item2vec 测试结果

1、使用半年内所有单据训练，并使用90天内的网单单据验证，训练集包含了验证集

precision, size, ns_exponent, ns

[(0.32391340274334823, 170, -0.04999999999999993, 9),
(0.3202776400594943, 350, -0.04999999999999993, 9),
(0.31994711617914395, 450, -0.1499999999999999, 9),
(0.31912080647826807, 390, -0.09999999999999998, 7),
(0.3189555445380929, 970, -0.1499999999999999, 9),
(0.3186250206577425, 170, -0.1499999999999999, 9),
(0.3186250206577425, 310, -0.04999999999999993, 9),
(0.3186250206577425, 890, -0.09999999999999998, 9),
(0.318129234837217, 250, -0.09999999999999998, 9),
(0.31746818707651625, 530, -0.04999999999999993, 9),
(0.3173029251363411, 650, -0.04999999999999993, 7),
(0.31647661543546524, 250, -0.04999999999999993, 9),
(0.3159808296149397, 270, -0.1499999999999999, 9),
(0.3159808296149397, 670, -0.1499999999999999, 9),
(0.3159808296149397, 850, -0.04999999999999993, 9),
(0.3158155676747645, 730, -0.1499999999999999, 7),
(0.3156503057345893, 430, -0.09999999999999998, 9),
(0.31548504379441417, 230, -0.09999999999999998, 9),
(0.315319781854239, 990, -0.1499999999999999, 9),
(0.3151545199140638, 150, -0.04999999999999993, 9),
(0.3151545199140638, 690, -0.04999999999999993, 9),
(0.3149892579738886, 90, -0.04999999999999993, 9),
(0.3149892579738886, 210, -0.1499999999999999, 9),
(0.3149892579738886, 290, -0.04999999999999993, 9),
(0.3149892579738886, 510, -0.1499999999999999, 9),
(0.3149892579738886, 710, -0.19999999999999996, 9),
(0.3149892579738886, 710, -0.09999999999999998, 9),
(0.3148239960337134, 390, -0.19999999999999996, 9),
(0.3148239960337134, 470, -0.09999999999999998, 9),
(0.3148239960337134, 490, -0.09999999999999998, 7)]

 

2、训练样本：过去180-30天的数据所有单据， 验证：近30天的网单

参数：size=170, ns_exponent=-0.2, negtive=9, window=999

预测结果分布
[('爆浆肉松包', 1744), ('咸蛋超人', 1033), ('虎皮蛋卷', 963), ('奶酪包', 871), ('甜甜圈', 667), ('原味吐司', 635), ('宇治抹茶', 487), ('手工蛋挞', 421), ('法式乳香片', 392), ('杂粮三明治', 342), ('牛角包', 340), ('椰奶餐包', 326), ('焦糖杏仁卷', 230), ('火辣包', 208), ('手工蛋挞（6粒装）', 204), ('车轮泡芙', 202), ('牛奶棒', 170), ('蔓越莓吐司', 168), ('黑眼豆豆', 161), ('酸奶（水果/坚果）', 156), ('推推乐', 121), ('奥利奥软欧', 115), ('全麦红提', 90), ('果蔬溶豆', 90), ('坚果软欧', 86), ('椰蓉手撕包', 81), ('全麦坚果吐司', 78), ('椰蓉吐司', 71), ('焦糖布朗尼', 69), ('雪媚娘', 63)]
正确结果分布
[('爆浆肉松包', 151), ('咸蛋超人', 95), ('虎皮蛋卷', 73), ('原味吐司', 49), ('奶酪包', 43), ('宇治抹茶', 35), ('甜甜圈', 31), ('牛角包', 24), ('杂粮三明治', 23), ('法式乳香片', 23), ('手工蛋挞', 22), ('车轮泡芙', 21), ('椰奶餐包', 15), ('果蔬溶豆', 14), ('推推乐', 12), ('手工蛋挞（6粒装）', 10), ('黑眼豆豆', 9), ('焦糖杏仁卷', 9), ('全麦坚果吐司', 9), ('全麦红提', 8), ('奥利奥软欧', 8), ('雪媚娘', 8), ('酸奶（水果/坚果）', 7), ('纽扣饼干', 7), ('火辣包', 6), ('蔓越莓吐司', 5), ('焦糖布朗尼', 5), ('雪花酥', 4), ('提拉米苏', 3), ('牛奶棒', 3)]
(3755, 0.2029294274300932)

3571448

pr.update_model(size=170, ns_exponent=0, negtive=50, window=999)

预测结果分布
[('千层手工蛋挞（新款升级口味）', 1805), ('泡芙', 1206), ('沙拉三明治', 1115), ('毛毛虫', 920), ('火腿王', 823), ('乳香包', 679), ('岩烧', 631), ('酸奶包', 602), ('雪媚娘', 592), ('脏脏蛋挞', 592), ('菠萝蜜', 583), ('肉松卷', 580), ('枣糕', 570), ('巧克力甜甜圈', 530), ('奶香片', 521), ('推推乐', 501), ('椰香小吐司', 482), ('可可奶昔包', 470), ('大理石', 468), ('北海道', 436), ('小牛角包', 427), ('脏脏包', 389), ('老面包（原味）', 367), ('早餐包', 350), ('脆皮草莓', 322), ('红豆切片', 322), ('黑森林切角', 318), ('草莓甜甜圈', 312), ('奶昔包', 278), ('唱片', 268)]
正确结果分布
[('泡芙', 45), ('沙拉三明治', 39), ('千层手工蛋挞（新款升级口味）', 38), ('雪媚娘', 27), ('毛毛虫', 27), ('肉松卷', 25), ('乳香包', 24), ('黑森林切角', 20), ('枣糕', 20), ('火腿王', 18), ('推推乐', 17), ('酸奶包', 16), ('岩烧', 16), ('大理石', 15), ('巧克力甜甜圈', 15), ('北海道', 14), ('可可奶昔包', 14), ('奶香片', 12), ('老面包（原味）', 12), ('水果蛋糕切角', 11), ('奶昔包', 11), ('菠萝蜜', 11), ('红豆切片', 11), ('早餐包', 9), ('脏脏蛋挞', 9), ('原味甜甜圈', 8), ('樱桃之恋', 8), ('梦龙卷', 8), ('椰香小吐司', 7), ('抹茶蜜豆千层切角', 7)]
Out[525]: (4857, 0.13218035824583077)

 

pr.update_model(size=170, ns_exponent=-0.2, negtive=9, window=999)

预测结果分布
[('千层手工蛋挞（新款升级口味）', 3001), ('泡芙', 1502), ('毛毛虫', 1297), ('雪媚娘', 1069), ('火腿王', 941), ('脏脏蛋挞', 925), ('沙拉三明治', 862), ('肉松卷', 820), ('巧克力甜甜圈', 723), ('菠萝蜜', 654), ('乳香包', 599), ('原味甜甜圈', 596), ('小牛角包', 498), ('酸奶包', 493), ('大理石', 468), ('椰香小吐司', 427), ('岩烧', 425), ('推推乐', 417), ('可可奶昔包', 416), ('北海道', 389), ('老面包（原味）', 382), ('奶香片', 357), ('黑森林切角', 344), ('香蕉抱抱', 273), ('三明治', 252), ('榴莲酥', 245), ('草莓甜甜圈', 241), ('原味切片', 238), ('早餐包', 237), ('老面包（豆沙）', 236)]
正确结果分布
[('千层手工蛋挞（新款升级口味）', 55), ('雪媚娘', 54), ('泡芙', 43), ('毛毛虫', 36), ('肉松卷', 36), ('乳香包', 27), ('沙拉三明治', 23), ('巧克力甜甜圈', 22), ('大理石', 22), ('黑森林切角', 21), ('火腿王', 20), ('原味甜甜圈', 17), ('岩烧', 16), ('推推乐', 15), ('北海道', 14), ('可可奶昔包', 14), ('原味切片', 14), ('菠萝蜜', 12), ('酸奶包', 12), ('脏脏蛋挞', 11), ('奶昔包', 10), ('老面包（原味）', 10), ('奶香片', 10), ('水果蛋糕切角', 8), ('樱桃之恋', 8), ('椰香小吐司', 7), ('鲜花饼', 7), ('小牛角包', 7), ('早餐包', 7), ('枣糕', 7)]
Out[527]: (4857, 0.1362981264154828)