try cross features for tractate

src/main_portrait.py

@@ -34,7 +34,7 @@ def user_portrait_scan_info(device_dict, diary_dict, tractate_dict, diary_predic
                         all_count += 1
                         print(str(all_count) + ": " + device_id)
                         offline_predict_diary(device_id, device_dict, diary_dict, diary_predict_fn, size=predict_size)
-                        print("---------------------")
+                        print("-------------------")
                         offline_predict_tractate(device_id, device_dict, tractate_dict, tractate_predict_fn, size=predict_size)
                         print("=========================================\n")
 
@@ -61,7 +61,7 @@ def user_portrait_increment_scan_info(device_dict, diary_dict, tractate_dict, di
                         all_count += 1
                         print(str(all_count) + ": " + device_id)
                         offline_predict_diary(device_id, device_dict, diary_dict, diary_predict_fn, size=200)
-                        print("---------------------")
+                        print("-------------------")
                         offline_predict_tractate(device_id, device_dict, tractate_dict, tractate_predict_fn, size=200)
                         print("=========================================\n")
 

src/models/esmm/diary_model.py

@@ -175,20 +175,20 @@ _categorical_columns = [
     "click_diary_id4",
     "click_diary_id5",
     "service_city",
-    # "device_fd2",
-    # "device_sd2",
-    # "device_fs2",
-    # "device_ss2",
-    # "device_fp2",
-    # "device_sp2",
-    # "device_p2",
-    # "device_fd3",
-    # "device_sd3",
-    # "device_fs3",
-    # "device_ss3",
-    # "device_fp3",
-    # "device_sp3",
-    # "device_p3",
+    "device_fd2",
+    "device_sd2",
+    "device_fs2",
+    "device_ss2",
+    "device_fp2",
+    "device_sp2",
+    "device_p2",
+    "device_fd3",
+    "device_sd3",
+    "device_fs3",
+    "device_ss3",
+    "device_fp3",
+    "device_sp3",
+    "device_p3",
 ]
 
 PREDICTION_ALL_COLUMNS = _int_columns + _float_columns + _categorical_columns

src/models/esmm/fe/diary_fe.py

@@ -303,20 +303,20 @@ CATEGORICAL_COLUMNS = [
     "click_diary_id5",
     "is_related_service",
     "service_city",
-    # "device_fd2",
-    # "device_sd2",
-    # "device_fs2",
-    # "device_ss2",
-    # "device_fp2",
-    # "device_sp2",
-    # "device_p2",
-    # "device_fd3",
-    # "device_sd3",
-    # "device_fs3",
-    # "device_ss3",
-    # "device_fp3",
-    # "device_sp3",
-    # "device_p3",
+    "device_fd2",
+    "device_sd2",
+    "device_fs2",
+    "device_ss2",
+    "device_fp2",
+    "device_sp2",
+    "device_p2",
+    "device_fd3",
+    "device_sd3",
+    "device_fs3",
+    "device_ss3",
+    "device_fp3",
+    "device_sp3",
+    "device_p3",
 ]
 CROSS_COLUMNS = [
     ["device_fd", "content_fd"],
@@ -326,20 +326,20 @@ CROSS_COLUMNS = [
     ["device_fp", "content_fp"],
     ["device_sp", "content_sp"],
     ["device_p", "content_p"],
-    # ["device_fd2", "content_fd"],
-    # ["device_sd2", "content_sd"],
-    # ["device_fs2", "content_fs"],
-    # ["device_ss2", "content_ss"],
-    # ["device_fp2", "content_fp"],
-    # ["device_sp2", "content_sp"],
-    # ["device_p2", "content_p"],
-    # ["device_fd3", "content_fd"],
-    # ["device_sd3", "content_sd"],
-    # ["device_fs3", "content_fs"],
-    # ["device_ss3", "content_ss"],
-    # ["device_fp3", "content_fp"],
-    # ["device_sp3", "content_sp"],
-    # ["device_p3", "content_p"],
+    ["device_fd2", "content_fd"],
+    ["device_sd2", "content_sd"],
+    ["device_fs2", "content_fs"],
+    ["device_ss2", "content_ss"],
+    ["device_fp2", "content_fp"],
+    ["device_sp2", "content_sp"],
+    ["device_p2", "content_p"],
+    ["device_fd3", "content_fd"],
+    ["device_sd3", "content_sd"],
+    ["device_fs3", "content_fs"],
+    ["device_ss3", "content_ss"],
+    ["device_fp3", "content_fp"],
+    ["device_sp3", "content_sp"],
+    ["device_p3", "content_p"],
 ]
 
 
@@ -448,21 +448,21 @@ def join_features(device_df, diary_df, cc_df):
     df["device_sp"] = df["second_positions_x"].apply(lambda x: nth_element(x, 0))
     df["device_p"] = df["projects_x"].apply(lambda x: nth_element(x, 0))
 
-    # df["device_fd2"] = df["first_demands_x"].apply(lambda x: nth_element(x, 1))
-    # df["device_sd2"] = df["second_demands_x"].apply(lambda x: nth_element(x, 1))
-    # df["device_fs2"] = df["first_solutions_x"].apply(lambda x: nth_element(x, 1))
-    # df["device_ss2"] = df["second_solutions_x"].apply(lambda x: nth_element(x, 1))
-    # df["device_fp2"] = df["first_positions_x"].apply(lambda x: nth_element(x, 1))
-    # df["device_sp2"] = df["second_positions_x"].apply(lambda x: nth_element(x, 1))
-    # df["device_p2"] = df["projects_x"].apply(lambda x: nth_element(x, 1))
+    df["device_fd2"] = df["first_demands_x"].apply(lambda x: nth_element(x, 1))
+    df["device_sd2"] = df["second_demands_x"].apply(lambda x: nth_element(x, 1))
+    df["device_fs2"] = df["first_solutions_x"].apply(lambda x: nth_element(x, 1))
+    df["device_ss2"] = df["second_solutions_x"].apply(lambda x: nth_element(x, 1))
+    df["device_fp2"] = df["first_positions_x"].apply(lambda x: nth_element(x, 1))
+    df["device_sp2"] = df["second_positions_x"].apply(lambda x: nth_element(x, 1))
+    df["device_p2"] = df["projects_x"].apply(lambda x: nth_element(x, 1))
 
-    # df["device_fd3"] = df["first_demands_x"].apply(lambda x: nth_element(x, 2))
-    # df["device_sd3"] = df["second_demands_x"].apply(lambda x: nth_element(x, 2))
-    # df["device_fs3"] = df["first_solutions_x"].apply(lambda x: nth_element(x, 2))
-    # df["device_ss3"] = df["second_solutions_x"].apply(lambda x: nth_element(x, 2))
-    # df["device_fp3"] = df["first_positions_x"].apply(lambda x: nth_element(x, 2))
-    # df["device_sp3"] = df["second_positions_x"].apply(lambda x: nth_element(x, 2))
-    # df["device_p3"] = df["projects_x"].apply(lambda x: nth_element(x, 2))
+    df["device_fd3"] = df["first_demands_x"].apply(lambda x: nth_element(x, 2))
+    df["device_sd3"] = df["second_demands_x"].apply(lambda x: nth_element(x, 2))
+    df["device_fs3"] = df["first_solutions_x"].apply(lambda x: nth_element(x, 2))
+    df["device_ss3"] = df["second_solutions_x"].apply(lambda x: nth_element(x, 2))
+    df["device_fp3"] = df["first_positions_x"].apply(lambda x: nth_element(x, 2))
+    df["device_sp3"] = df["second_positions_x"].apply(lambda x: nth_element(x, 2))
+    df["device_p3"] = df["projects_x"].apply(lambda x: nth_element(x, 2))
 
     df["content_fd"] = df["first_demands_y"].apply(lambda x: nth_element(x, 0))
     df["content_sd"] = df["second_demands_y"].apply(lambda x: nth_element(x, 0))
@@ -551,20 +551,20 @@ def device_diary_fe(device_id, diary_ids, device_dict, diary_dict):
     device_info["device_fp"] = nth_element(device_fp, 0)
     device_info["device_sp"] = nth_element(device_sp, 0)
     device_info["device_p"] = nth_element(device_p, 0)
-    # device_info["device_fd2"] = nth_element(device_fd, 1)
-    # device_info["device_sd2"] = nth_element(device_sd, 1)
-    # device_info["device_fs2"] = nth_element(device_fs, 1)
-    # device_info["device_ss2"] = nth_element(device_ss, 1)
-    # device_info["device_fp2"] = nth_element(device_fp, 1)
-    # device_info["device_sp2"] = nth_element(device_sp, 1)
-    # device_info["device_p2"] = nth_element(device_p, 1)
-    # device_info["device_fd3"] = nth_element(device_fd, 2)
-    # device_info["device_sd3"] = nth_element(device_sd, 2)
-    # device_info["device_fs3"] = nth_element(device_fs, 2)
-    # device_info["device_ss3"] = nth_element(device_ss, 2)
-    # device_info["device_fp3"] = nth_element(device_fp, 2)
-    # device_info["device_sp3"] = nth_element(device_sp, 2)
-    # device_info["device_p3"] = nth_element(device_p, 2)
+    device_info["device_fd2"] = nth_element(device_fd, 1)
+    device_info["device_sd2"] = nth_element(device_sd, 1)
+    device_info["device_fs2"] = nth_element(device_fs, 1)
+    device_info["device_ss2"] = nth_element(device_ss, 1)
+    device_info["device_fp2"] = nth_element(device_fp, 1)
+    device_info["device_sp2"] = nth_element(device_sp, 1)
+    device_info["device_p2"] = nth_element(device_p, 1)
+    device_info["device_fd3"] = nth_element(device_fd, 2)
+    device_info["device_sd3"] = nth_element(device_sd, 2)
+    device_info["device_fs3"] = nth_element(device_fs, 2)
+    device_info["device_ss3"] = nth_element(device_ss, 2)
+    device_info["device_fp3"] = nth_element(device_fp, 2)
+    device_info["device_sp3"] = nth_element(device_sp, 2)
+    device_info["device_p3"] = nth_element(device_p, 2)
     diary_lst = []
     diary_ids_res = []
     for id in diary_ids:

src/models/esmm/fe/tractate_fe.py

@@ -288,6 +288,20 @@ CATEGORICAL_COLUMNS = [
     "click_tractate_id5",
     "is_related_service",
     "service_city",
+    "device_fd2",
+    "device_sd2",
+    "device_fs2",
+    "device_ss2",
+    "device_fp2",
+    "device_sp2",
+    "device_p2",
+    "device_fd3",
+    "device_sd3",
+    "device_fs3",
+    "device_ss3",
+    "device_fp3",
+    "device_sp3",
+    "device_p3",
 ]
 CROSS_COLUMNS = [
     ["device_fd", "content_fd"],
@@ -297,6 +311,20 @@ CROSS_COLUMNS = [
     ["device_fp", "content_fp"],
     ["device_sp", "content_sp"],
     ["device_p", "content_p"],
+    ["device_fd2", "content_fd"],
+    ["device_sd2", "content_sd"],
+    ["device_fs2", "content_fs"],
+    ["device_ss2", "content_ss"],
+    ["device_fp2", "content_fp"],
+    ["device_sp2", "content_sp"],
+    ["device_p2", "content_p"],
+    ["device_fd3", "content_fd"],
+    ["device_sd3", "content_sd"],
+    ["device_fs3", "content_fs"],
+    ["device_ss3", "content_ss"],
+    ["device_fp3", "content_fp"],
+    ["device_sp3", "content_sp"],
+    ["device_p3", "content_p"],
 ]
 
 
@@ -406,6 +434,22 @@ def join_features(device_df, tractate_df, cc_df):
     df["device_sp"] = df["second_positions_x"].apply(lambda x: nth_element(x, 0))
     df["device_p"] = df["projects_x"].apply(lambda x: nth_element(x, 0))
 
+    df["device_fd2"] = df["first_demands_x"].apply(lambda x: nth_element(x, 1))
+    df["device_sd2"] = df["second_demands_x"].apply(lambda x: nth_element(x, 1))
+    df["device_fs2"] = df["first_solutions_x"].apply(lambda x: nth_element(x, 1))
+    df["device_ss2"] = df["second_solutions_x"].apply(lambda x: nth_element(x, 1))
+    df["device_fp2"] = df["first_positions_x"].apply(lambda x: nth_element(x, 1))
+    df["device_sp2"] = df["second_positions_x"].apply(lambda x: nth_element(x, 1))
+    df["device_p2"] = df["projects_x"].apply(lambda x: nth_element(x, 1))
+
+    df["device_fd3"] = df["first_demands_x"].apply(lambda x: nth_element(x, 2))
+    df["device_sd3"] = df["second_demands_x"].apply(lambda x: nth_element(x, 2))
+    df["device_fs3"] = df["first_solutions_x"].apply(lambda x: nth_element(x, 2))
+    df["device_ss3"] = df["second_solutions_x"].apply(lambda x: nth_element(x, 2))
+    df["device_fp3"] = df["first_positions_x"].apply(lambda x: nth_element(x, 2))
+    df["device_sp3"] = df["second_positions_x"].apply(lambda x: nth_element(x, 2))
+    df["device_p3"] = df["projects_x"].apply(lambda x: nth_element(x, 2))
+
     df["content_fd"] = df["first_demands_y"].apply(lambda x: nth_element(x, 0))
     df["content_sd"] = df["second_demands_y"].apply(lambda x: nth_element(x, 0))
     df["content_fs"] = df["first_solutions_y"].apply(lambda x: nth_element(x, 0))
@@ -494,6 +538,20 @@ def device_tractate_fe(device_id, tractate_ids, device_dict, tractate_dict):
     device_info["device_fp"] = nth_element(device_fp, 0)
     device_info["device_sp"] = nth_element(device_sp, 0)
     device_info["device_p"] = nth_element(device_p, 0)
+    device_info["device_fd2"] = nth_element(device_fd, 1)
+    device_info["device_sd2"] = nth_element(device_sd, 1)
+    device_info["device_fs2"] = nth_element(device_fs, 1)
+    device_info["device_ss2"] = nth_element(device_ss, 1)
+    device_info["device_fp2"] = nth_element(device_fp, 1)
+    device_info["device_sp2"] = nth_element(device_sp, 1)
+    device_info["device_p2"] = nth_element(device_p, 1)
+    device_info["device_fd3"] = nth_element(device_fd, 2)
+    device_info["device_sd3"] = nth_element(device_sd, 2)
+    device_info["device_fs3"] = nth_element(device_fs, 2)
+    device_info["device_ss3"] = nth_element(device_ss, 2)
+    device_info["device_fp3"] = nth_element(device_fp, 2)
+    device_info["device_sp3"] = nth_element(device_sp, 2)
+    device_info["device_p3"] = nth_element(device_p, 2)
     tractate_lst = []
     tractate_ids_res = []
     for id in tractate_ids:

src/models/esmm/tractate_model.py

@@ -167,6 +167,20 @@ _categorical_columns = [
     "click_tractate_id4",
     "click_tractate_id5",
     "service_city",
+    "device_fd2",
+    "device_sd2",
+    "device_fs2",
+    "device_ss2",
+    "device_fp2",
+    "device_sp2",
+    "device_p2",
+    "device_fd3",
+    "device_sd3",
+    "device_fs3",
+    "device_ss3",
+    "device_fp3",
+    "device_sp3",
+    "device_p3",
 ]
 
 PREDICTION_ALL_COLUMNS = _int_columns + _float_columns + _categorical_columns

src/train_diary.py

@@ -59,7 +59,9 @@ def main():
 
     session_config = tf.compat.v1.ConfigProto()
     session_config.gpu_options.allow_growth = True
-    session_config.gpu_options.per_process_gpu_memory_fraction = 0.9
+    session_config.gpu_options.per_process_gpu_memory_fraction = 0.7
+    session_config.inter_op_parallelism_threads = 1
+    session_config.intra_op_parallelism_threads = 1
     estimator_config = tf.estimator.RunConfig(session_config=session_config)
 
     model = tf.estimator.Estimator(model_fn=esmm_model_fn, params=params, model_dir=model_path, config=estimator_config)

src/train_tractate.py

@@ -56,11 +56,14 @@ def main():
 
     session_config = tf.compat.v1.ConfigProto()
     session_config.gpu_options.allow_growth = True
-    session_config.gpu_options.per_process_gpu_memory_fraction = 0.9
+    session_config.gpu_options.per_process_gpu_memory_fraction = 0.7
+    session_config.inter_op_parallelism_threads = 1
+    session_config.intra_op_parallelism_threads = 1
     estimator_config = tf.estimator.RunConfig(session_config=session_config)
 
     model = tf.estimator.Estimator(model_fn=esmm_model_fn, params=params, model_dir=model_path, config=estimator_config)
-    train_spec = tf.estimator.TrainSpec(input_fn=lambda: esmm_input_fn(train_df, shuffle=True), max_steps=50000)
+    # TODO 50000
+    train_spec = tf.estimator.TrainSpec(input_fn=lambda: esmm_input_fn(train_df, shuffle=True), max_steps=15000)
     eval_spec = tf.estimator.EvalSpec(input_fn=lambda: esmm_input_fn(val_df, shuffle=False))
     res = tf.estimator.train_and_evaluate(model, train_spec, eval_spec)
     print("@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@")
@@ -75,8 +78,9 @@ def main():
     model_export_path = str(Path("/data/files/models/tractate/").expanduser())
     save_path = model_export(model, all_features, model_export_path)
     print("save to: " + save_path)
-    set_essm_model_save_path("tractate", save_path)
-    record_esmm_auc_to_db("tractate", ctr_auc, ctcvr_auc, total_time, save_path)
+    # TODO save model
+    # set_essm_model_save_path("tractate", save_path)
+    # record_esmm_auc_to_db("tractate", ctr_auc, ctcvr_auc, total_time, save_path)
     print("============================================================")
 
     # save_path = get_essm_model_save_path("tractate")