关于代码107行

[sudongxiang]

trafficstars

deepFM.py 第107行为：self.y_deep = tf.nn.dropout(self.y_deep, self.dropout_keep_deep[0])，本人在实验中，删除该行，效果反而变好了很多。查阅资料后，发现dropout一般的对象是deep的中间节点，而该行是针对初始节点。

楼主可以自行实验，以加验证。

祝好~

[guotong1988]

THX

[guotong1988]

[1] train-result=0.0228, valid-result=0.0824 [12.5 s]
[2] train-result=0.2438, valid-result=0.2641 [12.9 s]
[3] train-result=0.2603, valid-result=0.2621 [11.1 s]
[4] train-result=0.2605, valid-result=0.2643 [11.4 s]
[5] train-result=0.2674, valid-result=0.2666 [11.5 s]
[6] train-result=0.2704, valid-result=0.2673 [12.3 s]
[7] train-result=0.2654, valid-result=0.2638 [11.2 s]
[8] train-result=0.2703, valid-result=0.2666 [13.0 s]
[9] train-result=0.2736, valid-result=0.2680 [11.6 s]
[10] train-result=0.2702, valid-result=0.2648 [11.3 s]
[11] train-result=0.2741, valid-result=0.2672 [12.3 s]
[12] train-result=0.2746, valid-result=0.2686 [13.2 s]
[13] train-result=0.2772, valid-result=0.2705 [12.0 s]
[14] train-result=0.2759, valid-result=0.2683 [13.5 s]
[15] train-result=0.2777, valid-result=0.2699 [11.9 s]
[16] train-result=0.2780, valid-result=0.2707 [12.2 s]
[17] train-result=0.2790, valid-result=0.2707 [11.9 s]
[18] train-result=0.2796, valid-result=0.2700 [12.0 s]
[19] train-result=0.2799, valid-result=0.2708 [13.5 s]
[20] train-result=0.2800, valid-result=0.2700 [11.6 s]
[21] train-result=0.2799, valid-result=0.2706 [11.8 s]
[22] train-result=0.2801, valid-result=0.2699 [11.5 s]
[23] train-result=0.2802, valid-result=0.2701 [11.8 s]
[24] train-result=0.2803, valid-result=0.2708 [11.5 s]
[25] train-result=0.2803, valid-result=0.2702 [13.4 s]
[26] train-result=0.2806, valid-result=0.2709 [11.3 s]
[27] train-result=0.2807, valid-result=0.2707 [11.5 s]
[28] train-result=0.2806, valid-result=0.2704 [11.1 s]
[29] train-result=0.2809, valid-result=0.2711 [12.8 s]
[30] train-result=0.2807, valid-result=0.2709 [13.4 s]
#params: 13483
[1] train-result=-0.0700, valid-result=0.0149 [12.0 s]
[2] train-result=0.2547, valid-result=0.2530 [13.0 s]
[3] train-result=0.2713, valid-result=0.2615 [12.6 s]
[4] train-result=0.2628, valid-result=0.2552 [13.2 s]
[5] train-result=0.2721, valid-result=0.2622 [13.4 s]
[6] train-result=0.2701, valid-result=0.2601 [12.6 s]
[7] train-result=0.2709, valid-result=0.2640 [11.8 s]
[8] train-result=0.2734, valid-result=0.2648 [11.7 s]
[9] train-result=0.2756, valid-result=0.2660 [12.1 s]
[10] train-result=0.2742, valid-result=0.2647 [13.4 s]
[11] train-result=0.2760, valid-result=0.2661 [11.6 s]
[12] train-result=0.2774, valid-result=0.2666 [11.9 s]
[13] train-result=0.2782, valid-result=0.2681 [11.6 s]
[14] train-result=0.2788, valid-result=0.2681 [11.7 s]
[15] train-result=0.2793, valid-result=0.2687 [12.1 s]
[16] train-result=0.2794, valid-result=0.2693 [13.3 s]
[17] train-result=0.2797, valid-result=0.2697 [12.4 s]
[18] train-result=0.2802, valid-result=0.2692 [11.7 s]
[19] train-result=0.2800, valid-result=0.2693 [11.6 s]
[20] train-result=0.2799, valid-result=0.2691 [12.0 s]
[21] train-result=0.2805, valid-result=0.2695 [13.5 s]
[22] train-result=0.2808, valid-result=0.2702 [11.6 s]
[23] train-result=0.2806, valid-result=0.2704 [11.6 s]
[24] train-result=0.2809, valid-result=0.2703 [12.0 s]
[25] train-result=0.2806, valid-result=0.2699 [11.3 s]
[26] train-result=0.2808, valid-result=0.2700 [11.8 s]
[27] train-result=0.2812, valid-result=0.2706 [12.6 s]
[28] train-result=0.2809, valid-result=0.2708 [11.3 s]
[29] train-result=0.2811, valid-result=0.2706 [11.4 s]
[30] train-result=0.2813, valid-result=0.2702 [12.0 s]
#params: 13483
[1] train-result=-0.0498, valid-result=-0.0266 [13.9 s]
[2] train-result=0.2441, valid-result=0.2428 [12.4 s]
[3] train-result=0.2744, valid-result=0.2502 [12.5 s]
[4] train-result=0.2723, valid-result=0.2561 [12.7 s]
[5] train-result=0.2723, valid-result=0.2541 [12.0 s]
[6] train-result=0.2746, valid-result=0.2572 [11.5 s]
[7] train-result=0.2753, valid-result=0.2566 [13.0 s]
[8] train-result=0.2766, valid-result=0.2567 [12.2 s]
[9] train-result=0.2800, valid-result=0.2587 [11.8 s]
[10] train-result=0.2801, valid-result=0.2591 [11.8 s]
[11] train-result=0.2816, valid-result=0.2594 [12.0 s]
[12] train-result=0.2821, valid-result=0.2604 [12.7 s]
[13] train-result=0.2829, valid-result=0.2605 [11.4 s]
[14] train-result=0.2836, valid-result=0.2611 [11.5 s]
[15] train-result=0.2837, valid-result=0.2608 [11.6 s]
[16] train-result=0.2842, valid-result=0.2611 [11.4 s]
[17] train-result=0.2844, valid-result=0.2613 [11.6 s]
[18] train-result=0.2845, valid-result=0.2610 [12.9 s]
[19] train-result=0.2847, valid-result=0.2612 [11.6 s]
[20] train-result=0.2847, valid-result=0.2609 [11.5 s]
[21] train-result=0.2845, valid-result=0.2613 [12.2 s]
[22] train-result=0.2848, valid-result=0.2617 [11.7 s]
[23] train-result=0.2851, valid-result=0.2614 [13.8 s]
[24] train-result=0.2852, valid-result=0.2620 [12.0 s]
[25] train-result=0.2849, valid-result=0.2613 [11.8 s]
[26] train-result=0.2853, valid-result=0.2611 [11.7 s]
[27] train-result=0.2853, valid-result=0.2621 [11.7 s]
[28] train-result=0.2855, valid-result=0.2617 [11.7 s]
[29] train-result=0.2855, valid-result=0.2615 [13.1 s]
[30] train-result=0.2856, valid-result=0.2614 [11.7 s]
DeepFM: 0.26753 (0.00431)

#params: 13451
[1] train-result=0.2479, valid-result=0.2485 [9.7 s]
[2] train-result=0.2646, valid-result=0.2613 [10.9 s]
[3] train-result=0.2694, valid-result=0.2650 [10.0 s]
[4] train-result=0.2730, valid-result=0.2675 [9.6 s]
[5] train-result=0.2750, valid-result=0.2688 [9.7 s]
[6] train-result=0.2764, valid-result=0.2688 [9.9 s]
[7] train-result=0.2772, valid-result=0.2701 [11.4 s]
[8] train-result=0.2788, valid-result=0.2705 [9.9 s]
[9] train-result=0.2785, valid-result=0.2704 [10.1 s]
[10] train-result=0.2787, valid-result=0.2708 [9.1 s]
[11] train-result=0.2797, valid-result=0.2704 [9.7 s]
[12] train-result=0.2803, valid-result=0.2700 [9.6 s]
[13] train-result=0.2806, valid-result=0.2701 [11.0 s]
[14] train-result=0.2808, valid-result=0.2711 [9.1 s]
[15] train-result=0.2805, valid-result=0.2719 [10.0 s]
[16] train-result=0.2816, valid-result=0.2712 [9.7 s]
[17] train-result=0.2816, valid-result=0.2719 [9.4 s]
[18] train-result=0.2817, valid-result=0.2725 [11.3 s]
[19] train-result=0.2826, valid-result=0.2725 [9.5 s]
[20] train-result=0.2833, valid-result=0.2725 [9.8 s]
[21] train-result=0.2834, valid-result=0.2721 [10.1 s]
[22] train-result=0.2848, valid-result=0.2732 [10.1 s]
[23] train-result=0.2858, valid-result=0.2743 [9.9 s]
[24] train-result=0.2876, valid-result=0.2740 [10.6 s]
[25] train-result=0.2890, valid-result=0.2733 [9.5 s]
[26] train-result=0.2896, valid-result=0.2754 [9.1 s]
[27] train-result=0.2914, valid-result=0.2749 [9.7 s]
[28] train-result=0.2932, valid-result=0.2744 [10.1 s]
[29] train-result=0.2950, valid-result=0.2739 [10.8 s]
[30] train-result=0.2977, valid-result=0.2767 [9.9 s]
#params: 13451
[1] train-result=0.2397, valid-result=0.2295 [10.0 s]
[2] train-result=0.2636, valid-result=0.2506 [9.8 s]
[3] train-result=0.2686, valid-result=0.2569 [10.1 s]
[4] train-result=0.2733, valid-result=0.2607 [11.4 s]
[5] train-result=0.2749, valid-result=0.2626 [9.7 s]
[6] train-result=0.2765, valid-result=0.2642 [10.1 s]
[7] train-result=0.2774, valid-result=0.2657 [9.6 s]
[8] train-result=0.2782, valid-result=0.2674 [9.9 s]
[9] train-result=0.2787, valid-result=0.2672 [10.4 s]
[10] train-result=0.2793, valid-result=0.2671 [9.7 s]
[11] train-result=0.2791, valid-result=0.2680 [9.5 s]
[12] train-result=0.2796, valid-result=0.2684 [9.0 s]
[13] train-result=0.2794, valid-result=0.2678 [9.9 s]
[14] train-result=0.2803, valid-result=0.2685 [9.1 s]
[15] train-result=0.2801, valid-result=0.2680 [10.9 s]
[16] train-result=0.2806, valid-result=0.2695 [9.8 s]
[17] train-result=0.2812, valid-result=0.2697 [9.8 s]
[18] train-result=0.2815, valid-result=0.2684 [9.6 s]
[19] train-result=0.2816, valid-result=0.2695 [9.6 s]
[20] train-result=0.2831, valid-result=0.2703 [10.9 s]
[21] train-result=0.2853, valid-result=0.2709 [9.7 s]
[22] train-result=0.2875, valid-result=0.2706 [9.2 s]
[23] train-result=0.2885, valid-result=0.2708 [9.1 s]
[24] train-result=0.2893, valid-result=0.2699 [9.7 s]
[25] train-result=0.2896, valid-result=0.2710 [9.4 s]
[26] train-result=0.2915, valid-result=0.2704 [10.4 s]
[27] train-result=0.2932, valid-result=0.2688 [9.0 s]
[28] train-result=0.2938, valid-result=0.2715 [9.0 s]
[29] train-result=0.2965, valid-result=0.2718 [9.1 s]
[30] train-result=0.2961, valid-result=0.2692 [9.8 s]
#params: 13451
[1] train-result=0.2335, valid-result=0.2223 [9.3 s]
[2] train-result=0.2600, valid-result=0.2425 [9.8 s]
[3] train-result=0.2696, valid-result=0.2494 [9.3 s]
[4] train-result=0.2751, valid-result=0.2532 [9.9 s]
[5] train-result=0.2781, valid-result=0.2552 [9.4 s]
[6] train-result=0.2813, valid-result=0.2583 [10.2 s]
[7] train-result=0.2825, valid-result=0.2596 [9.4 s]
[8] train-result=0.2836, valid-result=0.2604 [9.8 s]
[9] train-result=0.2841, valid-result=0.2609 [9.7 s]
[10] train-result=0.2847, valid-result=0.2602 [8.9 s]
[11] train-result=0.2846, valid-result=0.2611 [10.6 s]
[12] train-result=0.2848, valid-result=0.2610 [8.9 s]
[13] train-result=0.2854, valid-result=0.2617 [9.0 s]
[14] train-result=0.2852, valid-result=0.2614 [9.0 s]
[15] train-result=0.2854, valid-result=0.2618 [9.5 s]
[16] train-result=0.2854, valid-result=0.2624 [9.5 s]
[17] train-result=0.2857, valid-result=0.2611 [10.8 s]
[18] train-result=0.2858, valid-result=0.2615 [9.5 s]
[19] train-result=0.2859, valid-result=0.2616 [10.0 s]
[20] train-result=0.2860, valid-result=0.2616 [9.3 s]
[21] train-result=0.2877, valid-result=0.2628 [10.1 s]
[22] train-result=0.2910, valid-result=0.2636 [9.2 s]
[23] train-result=0.2917, valid-result=0.2636 [10.2 s]
[24] train-result=0.2912, valid-result=0.2607 [9.0 s]
[25] train-result=0.2938, valid-result=0.2636 [9.7 s]
[26] train-result=0.2942, valid-result=0.2633 [9.7 s]
[27] train-result=0.2943, valid-result=0.2630 [10.1 s]
[28] train-result=0.2944, valid-result=0.2630 [10.9 s]
[29] train-result=0.2959, valid-result=0.2620 [9.5 s]
[30] train-result=0.2954, valid-result=0.2619 [9.6 s]
FM: 0.26927 (0.00607)

#params: 13436
[1] train-result=-0.0333, valid-result=0.0096 [11.4 s]
[2] train-result=0.0446, valid-result=0.0529 [12.9 s]
[3] train-result=0.0639, valid-result=0.0766 [11.5 s]
[4] train-result=0.1272, valid-result=0.1024 [11.5 s]
[5] train-result=0.1414, valid-result=0.1055 [11.7 s]
[6] train-result=0.1279, valid-result=0.1000 [11.4 s]
[7] train-result=0.1457, valid-result=0.1188 [11.7 s]
[8] train-result=0.2061, valid-result=0.1793 [12.9 s]
[9] train-result=0.1677, valid-result=0.1347 [11.7 s]
[10] train-result=0.1612, valid-result=0.1376 [11.7 s]
[11] train-result=0.2260, valid-result=0.2019 [11.5 s]
[12] train-result=0.1751, valid-result=0.1346 [11.1 s]
[13] train-result=0.2566, valid-result=0.2289 [12.6 s]
[14] train-result=0.2670, valid-result=0.2467 [11.2 s]
[15] train-result=0.2576, valid-result=0.2522 [11.5 s]
[16] train-result=0.2573, valid-result=0.2502 [11.3 s]
[17] train-result=0.2713, valid-result=0.2553 [11.2 s]
[18] train-result=0.2645, valid-result=0.2492 [11.1 s]
[19] train-result=0.2823, valid-result=0.2688 [12.5 s]
[20] train-result=0.2795, valid-result=0.2704 [11.3 s]
[21] train-result=0.2798, valid-result=0.2681 [11.7 s]
[22] train-result=0.2775, valid-result=0.2660 [11.0 s]
[23] train-result=0.2835, valid-result=0.2693 [11.1 s]
[24] train-result=0.2826, valid-result=0.2697 [12.9 s]
[25] train-result=0.2828, valid-result=0.2709 [10.8 s]
[26] train-result=0.2819, valid-result=0.2666 [11.5 s]
[27] train-result=0.2786, valid-result=0.2660 [11.6 s]
[28] train-result=0.2808, valid-result=0.2719 [11.5 s]
[29] train-result=0.2831, valid-result=0.2694 [11.7 s]
[30] train-result=0.2822, valid-result=0.2703 [12.3 s]
#params: 13436
[1] train-result=0.0263, valid-result=0.0468 [11.2 s]
[2] train-result=0.1878, valid-result=0.2303 [11.0 s]
[3] train-result=0.2365, valid-result=0.2426 [10.9 s]
[4] train-result=0.2284, valid-result=0.2321 [12.6 s]
[5] train-result=0.2208, valid-result=0.2299 [11.5 s]
[6] train-result=0.2365, valid-result=0.2261 [11.2 s]
[7] train-result=0.2515, valid-result=0.2373 [11.5 s]
[8] train-result=0.2295, valid-result=0.2224 [10.8 s]
[9] train-result=0.2398, valid-result=0.2410 [11.4 s]
[10] train-result=0.2558, valid-result=0.2539 [12.9 s]
[11] train-result=0.2578, valid-result=0.2525 [11.4 s]
[12] train-result=0.2582, valid-result=0.2488 [11.6 s]
[13] train-result=0.2492, valid-result=0.2540 [11.2 s]
[14] train-result=0.2655, valid-result=0.2557 [11.7 s]
[15] train-result=0.2741, valid-result=0.2605 [12.3 s]
[16] train-result=0.2668, valid-result=0.2510 [11.5 s]
[17] train-result=0.2631, valid-result=0.2541 [11.1 s]
[18] train-result=0.2776, valid-result=0.2644 [11.5 s]
[19] train-result=0.2739, valid-result=0.2620 [11.8 s]
[20] train-result=0.2809, valid-result=0.2615 [11.1 s]
[21] train-result=0.2752, valid-result=0.2653 [12.5 s]
[22] train-result=0.2806, valid-result=0.2681 [11.0 s]
[23] train-result=0.2791, valid-result=0.2644 [11.5 s]
[24] train-result=0.2790, valid-result=0.2677 [11.3 s]
[25] train-result=0.2803, valid-result=0.2634 [11.3 s]
[26] train-result=0.2781, valid-result=0.2679 [12.3 s]
[27] train-result=0.2798, valid-result=0.2629 [11.2 s]
[28] train-result=0.2800, valid-result=0.2665 [11.4 s]
[29] train-result=0.2767, valid-result=0.2652 [11.1 s]
[30] train-result=0.2826, valid-result=0.2674 [11.5 s]
#params: 13436
[1] train-result=-0.0914, valid-result=-0.0736 [12.6 s]
[2] train-result=0.0825, valid-result=0.1460 [11.1 s]
[3] train-result=0.1831, valid-result=0.1728 [10.8 s]
[4] train-result=0.1506, valid-result=0.1436 [11.0 s]
[5] train-result=0.1385, valid-result=0.1440 [11.2 s]
[6] train-result=0.1579, valid-result=0.1478 [12.4 s]
[7] train-result=0.2130, valid-result=0.2043 [10.8 s]
[8] train-result=0.2439, valid-result=0.2317 [10.7 s]
[9] train-result=0.2195, valid-result=0.2110 [10.9 s]
[10] train-result=0.2519, valid-result=0.2322 [11.0 s]
[11] train-result=0.2486, valid-result=0.2354 [11.0 s]
[12] train-result=0.2657, valid-result=0.2444 [11.8 s]
[13] train-result=0.2543, valid-result=0.2376 [11.0 s]
[14] train-result=0.2701, valid-result=0.2445 [11.4 s]
[15] train-result=0.2753, valid-result=0.2564 [10.7 s]
[16] train-result=0.2776, valid-result=0.2594 [11.5 s]
[17] train-result=0.2643, valid-result=0.2542 [12.3 s]
[18] train-result=0.2778, valid-result=0.2611 [11.2 s]
[19] train-result=0.2796, valid-result=0.2643 [11.2 s]
[20] train-result=0.2831, valid-result=0.2604 [11.1 s]
[21] train-result=0.2820, valid-result=0.2608 [10.9 s]
[22] train-result=0.2841, valid-result=0.2609 [11.3 s]
[23] train-result=0.2813, valid-result=0.2566 [12.7 s]
[24] train-result=0.2861, valid-result=0.2609 [11.3 s]
[25] train-result=0.2843, valid-result=0.2636 [10.6 s]
[26] train-result=0.2849, valid-result=0.2591 [11.8 s]
[27] train-result=0.2815, valid-result=0.2619 [11.5 s]
[28] train-result=0.2850, valid-result=0.2614 [11.3 s]
[29] train-result=0.2866, valid-result=0.2570 [12.5 s]
[30] train-result=0.2818, valid-result=0.2599 [11.3 s]
DNN: 0.26585 (0.00445)

好像没有明显提升

[washujiang]

train-result 从0.6663变成了0.7175，valid-result 从0.4569变到了0.4378，不过AUC从0.73变到了0.72