成轴对称的两组图片能被分成两类吗？

将吸引子，排斥子，鞍点和反鞍点旋转90度形成轴对称，并将成轴对称的两组图片进行分类。

r1	r2
<1	<1	吸引子	c
>1	>1	排斥子	p
>1	<1	鞍点	a
<1	>1	反鞍点	fa

r1	r2
<1	<1		90c
>1	>1		90p
>1	<1		90a
<1	>1		90fa

 

做了10个二分类网络来验证。

d2（c，c）-4-4-2-（2*k）,k∈{0,1}

d2（c，p）-4-4-2-（2*k）,k∈{0,1}

d2（c，a）-4-4-2-（2*k）,k∈{0,1}

d2（c，fa）-4-4-2-（2*k）,k∈{0,1}

d2（p，p）-4-4-2-（2*k）,k∈{0,1}

d2（p，a）-4-4-2-（2*k）,k∈{0,1}

d2（p，fa）-4-4-2-（2*k）,k∈{0,1}

d2（a，a）-4-4-2-（2*k）,k∈{0,1}

d2（a，fa）-4-4-2-（2*k）,k∈{0,1}

d2（fa，fa）-4-4-2-（2*k）,k∈{0,1}

实验过程

以二分类吸引子和90c为例子

d2（c，90c）-4-4-2-（2*k）,k∈{0,1}

制作一个4*4*2的网络向这个的左侧输入吸引子C，并让左侧网络向1，0收敛；向右侧网络输入90C让右侧向0，1收敛，并让4*4*2部分权重共享，前面大量实验表明这种效果相当于将两个弹性系数为k1，k2的弹簧并联成一个弹性系数为k的弹簧，并且让k1=k2=k/2的过程。

 

这个网络的收敛标准是

if (Math.abs(f2[0]-y[0])< δ  &&  Math.abs(f2[1]-y[1])< δ   )

因为对应每个收敛标准δ都有一个特征的迭代次数n与之对应因此可以用迭代次数曲线n(δ)来评价网络性能。

本文尝试了δ从0.5到1e-6在内的26个值.

具体进样顺序
进样顺序	迭代次数
δ=0.5
c	1	判断是否达到收敛
90c	2	判断是否达到收敛
梯度下降
c	3	判断是否达到收敛
90c	4	判断是否达到收敛
梯度下降
……
达到收敛标准测量准确率,记录迭代次数，将这个过程重复199次
δ=0.4
…
δ=1e-6

 

将这个网络简写成

d2（c，90c）-4-4-2-（2*k）,k∈{0,1}

得到的数据

c90c
测试集	0.3	c
	0.7	90c
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.526543	0.520987	131.8744	0.684383	0.5	1.864322	404	0.006733	1
0.494236	0.506314	408.5628	0.998521	0.4	2.045226	407	0.006783	1
0.507004	0.492747	523.8543	1	0.3	2.01005	401	0.006683	1
0.52538	0.473589	651.4573	1	0.2	2.336683	465	0.00775	1
0.51796	0.48164	892.1508	1	0.1	2.798995	557	0.009283	1
0.507406	0.492649	3041.864	1	0.01	6.477387	1290	0.0215	1
0.522571	0.477433	18666.41	1	0.001	35.43719	7052	0.117533	1
0.522609	0.477391	154933	1	1.00E-04	265.5126	52869	0.88115	1
0.482415	0.517585	173681.9	1	9.00E-05	297.5528	59214	0.9869	1
0.507537	0.492464	193083.9	1	8.00E-05	326.7789	65029	1.083817	1
0.532659	0.467341	215305.9	1	7.00E-05	366.603	72954	1.2159	1
0.497488	0.502512	250412.3	1	6.00E-05	423.9648	84369	1.40615	1
0.447241	0.552759	301038.5	1	5.00E-05	510.4573	101597	1.693283	1
0.472364	0.527636	370228.5	1	4.00E-05	629.1307	125199	2.08665	1
0.487438	0.512562	490387.4	1	3.00E-05	842.4221	167658	2.7943	1
0.427139	0.572861	719951.8	1	2.00E-05	1190.337	236879	3.947983	1
0.522613	0.477387	1409685	1	1.00E-05	2291.794	456068	7.601133	1
0.487437	0.512563	1562235	1	9.00E-06	2540.839	505634	8.427233	1
0.527638	0.472362	1749759	1	8.00E-06	2293.593	456428	7.607133	1
0.517588	0.482412	2010267	1	7.00E-06	3270.432	650820	10.847	1
0.502513	0.497487	2372571	1	6.00E-06	3859.116	767968	12.79947	1
0.532663	0.467337	2831875	1	5.00E-06	4602.578	915919	15.26532	1
0.457287	0.542713	3497169	1	4.00E-06	5679.698	1130263	18.83772	1
0.522613	0.477387	4633150	1	3.00E-06	7525.598	1497601	24.96002	1
0.552764	0.447236	6887346	1	2.00E-06	11182.28	2225280	37.088	1
0.517588	0.482412	1.39E+07	1	1.00E-06	22662.01	4509744	75.1624	1

平均准确率p-ave，是199次收敛的平均值

最大准确率p-max，是199次收敛的最大值

从数据观察到p-ave和p-max都很高甚至等于1，表明c和90c是可以分类的。

将所有10组数据汇总

迭代次数n
δ	c90c	c90p	c90a	c90fa	p90p	p90a	p90fa	a90a	a90fa	fa90fa
0.5	131.8744	114.8844	127.3116	126.5427	113.0302	115.402	116.2111	123.3015	119.8241	119.7337
0.4	408.5628	346.5879	372.5075	372.8593	306.1457	322.5377	328.7186	340.8241	337.8693	353.1307
0.3	523.8543	460.8191	497.2111	489.7588	405.4673	430.6935	431.5075	470.3869	457.6633	464.4573
0.2	651.4573	591.0352	608.3116	618.9146	524.1809	543.9648	550.2864	571.392	577.3618	574.5779
0.1	892.1508	791.5779	835.1558	839.5578	732.9849	759.8442	766.8945	798.8643	791.5528	789.7437
0.01	3041.864	2997.925	2948.799	2995.382	3036.03	2971.085	3027.683	3043.523	3026.075	2992.181
0.001	18666.41	18982.63	19199.14	18697.54	20160.65	20141.81	19973.77	19763.44	19902.08	19734.79
1.00E-04	154933	155119	157139.3	160623.4	180285.8	169889.1	169733.2	173807.2	172645.8	170339.3
9.00E-05	173681.9	175880.2	178394.2	174586.2	197241.3	191301	191646.1	186127.7	185716.9	193358.2
8.00E-05	193083.9	194356.2	192181.4	194583.5	220316.1	216489.9	211761.2	212362.1	211638.4	210133.1
7.00E-05	215305.9	217283.1	221122.4	220790.2	250422	244110.6	241865.8	240375.8	243420.3	241907
6.00E-05	250412.3	255396.8	253899.3	248430.4	296003.7	280588.5	284804.3	282585.6	280313	276559.3
5.00E-05	301038.5	310982.8	306293.8	307819.2	352459	332682.3	332686.2	331453.8	335069.5	331354.3
4.00E-05	370228.5	372387.4	377779	373825.7	434223.6	419922.5	412651.2	408452.5	416923.9	412179.2
3.00E-05	490387.4	491419.8	508109.6	486019.3	575105.8	555232.1	558235.3	554077.2	544859.7	547005
2.00E-05	719951.8	741632.8	743942.2	729207.4	851691.7	809831.7	827564.3	813810	809301.8	818629.8
1.00E-05	1409685	1443639	1469048	1455749	1673573	1611146	1615672	1605139	1616243	1610707
9.00E-06	1562235	1584642	1626100	1602979	1904267	1805135	1796785	1769139	1754074	1764101
8.00E-06	1749759	1776924	1811294	1796519	2102945	2009691	2009703	1989280	1963808	1985273
7.00E-06	2010267	2053422	2050795	2058895	2375729	2295291	2279402	2279703	2284711	2282847
6.00E-06	2372571	2393968	2356225	2428483	2798351	2652808	2698319	2645305	2631248	2661392
5.00E-06	2831875	2896488	2820220	2825876	3399647	3200305	3199980	3189939	3173364	3176851
4.00E-06	3497169	3516782	3535259	3516984	4242944	3989938	3929552	3925776	3927128	3967423
3.00E-06	4633150	4740260	4748654	4750814	5535770	5362736	5327771	5185330	5237892	5191960
2.00E-06	6887346	6930461	7105182	6972852	8327463	7906124	7959008	7860425	7909708	7753992
1.00E-06	1.39E+07	1.40E+07	1.40E+07	1.38E+07	1.67E+07	1.57E+07	1.56E+07	1.58E+07	1.59E+07	1.58E+07

表明这10个网络的迭代次数曲线是差不多的。

平均准确率p-ave
δ	c90c	c90p	c90a	c90fa	p90p	p90a	p90fa	a90a	a90fa	fa90fa
0.5	0.684383	0.767883	0.784805	0.759302	0.692924	0.629695	0.671631	0.730786	0.732049	0.711712
0.4	0.998521	0.989417	0.997012	0.993662	0.98996	0.987832	0.987797	0.986947	0.990221	0.995493
0.3	1	1	1	1	1	1	1	1	1	1
0.2	1	1	1	1	1	1	1	1	1	1
0.1	1	1	1	1	1	1	1	1	1	1
0.01	1	1	1	1	1	1	1	1	1	1
0.001	1	1	1	1	1	1	1	1	1	1
1.00E-04	1	1	1	1	1	1	1	1	1	1
9.00E-05	1	1	1	1	1	1	1	1	1	1
8.00E-05	1	1	1	1	1	1	1	1	1	1
7.00E-05	1	1	1	1	1	1	1	1	1	1
6.00E-05	1	1	1	1	1	1	1	1	1	1
5.00E-05	1	1	1	1	1	1	1	1	1	1
4.00E-05	1	1	1	1	1	1	1	1	1	1
3.00E-05	1	1	1	1	1	1	1	1	1	1
2.00E-05	1	1	1	1	1	1	1	1	1	1
1.00E-05	1	1	1	1	1	1	1	1	1	1
9.00E-06	1	1	1	1	1	1	1	1	1	1
8.00E-06	1	1	1	1	1	1	1	1	1	1
7.00E-06	1	1	1	1	1	1	1	1	1	1
6.00E-06	1	1	1	1	1	1	1	1	1	1
5.00E-06	1	1	1	1	1	1	1	1	1	1
4.00E-06	1	1	1	1	1	1	1	1	1	1
3.00E-06	1	1	1	1	1	1	1	1	1	1
2.00E-06	1	1	1	1	1	1	1	1	1	1
1.00E-06	1	1	1	1	1	1	1	1	1	1

 

而且这个10个网络都可以实现很高的分辨率。

表明成轴对称的两组图片有被分为两类的可能。

实验参数

学习率 0.1

权重初始化方式

Random rand1 =new Random();

int ti1=rand1.nextInt(98)+1;

tw[a][b]=xx*((double)ti1/100);

 

c90c
测试集	0.3	c
	0.7	90c
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.526543	0.520987	131.8744	0.684383	0.5	1.864322	404	0.006733	1
0.494236	0.506314	408.5628	0.998521	0.4	2.045226	407	0.006783	1
0.507004	0.492747	523.8543	1	0.3	2.01005	401	0.006683	1
0.52538	0.473589	651.4573	1	0.2	2.336683	465	0.00775	1
0.51796	0.48164	892.1508	1	0.1	2.798995	557	0.009283	1
0.507406	0.492649	3041.864	1	0.01	6.477387	1290	0.0215	1
0.522571	0.477433	18666.41	1	0.001	35.43719	7052	0.117533	1
0.522609	0.477391	154933	1	1.00E-04	265.5126	52869	0.88115	1
0.482415	0.517585	173681.9	1	9.00E-05	297.5528	59214	0.9869	1
0.507537	0.492464	193083.9	1	8.00E-05	326.7789	65029	1.083817	1
0.532659	0.467341	215305.9	1	7.00E-05	366.603	72954	1.2159	1
0.497488	0.502512	250412.3	1	6.00E-05	423.9648	84369	1.40615	1
0.447241	0.552759	301038.5	1	5.00E-05	510.4573	101597	1.693283	1
0.472364	0.527636	370228.5	1	4.00E-05	629.1307	125199	2.08665	1
0.487438	0.512562	490387.4	1	3.00E-05	842.4221	167658	2.7943	1
0.427139	0.572861	719951.8	1	2.00E-05	1190.337	236879	3.947983	1
0.522613	0.477387	1409685	1	1.00E-05	2291.794	456068	7.601133	1
0.487437	0.512563	1562235	1	9.00E-06	2540.839	505634	8.427233	1
0.527638	0.472362	1749759	1	8.00E-06	2293.593	456428	7.607133	1
0.517588	0.482412	2010267	1	7.00E-06	3270.432	650820	10.847	1
0.502513	0.497487	2372571	1	6.00E-06	3859.116	767968	12.79947	1
0.532663	0.467337	2831875	1	5.00E-06	4602.578	915919	15.26532	1
0.457287	0.542713	3497169	1	4.00E-06	5679.698	1130263	18.83772	1
0.522613	0.477387	4633150	1	3.00E-06	7525.598	1497601	24.96002	1
0.552764	0.447236	6887346	1	2.00E-06	11182.28	2225280	37.088	1
0.517588	0.482412	1.39E+07	1	1.00E-06	22662.01	4509744	75.1624	1

 

c90p
测试集	0.3	c
	0.7	90p
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.508727	0.552175	114.8844	0.767883	0.5	2.065327	416	0.006933	1
0.444757	0.556683	346.5879	0.989417	0.4	2.090452	418	0.006967	1
0.414325	0.586769	460.8191	1	0.3	2.366834	475	0.007917	1
0.423193	0.581004	591.0352	1	0.2	2.020101	407	0.006783	1
0.494651	0.505552	791.5779	1	0.1	2.341709	469	0.007817	1
0.704623	0.295447	2997.925	1	0.01	6.075377	1216	0.020267	1
0.783366	0.216635	18982.63	1	0.001	33.75879	6727	0.112117	1
0.808984	0.191016	155119	1	1.00E-04	251.9749	50153	0.835883	1
0.80899	0.19101	175880.2	1	9.00E-05	286.1106	56942	0.949033	1
0.783875	0.216125	194356.2	1	8.00E-05	314.608	62613	1.04355	1
0.819051	0.180949	217283.1	1	7.00E-05	352.0653	70068	1.1678	1
0.814033	0.185967	255396.8	1	6.00E-05	413.3769	82270	1.371167	1
0.773842	0.226158	310982.8	1	5.00E-05	503.2764	100155	1.66925	1
0.844194	0.155806	372387.4	1	4.00E-05	602.0854	119824	1.997067	1
0.839176	0.160824	491419.8	1	3.00E-05	794.9045	158194	2.636567	1
0.773859	0.226141	741632.8	1	2.00E-05	1202.442	239294	3.988233	1
0.844214	0.155786	1443639	1	1.00E-05	2355.065	468660	7.811	1
0.864315	0.135685	1584642	1	9.00E-06	2568.543	511145	8.519083	1
0.839191	0.160809	1776924	1	8.00E-06	2878.462	572821	9.547017	1
0.844216	0.155784	2053422	1	7.00E-06	3326.492	661977	11.03295	1
0.854267	0.145733	2393968	1	6.00E-06	3884.492	773017	12.88362	1
0.773867	0.226133	2896488	1	5.00E-06	4698.794	935063	15.58438	1
0.788942	0.211058	3516782	1	4.00E-06	5902.538	1174605	19.57675	1
0.834169	0.165831	4740260	1	3.00E-06	8036.573	1599279	26.65465	1
0.824119	0.175881	6930461	1	2.00E-06	11170.96	2223025	37.05042	1
0.804019	0.195981	1.40E+07	1	1.00E-06	24117.63	4799409	79.99015	1

 

c90a
测试集	0.3	c
	0.7	90a
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.512295	0.538503	127.3116	0.784805	0.5	1.934673	400	0.006667	1
0.487975	0.516048	372.5075	0.997012	0.4	1.974874	393	0.00655	1
0.459986	0.540066	497.2111	1	0.3	2.115578	423	0.00705	1
0.426574	0.57352	608.3116	1	0.2	2.165829	431	0.007183	1
0.568285	0.431977	835.1558	1	0.1	2.487437	495	0.00825	1
0.753884	0.246186	2948.799	1	0.01	6.311558	1256	0.020933	1
0.798414	0.201591	19199.14	1	0.001	36.51256	7281	0.12135	1
0.814008	0.185992	157139.3	1	1.00E-04	265.4221	52820	0.880333	1
0.758748	0.241252	178394.2	1	9.00E-05	303.5628	60411	1.00685	1
0.803972	0.196028	192181.4	1	8.00E-05	326.9296	65060	1.084333	1
0.814027	0.185973	221122.4	1	7.00E-05	376.0553	74851	1.247517	1
0.778861	0.221139	253899.3	1	6.00E-05	429.3518	85441	1.424017	1
0.793941	0.206059	306293.8	1	5.00E-05	521.5829	103796	1.729933	1
0.773848	0.226152	377779	1	4.00E-05	639.392	127255	2.120917	1
0.773853	0.226147	508109.6	1	3.00E-05	857.4774	170638	2.843967	1
0.798983	0.201017	743942.2	1	2.00E-05	1249.342	248619	4.14365	1
0.778889	0.221111	1469048	1	1.00E-05	2461.035	489746	8.162433	1
0.793965	0.206035	1626100	1	9.00E-06	2765.744	550387	9.173117	1
0.804015	0.195985	1811294	1	8.00E-06	3106.276	618149	10.30248	1
0.814066	0.185934	2050795	1	7.00E-06	3405.111	677633	11.29388	1
0.804016	0.195984	2356225	1	6.00E-06	4025.362	801048	13.3508	1
0.748741	0.251259	2820220	1	5.00E-06	4829.372	961047	16.01745	1
0.809043	0.190957	3535259	1	4.00E-06	6056.794	1205302	20.08837	1
0.793968	0.206032	4748654	1	3.00E-06	8096.116	1611143	26.85238	1
0.804019	0.195981	7105182	1	2.00E-06	12078.41	2403605	40.06008	1
0.778894	0.221106	1.40E+07	1	1.00E-06	23854.79	4747104	79.1184	1

 

 

c90fa
测试集	0.3	c
	0.7	90fa
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.513475	0.539712	126.5427	0.759302	0.5	2.266332	451	0.007517	1
0.472083	0.532023	372.8593	0.993662	0.4	2.276382	455	0.007583	1
0.458539	0.543047	489.7588	1	0.3	2.668342	532	0.008867	1
0.484899	0.516832	618.9146	1	0.2	2.231156	447	0.00745	1
0.495018	0.50587	839.5578	1	0.1	2.487437	498	0.0083	1
0.694713	0.305296	2995.382	1	0.01	6.417085	1279	0.021317	1
0.748257	0.251742	18697.54	1	0.001	34.39196	6849	0.11415	1
0.728598	0.271402	160623.4	1	1.00E-04	278.4824	55422	0.9237	1
0.758748	0.241252	174586.2	1	9.00E-05	289.4925	57614	0.960233	1
0.753729	0.246271	194583.5	1	8.00E-05	326.6583	65010	1.0835	1
0.809003	0.190998	220790.2	1	7.00E-05	359.1407	71473	1.191217	1
0.814033	0.185967	248430.4	1	6.00E-05	423.2563	84230	1.403833	1
0.773842	0.226158	307819.2	1	5.00E-05	526.5176	104782	1.746367	1
0.824095	0.175905	373825.7	1	4.00E-05	636.392	126661	2.111017	1
0.814052	0.185948	486019.3	1	3.00E-05	835.5477	166276	2.771267	1
0.798983	0.201017	729207.4	1	2.00E-05	1260.789	250897	4.181617	1
0.829139	0.170861	1455749	1	1.00E-05	2541.271	505714	8.428567	1
0.804015	0.195985	1602979	1	9.00E-06	2751.663	547586	9.126433	1
0.814065	0.185935	1796519	1	8.00E-06	2953.281	587706	9.7951	1
0.72864	0.27136	2058895	1	7.00E-06	3384.136	673457	11.22428	1
0.778891	0.221109	2428483	1	6.00E-06	4049.347	805825	13.43042	1
0.798992	0.201008	2825876	1	5.00E-06	4701.688	935640	15.594	1
0.829143	0.170857	3516984	1	4.00E-06	5864.769	1167123	19.45205	1
0.839194	0.160806	4750814	1	3.00E-06	8122.291	1616339	26.93898	1
0.778893	0.221107	6972852	1	2.00E-06	12089.77	2405866	40.09777	1
0.783919	0.216081	1.38E+07	1	1.00E-06	23456.4	4667829	77.79715	1

 

p90p
测试集	0.3	p
	0.7	90p
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.529785	0.529507	113.0302	0.692924	0.5	2.090452	433	0.007217	1
0.504399	0.500345	306.1457	0.98996	0.4	1.98995	396	0.0066	1
0.491758	0.505517	405.4673	1	0.3	2.281407	454	0.007567	1
0.514413	0.4867	524.1809	1	0.2	2.175879	439	0.007317	1
0.510086	0.490508	732.9849	1	0.1	2.331658	465	0.00775	1
0.502471	0.497527	3036.03	1	0.01	6.366834	1269	0.02115	1
0.502509	0.497494	20160.65	1	0.001	37.86432	7536	0.1256	1
0.507536	0.492464	180285.8	1	1.00E-04	307.8693	61268	1.021133	1
0.427149	0.572851	197241.3	1	9.00E-05	337.191	67101	1.11835	1
0.522609	0.477391	220316.1	1	8.00E-05	378.4724	75317	1.255283	1
0.472366	0.527634	250422	1	7.00E-05	431.3417	85855	1.430917	1
0.527635	0.472365	296003.7	1	6.00E-05	507.1256	100944	1.6824	1
0.497488	0.502512	352459	1	5.00E-05	607.0553	120807	2.01345	1
0.467339	0.532661	434223.6	1	4.00E-05	747.4774	148753	2.479217	1
0.497488	0.502512	575105.8	1	3.00E-05	-59.0804	-11755	-0.19592	1
0.552762	0.447238	851691.7	1	2.00E-05	1447.693	288093	4.80155	1
0.522613	0.477387	1673573	1	1.00E-05	2835.04	564174	9.4029	1
0.557788	0.442212	1904267	1	9.00E-06	3347.92	666244	11.10407	1
0.487437	0.512563	2102945	1	8.00E-06	3698.633	736028	12.26713	1
0.502513	0.497487	2375729	1	7.00E-06	4176.653	831154	13.85257	1
0.517588	0.482412	2798351	1	6.00E-06	4739.538	943168	15.71947	1
0.492462	0.507538	3399647	1	5.00E-06	5693.402	1132990	18.88317	1
0.537688	0.462312	4242944	1	4.00E-06	7055.603	1404072	23.4012	1
0.512563	0.487437	5535770	1	3.00E-06	9299.347	1850572	30.84287	1
0.507538	0.492462	8327463	1	2.00E-06	14069.05	2799774	46.6629	1
0.497487	0.502513	1.67E+07	1	1.00E-06	27518.72	5476225	91.27042	1

 

 

p90a
测试集	0.3	p
	0.7	90a
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.537917	0.520255	115.402	0.629695	0.5	2.040201	406	0.006767	1
0.541042	0.462075	322.5377	0.987832	0.4	2.050251	408	0.0068	1
0.573857	0.426363	430.6935	1	0.3	2.361809	470	0.007833	1
0.547257	0.45223	543.9648	1	0.2	2.045226	407	0.006783	1
0.514366	0.486225	759.8442	1	0.1	2.361809	470	0.007833	1
0.34478	0.655264	2971.085	1	0.01	6.231156	1240	0.020667	1
0.367096	0.632905	20141.81	1	0.001	36.0201	7184	0.119733	1
0.28145	0.71855	169889.1	1	1.00E-04	284.2563	56567	0.942783	1
0.256325	0.743675	191301	1	9.00E-05	320.1106	63702	1.0617	1
0.271393	0.728607	216489.9	1	8.00E-05	361.3065	71916	1.1986	1
0.236218	0.763783	244110.6	1	7.00E-05	406.7387	80941	1.349017	1
0.266359	0.733641	280588.5	1	6.00E-05	464.4874	92433	1.54055	1
0.276404	0.723596	332682.3	1	5.00E-05	550.9849	109646	1.827433	1
0.246251	0.753749	419922.5	1	4.00E-05	694.5829	138238	2.303967	1
0.266346	0.733654	555232.1	1	3.00E-05	923.7638	183846	3.0641	1
0.251266	0.748734	809831.7	1	2.00E-05	1354.236	269493	4.49155	1
0.211061	0.788939	1611146	1	1.00E-05	2713.834	540069	9.00115	1
0.226136	0.773864	1805135	1	9.00E-06	2946.291	586312	9.771867	1
0.175885	0.824115	2009691	1	8.00E-06	3365.261	669687	11.16145	1
0.246235	0.753765	2295291	1	7.00E-06	3860.106	768162	12.8027	1
0.206034	0.793966	2652808	1	6.00E-06	4441.98	883969	14.73282	1
0.221108	0.778892	3200305	1	5.00E-06	4976.432	990310	16.50517	1
0.180907	0.819093	3989938	1	4.00E-06	6705.261	1334347	22.23912	1
0.201007	0.798993	5362736	1	3.00E-06	9015.276	1794040	29.90067	1
0.216082	0.783918	7906124	1	2.00E-06	13247.04	2636160	43.936	1
0.190955	0.809045	1.57E+07	1	1.00E-06	26395.33	5252671	87.54452	1

 

p90fa
测试集	0.3	p
	0.7	90fa
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.540787	0.517055	116.2111	0.671631	0.5	1.964824	391	0.006517	1
0.534404	0.467007	328.7186	0.987797	0.4	1.964824	391	0.006517	1
0.563833	0.434837	431.5075	1	0.3	2.326633	463	0.007717	1
0.529733	0.47056	550.2864	1	0.2	2.472362	492	0.0082	1
0.510557	0.491419	766.8945	1	0.1	2.366834	471	0.00785	1
0.517234	0.482795	3027.683	1	0.01	6.221106	1238	0.020633	1
0.321959	0.678042	19973.77	1	0.001	36.48744	7261	0.121017	1
0.206089	0.793912	169733.2	1	1.00E-04	284.0201	56520	0.942	1
0.306567	0.693433	191646.1	1	9.00E-05	319.5226	63585	1.05975	1
0.231198	0.768802	211761.2	1	8.00E-05	351.1055	69870	1.1645	1
0.256315	0.743685	241865.8	1	7.00E-05	403.608	80318	1.338633	1
0.226163	0.773837	284804.3	1	6.00E-05	476.7286	94869	1.58115	1
0.226158	0.773842	332686.2	1	5.00E-05	553.6633	110179	1.836317	1
0.261326	0.738675	412651.2	1	4.00E-05	695.8141	138467	2.307783	1
0.27137	0.72863	558235.3	1	3.00E-05	943.0352	187680	3.128	1
0.256291	0.743709	827564.3	1	2.00E-05	1401.121	278823	4.64705	1
0.266336	0.733664	1615672	1	1.00E-05	2732.467	543777	9.06295	1
0.276386	0.723614	1796785	1	9.00E-06	3031.291	603242	10.05403	1
0.23116	0.76884	2009703	1	8.00E-06	3311.884	659065	10.98442	1
0.286435	0.713565	2279402	1	7.00E-06	3852.291	766606	12.77677	1
0.211059	0.788941	2698319	1	6.00E-06	4552.648	905977	15.09962	1
0.246234	0.753766	3199980	1	5.00E-06	5331.804	1061060	17.68433	1
0.180907	0.819093	3929552	1	4.00E-06	6434.397	1280461	21.34102	1
0.201007	0.798993	5327771	1	3.00E-06	9241.382	1839051	30.65085	1
0.221107	0.778893	7959008	1	2.00E-06	13394.4	2665493	44.42488	1
0.18593	0.81407	1.56E+07	1	1.00E-06	26746.51	5322556	88.70927	1

 

a90a
测试集	0.3	a
	0.7	90a
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.525283	0.524947	123.3015	0.730786	0.5	1.929648	400	0.006667	1
0.498771	0.502565	340.8241	0.986947	0.4	2.015075	401	0.006683	1
0.523903	0.476954	470.3869	1	0.3	2.241206	446	0.007433	1
0.522427	0.47661	571.392	1	0.2	2.713568	556	0.009267	1
0.485956	0.514318	798.8643	1	0.1	2.592965	516	0.0086	1
0.527135	0.4729	3043.523	1	0.01	6.321608	1259	0.020983	1
0.522568	0.477432	19763.44	1	0.001	35.75377	7115	0.118583	1
0.502512	0.497488	173807.2	1	1.00E-04	292.7085	58250	0.970833	1
0.522609	0.477391	186127.7	1	9.00E-05	316.1709	62919	1.04865	1
0.462318	0.537683	212362.1	1	8.00E-05	361.6583	71970	1.1995	1
0.492463	0.507537	240375.8	1	7.00E-05	407.8593	81180	1.353	1
0.512561	0.487439	282585.6	1	6.00E-05	477.6834	95059	1.584317	1
0.46734	0.53266	331453.8	1	5.00E-05	561.1658	111687	1.86145	1
0.442216	0.557784	408452.5	1	4.00E-05	694.201	138146	2.302433	1
0.492463	0.507537	554077.2	1	3.00E-05	943.5025	187757	3.129283	1
0.462313	0.537687	813810	1	2.00E-05	1381.528	274940	4.582333	1
0.512563	0.487437	1605139	1	1.00E-05	2717.116	540706	9.011767	1
0.517588	0.482412	1769139	1	9.00E-06	2982.07	593432	9.890533	1
0.512563	0.487437	1989280	1	8.00E-06	3294.447	655595	10.92658	1
0.502513	0.497487	2279703	1	7.00E-06	3847.085	765570	12.7595	1
0.447237	0.552763	2645305	1	6.00E-06	4467.864	889105	14.81842	1
0.502513	0.497487	3189939	1	5.00E-06	5382.714	1071192	17.8532	1
0.577889	0.422111	3925776	1	4.00E-06	6594.513	1312324	21.87207	1
0.517588	0.482412	5185330	1	3.00E-06	9035.729	1798110	29.9685	1
0.507538	0.492462	7860425	1	2.00E-06	13609.97	2708391	45.13985	1
0.502513	0.497487	1.58E+07	1	1.00E-06	26440.74	5261730	87.6955	1

 

 

a90fa
测试集	0.3	a
	0.7	90fa
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.525959	0.526872	119.8241	0.732049	0.5	1.969849	407	0.006783	1
0.506122	0.496059	337.8693	0.990221	0.4	2.135678	440	0.007333	1
0.50456	0.497614	457.6633	1	0.3	2.276382	453	0.00755	1
0.536599	0.46456	577.3618	1	0.2	2.125628	423	0.00705	1
0.522535	0.477742	791.5528	1	0.1	2.386935	475	0.007917	1
0.566561	0.433481	3026.075	1	0.01	6.055276	1205	0.020083	1
0.502508	0.497492	19902.08	1	0.001	35.54271	7073	0.117883	1
0.527633	0.472367	172645.8	1	1.00E-04	286.3266	56979	0.94965	1
0.502512	0.497488	185716.9	1	9.00E-05	304.5829	60628	1.010467	1
0.472366	0.527634	211638.4	1	8.00E-05	360.0251	71645	1.194083	1
0.462317	0.537683	243420.3	1	7.00E-05	411.6633	81923	1.365383	1
0.52261	0.47739	280313	1	6.00E-05	470.4623	93638	1.560633	1
0.53266	0.46734	335069.5	1	5.00E-05	553.5477	110157	1.83595	1
0.507537	0.492463	416923.9	1	4.00E-05	712.1759	141773	2.362883	1
0.492463	0.507537	544859.7	1	3.00E-05	931.4121	185351	3.089183	1
0.462313	0.537687	809301.8	1	2.00E-05	1391.573	276939	4.61565	1
0.457287	0.542713	1616243	1	1.00E-05	2720.95	541472	9.024533	1
0.527638	0.472362	1754074	1	9.00E-06	3027.362	602445	10.04075	1
0.487437	0.512563	1963808	1	8.00E-06	3364.889	669615	11.16025	1
0.547738	0.452262	2284711	1	7.00E-06	3762.799	748797	12.47995	1
0.432162	0.567838	2631248	1	6.00E-06	4358.884	867418	14.45697	1
0.442212	0.557788	3173364	1	5.00E-06	5300.457	1054793	17.57988	1
0.507538	0.492462	3927128	1	4.00E-06	6803.618	1353921	22.56535	1
0.477387	0.522613	5237892	1	3.00E-06	7841.457	1560465	26.00775	1
0.396985	0.603015	7909708	1	2.00E-06	13046.52	2596272	43.2712	1
0.467337	0.532663	1.59E+07	1	1.00E-06	25892.07	5152523	85.87538	1

 

 

fa90fa
测试集	0.3	fa
	0.7	90fa
f2[0]	f2[1]	迭代次数n	平均准确率p-ave	δ	耗时ms/次	耗时ms/199次	耗时 min/199	最大准确率p-max
0.526581	0.530115	119.7337	0.711712	0.5	2.045226	407	0.006783	1
0.522076	0.483264	353.1307	0.995493	0.4	2.125628	423	0.00705	1
0.520863	0.480805	464.4573	1	0.3	2.201005	438	0.0073	1
0.523406	0.476828	574.5779	1	0.2	2.040201	406	0.006767	1
0.490214	0.510307	789.7437	1	0.1	2.532663	520	0.008667	1
0.581335	0.418703	2992.181	1	0.01	5.994975	1193	0.019883	1
0.467401	0.5326	19734.79	1	0.001	33.96482	6759	0.11265	1
0.442223	0.557778	170339.3	1	1.00E-04	278.8844	55498	0.924967	1
0.4171	0.5829	193358.2	1	9.00E-05	316.7789	63040	1.050667	1
0.512561	0.487439	210133.1	1	8.00E-05	348.7085	69394	1.156567	1
0.532659	0.467341	241907	1	7.00E-05	415.2412	82636	1.377267	1
0.482414	0.517586	276559.3	1	6.00E-05	461.4121	91822	1.530367	1
0.522611	0.477389	331354.3	1	5.00E-05	551.2362	109696	1.828267	1
0.482414	0.517587	412179.2	1	4.00E-05	686.8794	136704	2.2784	1
0.532661	0.467339	547005	1	3.00E-05	901.402	179379	2.98965	1
0.522612	0.477388	818629.8	1	2.00E-05	1357.457	270134	4.502233	1
0.472362	0.527638	1610707	1	1.00E-05	2674.236	532173	8.86955	1
0.507538	0.492462	1764101	1	9.00E-06	2873.342	571795	9.529917	1
0.517588	0.482412	1985273	1	8.00E-06	3326.422	661958	11.03263	1
0.472362	0.527638	2282847	1	7.00E-06	3824.387	761053	12.68422	1
0.482412	0.517588	2661392	1	6.00E-06	4453.96	886354	14.77257	1
0.592964	0.407036	3176851	1	5.00E-06	5311.141	1056917	17.61528	1
0.552763	0.447237	3967423	1	4.00E-06	6626.93	1318759	21.97932	1
0.527638	0.472362	5191960	1	3.00E-06	8756.98	1742640	29.044	1
0.522613	0.477387	7753992	1	2.00E-06	13257.36	2638216	43.97027	1
0.462312	0.537688	1.58E+07	1	1.00E-06	24996.18	4974240	82.904	1