模型预测

predict/dict_predict.py

 #!/usr/bin/env python
 # -*- coding:utf-8 -*-
+from collections import Counter
+
 from predict.base_handler import BasePredictor
 
 # 可能的头部字段
@@ -16,9 +18,11 @@ fields_map = {"序号": ["序号"],
               "封装": ["封装", "封装规格"],
               "品牌": ["品牌", "品牌/厂商", "参考品牌", "参考供应商", "厂商", "参考供应商", "参考厂商"]}
 
+# 必须返回也必须验证的标准字段
+MUST_STD_FIELDS = ['参数', '数量']
 #
 order_list = ['序号']
-
+en_to_zh_map = {'brand_name': '品牌', 'param': '参数', 'goods_name': '型号'}
 
 def fun(seri):
     li_seri = seri.tolist()
@@ -31,7 +35,8 @@ def fun(seri):
 HEAD_ROW = 7
 # 空置率阈值
 NAN_RATE = 0.8
-
+# 占位符
+PLACEHOLDER = '?'
 
 
 class DicPredict(BasePredictor):
@@ -63,14 +68,6 @@ class DicPredict(BasePredictor):
         columns = []
         li_data = []
         for k, v in dict_data.items():
-
-            # 去掉空置率大于等于0.8的列
-            counter = 0
-            for item in v:
-                if not str(item).strip():
-                    counter += 1
-            if counter / len(v) >= NAN_RATE:
-                continue
             columns.append(k)
             li_data.append(v)
 
@@ -87,16 +84,44 @@ class DicPredict(BasePredictor):
             else:
                 dic_ab_res = {i['column_name']: i['pronoun']}
                 ab_result.append(dic_ab_res)
-        id_res = {
+        pre_id_res = {
             'std_result': std_result,
             'ab_result': ab_result,
         }
-        return id_res
+        return pre_id_res
+
+    def model_predict(self, dic_data):
+        """
+        该方法目前只对[参数, 型号, 数量, 品牌]进行预测
+        :param dic_data:
+        :return:
+        """
+        prob_columns = []
+        for k, v in dic_data.items():
+            if self.valid_chain(v):
+                prob_columns.append(k)
+
+        temp_dic_data = {k: list(filter(lambda x: x != PLACEHOLDER, dic_data[k]))for k in prob_columns}
+        temp_pre_model_res = {}
+        for k, v in temp_dic_data.items():
+            li_single_pred_res = []
+            for string in v:
+                single_pred_res, probdic = self.get_single_predict(string)
+                li_single_pred_res.append(single_pred_res)
+            result = Counter(li_single_pred_res)
+            # [('brand_name', 4), ('goods_name', 3), ('param', 2)]
+            li_sort = sorted(result.items(), key=lambda x: x[1], reverse=True)
+            temp_pre_model_res[k] = en_to_zh_map[li_sort[0][0]]
+        print(temp_pre_model_res)
+
+        # pre_model_res = {}
 
     def predict(self, dic_data):
-        res = self.pre_predict(dic_data)
-        if res:
-            return res
+        dic_data = self.pre_deal(dic_data)
+        pre_id_res = self.pre_predict(dic_data)
+        if pre_id_res:
+            return pre_id_res
+
         if len(dic_data) > 0:
             self.order_predict(dic_data[0])
         pass
@@ -115,3 +140,19 @@ class DicPredict(BasePredictor):
         if size == 1:
             return True
         return (arr[size - 1] >= arr[size - 2]) and self.IsIncrease(arr, size - 1)
+
+    def pre_deal(self, dic_data):
+        new_dic_data = {}
+        for k, v in dic_data.items():
+            # 去掉空置率大于等于0.8的列
+            counter = 0
+            for item in v:
+                if str(item).strip() == PLACEHOLDER:
+                    counter += 1
+            if counter / len(v) <= NAN_RATE:
+                new_dic_data[k] = v
+
+        return new_dic_data
+
+    def valid_chain(self, li):
+        pass
\ No newline at end of file