try cross columns for diary

src/models/esmm/fe/diary_fe.py

@@ -304,6 +304,15 @@ CATEGORICAL_COLUMNS = [
     "is_related_service",
     "service_city",
 ]
+CROSS_COLUMNS = [
+    ["device_fd", "content_fd"],
+    ["device_sd", "content_sd"],
+    ["device_fs", "content_fs"],
+    ["device_ss", "content_ss"],
+    ["device_fp", "content_fp"],
+    ["device_sp", "content_sp"],
+    ["device_p", "content_p"],
+]
 
 
 def read_csv_data(dataset_path):

src/models/esmm/fe/fe.py

@@ -4,7 +4,15 @@ from tensorflow import feature_column as fc
 from ..utils import create_boundaries, create_vocabulary_list
 
 
-def build_features(df, int_columns, float_columns, categorical_columns):
+def build_features(df, int_columns, float_columns, categorical_columns, cross_columns=[]):
+    """
+    df:
+    int_columns:
+    float_columns:
+    categorical_columns:
+    cross_columns=[]:
+    return:
+    """
     numeric_features = []
     for col in (int_columns + float_columns):
         if col in int_columns:
@@ -33,11 +41,13 @@ def build_features(df, int_columns, float_columns, categorical_columns):
             categorical_features.append(
                 fc.indicator_column(fc.categorical_column_with_vocabulary_list(col, create_vocabulary_list(df, col))))
 
-    # TODO try cross feature
-    a = fc.categorical_column_with_vocabulary_list("device_fd", create_vocabulary_list(df, "device_fd"))
-    b = fc.categorical_column_with_vocabulary_list("content_fd", create_vocabulary_list(df, "content_fd"))
-    c = tf.feature_column.crossed_column([a, b], hash_bucket_size=10)
-    categorical_features.append(tf.feature_column.indicator_column(c))
+    for lst in cross_columns:
+        column_a = lst[0]
+        column_b = lst[1]
+        tmp_a = fc.categorical_column_with_vocabulary_list(column_a, create_vocabulary_list(df, column_a))
+        tmp_b = fc.categorical_column_with_vocabulary_list(column_b, create_vocabulary_list(df, column_b))
+        tmp_res = tf.feature_column.crossed_column([tmp_a, tmp_b], hash_bucket_size=10)
+        categorical_features.append(tf.feature_column.indicator_column(tmp_res))
 
     all_features = (numeric_features + categorical_features)
     return all_features

src/train_diary.py

@@ -49,7 +49,8 @@ def main():
     train_df, test_df = train_test_split(df, test_size=0.2)
     train_df, val_df = train_test_split(train_df, test_size=0.2)
 
-    all_features = fe.build_features(df, diary_fe.INT_COLUMNS, diary_fe.FLOAT_COLUMNS, diary_fe.CATEGORICAL_COLUMNS)
+    all_features = fe.build_features(df, diary_fe.INT_COLUMNS, diary_fe.FLOAT_COLUMNS, diary_fe.CATEGORICAL_COLUMNS,
+                                     diary_fe.CROSS_COLUMNS)
     params = {"feature_columns": all_features, "hidden_units": [360, 200, 80, 2], "learning_rate": 0.2}
     model_path = str(Path("/data/files/model_tmp/diary/").expanduser())
     if os.path.exists(model_path):
@@ -79,7 +80,7 @@ def main():
     print("============================================================")
 
     # save_path = str(Path("~/Desktop/models/1596012827").expanduser())  # local
-    # save_path = "/data/files/models/diary/1597390452"  # server
+    # save_path = "/data/files/models/diary/1598948682"  # server
 
     # tf.saved_model.load