5,15 @@ import numpy as np
 import pandas as pd
 
 
+def add_row_train(df, list_tid):
+    new_pid_add = df.iloc[-1].name +1
+    list_tid_add = list_tid
+    list_pos_add = list(range(len(list_tid_add)))
+
+    df.loc[new_pid_add] = {'tid': list_tid_add,'pos': list_pos_add}
+    return df
+
+
 def inference_row(list_tid, ps_matrix):
     ps_matrix_norm = pp.normalize(ps_matrix, axis=1)
     length_tid = len(list_tid)
@@ -17,6 +26,9 @@ def inference_row(list_tid, ps_matrix):
 
 def get_best_tid(current_list, ps_matrix_row, K=50, MAX_tid=10):
     df_ps_train = pd.read_hdf('model/df_ps_train_new.hdf')
+    df_ps_train_extra = pd.read_hdf('model/df_ps_train_extra.hdf')
+    df_ps_train = pd.concat([df_ps_train,df_ps_train_extra])
+    
     sim_vector, sparse_row = inference_row(current_list, ps_matrix_row)
     sim_vector = sim_vector.toarray()[0].tolist()
 
@@ -55,7 +67,9 @@ def get_best_tid(current_list, ps_matrix_row, K=50, MAX_tid=10):
         if (n == K):
             break
 
-    return new_list
+    df_ps_train_extra = add_row_train(df_ps_train_extra, current_list)
+    df_ps_train_extra.to_hdf('model/df_ps_train_extra.hdf', key='abc')
+    return new_list, sparse_row
 
 
 def inference_from_tid(list_tid, K=50, MAX_tid=10):
@@ -64,10 +78,19 @@ def inference_from_tid(list_tid, K=50, MAX_tid=10):
 
     with open(pickle_path, 'rb') as f:
         ps_matrix = pickle.load(f)
-
-    ps_matrix_row = ps_matrix.tocsr()
-
-    return get_best_tid(list_tid, ps_matrix.tocsr(), K, MAX_tid)
+        
+    with open("model/giantMatrix_extra.pickle",'rb') as f:
+        ps_matrix_extra = pickle.load(f)
+    
+    ps_matrix = vstack((ps_matrix,ps_matrix_extra))
+
+    result, sparse_row = get_best_tid(list_tid, ps_matrix.tocsr(), K, MAX_tid)
+    ps_matrix_extra = vstack((ps_matrix_extra,sparse_row.todok()))
+    
+    with open("model/giantMatrix_extra.pickle", 'wb') as f:
+        pickle.dump(ps_matrix_extra, f)
+    
+    return result
 
 
 def inference_from_uri(list_uri, K=50, MAX_tid=10):