为什么DeepFM的Normalized Gini反而小于DNN和FM？

[QiDekang]

Normalized Gini越大效果越好，但是您给出的Performance中三张图片可以看出，DeepFM的Normalized Gini反而小于DNN和FM，说明DeepFM没有什么提升呀。能否请您解释一下，谢谢。

[guotong1988]

同问

[guotong1988]

#params: 13483
[1] train-result=-0.0392, valid-result=-0.0044 [14.6 s]
[2] train-result=0.1155, valid-result=0.2008 [16.5 s]
[3] train-result=0.2071, valid-result=0.2234 [12.4 s]
[4] train-result=0.2474, valid-result=0.2525 [12.1 s]
[5] train-result=0.2492, valid-result=0.2547 [12.0 s]
[6] train-result=0.2416, valid-result=0.2411 [12.1 s]
[7] train-result=0.2582, valid-result=0.2556 [12.2 s]
[8] train-result=0.2552, valid-result=0.2567 [13.8 s]
[9] train-result=0.2608, valid-result=0.2605 [12.4 s]
[10] train-result=0.2598, valid-result=0.2603 [11.8 s]
[11] train-result=0.2588, valid-result=0.2576 [13.2 s]
[12] train-result=0.2634, valid-result=0.2623 [11.9 s]
[13] train-result=0.2595, valid-result=0.2580 [12.0 s]
[14] train-result=0.2641, valid-result=0.2635 [13.8 s]
[15] train-result=0.2691, valid-result=0.2666 [11.9 s]
[16] train-result=0.2667, valid-result=0.2668 [12.0 s]
[17] train-result=0.2691, valid-result=0.2655 [14.1 s]
[18] train-result=0.2677, valid-result=0.2634 [14.0 s]
[19] train-result=0.2695, valid-result=0.2671 [16.3 s]
[20] train-result=0.2690, valid-result=0.2656 [12.2 s]
[21] train-result=0.2673, valid-result=0.2650 [12.3 s]
[22] train-result=0.2711, valid-result=0.2685 [13.2 s]
[23] train-result=0.2684, valid-result=0.2656 [12.1 s]
[24] train-result=0.2718, valid-result=0.2685 [12.2 s]
[25] train-result=0.2709, valid-result=0.2678 [13.8 s]
[26] train-result=0.2718, valid-result=0.2681 [12.1 s]
[27] train-result=0.2748, valid-result=0.2700 [12.3 s]
[28] train-result=0.2761, valid-result=0.2708 [12.1 s]
[29] train-result=0.2771, valid-result=0.2702 [12.2 s]
[30] train-result=0.2777, valid-result=0.2716 [14.4 s]
#params: 13483
[1] train-result=0.0686, valid-result=0.1276 [12.4 s]
[2] train-result=0.2211, valid-result=0.2289 [12.4 s]
[3] train-result=0.2590, valid-result=0.2534 [12.3 s]
[4] train-result=0.2614, valid-result=0.2472 [12.6 s]
[5] train-result=0.2589, valid-result=0.2521 [14.3 s]
[6] train-result=0.2613, valid-result=0.2549 [12.3 s]
[7] train-result=0.2632, valid-result=0.2494 [12.1 s]
[8] train-result=0.2648, valid-result=0.2515 [12.6 s]
[9] train-result=0.2670, valid-result=0.2554 [12.1 s]
[10] train-result=0.2645, valid-result=0.2534 [13.7 s]
[11] train-result=0.2666, valid-result=0.2553 [13.2 s]
[12] train-result=0.2704, valid-result=0.2608 [13.1 s]
[13] train-result=0.2724, valid-result=0.2632 [12.3 s]
[14] train-result=0.2748, valid-result=0.2647 [12.5 s]
[15] train-result=0.2759, valid-result=0.2652 [12.3 s]
[16] train-result=0.2773, valid-result=0.2670 [14.4 s]
[17] train-result=0.2779, valid-result=0.2672 [12.8 s]
[18] train-result=0.2785, valid-result=0.2682 [12.5 s]
[19] train-result=0.2789, valid-result=0.2685 [12.7 s]
[20] train-result=0.2791, valid-result=0.2685 [11.9 s]
[21] train-result=0.2793, valid-result=0.2688 [14.7 s]
[22] train-result=0.2794, valid-result=0.2689 [12.1 s]
[23] train-result=0.2797, valid-result=0.2684 [12.5 s]
[24] train-result=0.2797, valid-result=0.2687 [12.2 s]
[25] train-result=0.2800, valid-result=0.2693 [12.3 s]
[26] train-result=0.2800, valid-result=0.2692 [13.0 s]
[27] train-result=0.2802, valid-result=0.2695 [13.6 s]
[28] train-result=0.2803, valid-result=0.2690 [12.2 s]
[29] train-result=0.2801, valid-result=0.2685 [12.0 s]
[30] train-result=0.2804, valid-result=0.2695 [12.2 s]
#params: 13483
[1] train-result=0.0570, valid-result=0.1398 [13.9 s]
[2] train-result=0.2418, valid-result=0.2305 [12.3 s]
[3] train-result=0.2592, valid-result=0.2424 [12.2 s]
[4] train-result=0.2635, valid-result=0.2477 [12.1 s]
[5] train-result=0.2696, valid-result=0.2524 [12.1 s]
[6] train-result=0.2740, valid-result=0.2549 [12.2 s]
[7] train-result=0.2763, valid-result=0.2563 [13.4 s]
[8] train-result=0.2785, valid-result=0.2583 [11.9 s]
[9] train-result=0.2801, valid-result=0.2588 [12.0 s]
[10] train-result=0.2810, valid-result=0.2588 [12.2 s]
[11] train-result=0.2818, valid-result=0.2598 [12.3 s]
[12] train-result=0.2826, valid-result=0.2596 [13.6 s]
[13] train-result=0.2833, valid-result=0.2601 [12.0 s]
[14] train-result=0.2835, valid-result=0.2601 [12.2 s]
[15] train-result=0.2838, valid-result=0.2605 [12.2 s]
[16] train-result=0.2840, valid-result=0.2607 [12.2 s]
[17] train-result=0.2841, valid-result=0.2603 [12.1 s]
[18] train-result=0.2843, valid-result=0.2607 [14.5 s]
[19] train-result=0.2846, valid-result=0.2614 [13.0 s]
[20] train-result=0.2846, valid-result=0.2610 [13.1 s]
[21] train-result=0.2849, valid-result=0.2610 [12.7 s]
[22] train-result=0.2849, valid-result=0.2609 [12.5 s]
[23] train-result=0.2850, valid-result=0.2609 [13.9 s]
[24] train-result=0.2851, valid-result=0.2612 [13.0 s]
[25] train-result=0.2852, valid-result=0.2613 [12.5 s]
[26] train-result=0.2851, valid-result=0.2616 [12.7 s]
[27] train-result=0.2852, valid-result=0.2618 [12.7 s]
[28] train-result=0.2854, valid-result=0.2619 [12.7 s]
[29] train-result=0.2855, valid-result=0.2621 [14.1 s]
[30] train-result=0.2854, valid-result=0.2616 [13.0 s]
DeepFM: 0.26758 (0.00431)

#params: 13451
[1] train-result=0.2561, valid-result=0.2535 [10.7 s]
[2] train-result=0.2696, valid-result=0.2658 [11.3 s]
[3] train-result=0.2732, valid-result=0.2688 [9.6 s]
[4] train-result=0.2746, valid-result=0.2691 [10.7 s]
[5] train-result=0.2757, valid-result=0.2698 [10.2 s]
[6] train-result=0.2771, valid-result=0.2706 [9.6 s]
[7] train-result=0.2780, valid-result=0.2697 [11.1 s]
[8] train-result=0.2787, valid-result=0.2718 [9.7 s]
[9] train-result=0.2791, valid-result=0.2718 [9.6 s]
[10] train-result=0.2810, valid-result=0.2716 [9.8 s]
[11] train-result=0.2840, valid-result=0.2723 [10.1 s]
[12] train-result=0.2843, valid-result=0.2737 [9.9 s]
[13] train-result=0.2849, valid-result=0.2740 [11.6 s]
[14] train-result=0.2864, valid-result=0.2744 [10.1 s]
[15] train-result=0.2874, valid-result=0.2737 [9.7 s]
[16] train-result=0.2877, valid-result=0.2732 [9.8 s]
[17] train-result=0.2864, valid-result=0.2750 [9.7 s]
[18] train-result=0.2880, valid-result=0.2729 [11.2 s]
[19] train-result=0.2885, valid-result=0.2746 [9.3 s]
[20] train-result=0.2894, valid-result=0.2737 [9.8 s]
[21] train-result=0.2895, valid-result=0.2744 [9.5 s]
[22] train-result=0.2897, valid-result=0.2734 [10.3 s]
[23] train-result=0.2904, valid-result=0.2740 [10.3 s]
[24] train-result=0.2908, valid-result=0.2723 [11.2 s]
[25] train-result=0.2911, valid-result=0.2740 [9.6 s]
[26] train-result=0.2920, valid-result=0.2744 [9.6 s]
[27] train-result=0.2923, valid-result=0.2737 [9.8 s]
[28] train-result=0.2931, valid-result=0.2735 [9.8 s]
[29] train-result=0.2931, valid-result=0.2739 [11.1 s]
[30] train-result=0.2939, valid-result=0.2742 [9.9 s]
#params: 13451
[1] train-result=0.2587, valid-result=0.2499 [10.0 s]
[2] train-result=0.2693, valid-result=0.2590 [10.2 s]
[3] train-result=0.2737, valid-result=0.2622 [10.2 s]
[4] train-result=0.2754, valid-result=0.2653 [11.8 s]
[5] train-result=0.2778, valid-result=0.2662 [10.2 s]
[6] train-result=0.2784, valid-result=0.2669 [10.0 s]
[7] train-result=0.2787, valid-result=0.2686 [9.6 s]
[8] train-result=0.2793, valid-result=0.2678 [9.3 s]
[9] train-result=0.2797, valid-result=0.2700 [11.5 s]
[10] train-result=0.2803, valid-result=0.2704 [10.5 s]
[11] train-result=0.2803, valid-result=0.2700 [10.0 s]
[12] train-result=0.2803, valid-result=0.2698 [9.7 s]
[13] train-result=0.2808, valid-result=0.2703 [9.4 s]
[14] train-result=0.2807, valid-result=0.2696 [9.7 s]
[15] train-result=0.2815, valid-result=0.2704 [12.3 s]
[16] train-result=0.2827, valid-result=0.2705 [10.3 s]
[17] train-result=0.2834, valid-result=0.2707 [10.9 s]
[18] train-result=0.2850, valid-result=0.2723 [9.6 s]
[19] train-result=0.2863, valid-result=0.2725 [9.4 s]
[20] train-result=0.2881, valid-result=0.2708 [11.1 s]
[21] train-result=0.2906, valid-result=0.2722 [9.6 s]
[22] train-result=0.2913, valid-result=0.2735 [9.7 s]
[23] train-result=0.2924, valid-result=0.2734 [9.9 s]
[24] train-result=0.2932, valid-result=0.2741 [9.5 s]
[25] train-result=0.2951, valid-result=0.2733 [10.5 s]
[26] train-result=0.2957, valid-result=0.2730 [11.7 s]
[27] train-result=0.2965, valid-result=0.2725 [9.7 s]
[28] train-result=0.2979, valid-result=0.2722 [10.1 s]
[29] train-result=0.2989, valid-result=0.2719 [9.7 s]
[30] train-result=0.2995, valid-result=0.2713 [9.6 s]
#params: 13451
[1] train-result=0.2550, valid-result=0.2375 [9.8 s]
[2] train-result=0.2718, valid-result=0.2515 [9.5 s]
[3] train-result=0.2777, valid-result=0.2558 [9.8 s]
[4] train-result=0.2784, valid-result=0.2565 [9.4 s]
[5] train-result=0.2803, valid-result=0.2575 [9.6 s]
[6] train-result=0.2814, valid-result=0.2594 [11.2 s]
[7] train-result=0.2819, valid-result=0.2595 [10.2 s]
[8] train-result=0.2828, valid-result=0.2592 [9.7 s]
[9] train-result=0.2837, valid-result=0.2605 [9.8 s]
[10] train-result=0.2839, valid-result=0.2597 [9.7 s]
[11] train-result=0.2842, valid-result=0.2603 [11.0 s]
[12] train-result=0.2847, valid-result=0.2612 [9.7 s]
[13] train-result=0.2844, valid-result=0.2594 [9.5 s]
[14] train-result=0.2858, valid-result=0.2626 [9.5 s]
[15] train-result=0.2867, valid-result=0.2619 [9.9 s]
[16] train-result=0.2883, valid-result=0.2628 [10.0 s]
[17] train-result=0.2894, valid-result=0.2639 [11.1 s]
[18] train-result=0.2907, valid-result=0.2640 [9.5 s]
[19] train-result=0.2916, valid-result=0.2640 [9.5 s]
[20] train-result=0.2934, valid-result=0.2634 [9.6 s]
[21] train-result=0.2935, valid-result=0.2631 [10.1 s]
[22] train-result=0.2958, valid-result=0.2648 [10.1 s]
[23] train-result=0.2968, valid-result=0.2649 [12.1 s]
[24] train-result=0.2988, valid-result=0.2655 [10.2 s]
[25] train-result=0.3017, valid-result=0.2636 [10.1 s]
[26] train-result=0.3039, valid-result=0.2644 [9.9 s]
[27] train-result=0.3040, valid-result=0.2649 [9.8 s]
[28] train-result=0.3065, valid-result=0.2634 [11.3 s]
[29] train-result=0.3079, valid-result=0.2650 [9.9 s]
[30] train-result=0.3090, valid-result=0.2644 [9.9 s]
FM: 0.26998 (0.00413)


#params: 13436
[1] train-result=0.0063, valid-result=-0.0136 [12.5 s]
[2] train-result=-0.0116, valid-result=0.0828 [13.6 s]
[3] train-result=0.1636, valid-result=0.0883 [12.4 s]
[4] train-result=0.1218, valid-result=0.1412 [12.2 s]
[5] train-result=0.0932, valid-result=0.0798 [12.1 s]
[6] train-result=0.0372, valid-result=0.0577 [12.1 s]
[7] train-result=0.0665, valid-result=0.0691 [12.3 s]
[8] train-result=0.2019, valid-result=0.1829 [13.8 s]
[9] train-result=0.1617, valid-result=0.1506 [11.7 s]
[10] train-result=0.0629, valid-result=0.0895 [12.2 s]
[11] train-result=0.2188, valid-result=0.2101 [11.9 s]
[12] train-result=0.2133, valid-result=0.2032 [12.2 s]
[13] train-result=0.2398, valid-result=0.2286 [13.5 s]
[14] train-result=0.2368, valid-result=0.2149 [12.3 s]
[15] train-result=0.2388, valid-result=0.2023 [12.3 s]
[16] train-result=0.2250, valid-result=0.2145 [12.2 s]
[17] train-result=0.2651, valid-result=0.2518 [12.3 s]
[18] train-result=0.2675, valid-result=0.2630 [11.8 s]
[19] train-result=0.2665, valid-result=0.2573 [13.5 s]
[20] train-result=0.2695, valid-result=0.2667 [11.8 s]
[21] train-result=0.2684, valid-result=0.2609 [11.8 s]
[22] train-result=0.2694, valid-result=0.2644 [12.5 s]
[23] train-result=0.2690, valid-result=0.2671 [12.4 s]
[24] train-result=0.2691, valid-result=0.2637 [13.3 s]
[25] train-result=0.2691, valid-result=0.2626 [11.8 s]
[26] train-result=0.2684, valid-result=0.2616 [11.6 s]
[27] train-result=0.2689, valid-result=0.2641 [11.6 s]
[28] train-result=0.2731, valid-result=0.2664 [11.7 s]
[29] train-result=0.2743, valid-result=0.2675 [12.2 s]
[30] train-result=0.2695, valid-result=0.2635 [13.5 s]
#params: 13436
[1] train-result=-0.0488, valid-result=-0.0599 [12.1 s]
[2] train-result=-0.0228, valid-result=0.0495 [11.6 s]
[3] train-result=0.0475, valid-result=0.0737 [11.6 s]
[4] train-result=0.0525, valid-result=0.0824 [13.4 s]
[5] train-result=0.0822, valid-result=0.0832 [12.3 s]
[6] train-result=0.1182, valid-result=0.1217 [11.7 s]
[7] train-result=0.0537, valid-result=0.0999 [11.6 s]
[8] train-result=0.1158, valid-result=0.1194 [11.8 s]
[9] train-result=0.0882, valid-result=0.1156 [12.3 s]
[10] train-result=0.1039, valid-result=0.1080 [13.5 s]
[11] train-result=0.1511, valid-result=0.1534 [11.7 s]
[12] train-result=0.1728, valid-result=0.1652 [12.2 s]
[13] train-result=0.1656, valid-result=0.1655 [11.8 s]
[14] train-result=0.1809, valid-result=0.1734 [12.3 s]
[15] train-result=0.2066, valid-result=0.1979 [13.3 s]
[16] train-result=0.2116, valid-result=0.1975 [12.6 s]
[17] train-result=0.2202, valid-result=0.2105 [12.6 s]
[18] train-result=0.2409, valid-result=0.2296 [13.8 s]
[19] train-result=0.2645, valid-result=0.2497 [12.4 s]
[20] train-result=0.2693, valid-result=0.2556 [12.4 s]
[21] train-result=0.2748, valid-result=0.2649 [13.9 s]
[22] train-result=0.2728, valid-result=0.2647 [11.9 s]
[23] train-result=0.2757, valid-result=0.2627 [11.8 s]
[24] train-result=0.2704, valid-result=0.2596 [11.9 s]
[25] train-result=0.2735, valid-result=0.2610 [11.9 s]
[26] train-result=0.2698, valid-result=0.2609 [11.7 s]
[27] train-result=0.2723, valid-result=0.2597 [13.0 s]
[28] train-result=0.2740, valid-result=0.2622 [12.0 s]
[29] train-result=0.2707, valid-result=0.2565 [12.0 s]
[30] train-result=0.2697, valid-result=0.2600 [12.0 s]
#params: 13436
[1] train-result=0.0104, valid-result=0.0764 [13.5 s]
[2] train-result=0.0649, valid-result=0.0848 [12.2 s]
[3] train-result=0.0714, valid-result=0.0771 [11.8 s]
[4] train-result=0.1314, valid-result=0.1457 [12.1 s]
[5] train-result=-0.0250, valid-result=-0.0095 [11.9 s]
[6] train-result=0.2239, valid-result=0.2153 [12.8 s]
[7] train-result=0.1514, valid-result=0.1481 [11.9 s]
[8] train-result=0.1887, valid-result=0.1893 [12.2 s]
[9] train-result=0.2378, valid-result=0.2133 [11.9 s]
[10] train-result=0.2207, valid-result=0.2195 [11.5 s]
[11] train-result=0.2400, valid-result=0.2112 [11.7 s]
[12] train-result=0.2476, valid-result=0.2240 [13.0 s]
[13] train-result=0.2571, valid-result=0.2358 [11.5 s]
[14] train-result=0.2634, valid-result=0.2450 [11.9 s]
[15] train-result=0.2711, valid-result=0.2522 [11.9 s]
[16] train-result=0.2711, valid-result=0.2544 [12.1 s]
[17] train-result=0.2690, valid-result=0.2537 [12.6 s]
[18] train-result=0.2756, valid-result=0.2540 [13.0 s]
[19] train-result=0.2738, valid-result=0.2536 [12.8 s]
[20] train-result=0.2769, valid-result=0.2555 [11.4 s]
[21] train-result=0.2744, valid-result=0.2567 [12.3 s]
[22] train-result=0.2743, valid-result=0.2516 [12.6 s]
[23] train-result=0.2784, valid-result=0.2578 [14.3 s]
[24] train-result=0.2749, valid-result=0.2485 [11.8 s]
[25] train-result=0.2783, valid-result=0.2581 [11.6 s]
[26] train-result=0.2769, valid-result=0.2545 [11.9 s]
[27] train-result=0.2756, valid-result=0.2513 [12.0 s]
[28] train-result=0.2794, valid-result=0.2559 [11.6 s]
[29] train-result=0.2797, valid-result=0.2564 [13.1 s]
[30] train-result=0.2769, valid-result=0.2548 [11.4 s]
DNN: 0.25931 (0.00348)