php - 我创建并训练了一个 PHP-FANN，但我没有得到想要的结果或准确性

Question

我在 geekgirljoy 的一些示例和教程的帮助下在 PHP 中创建了一个 FANN， 并基于php-fann-repo中的 ocr 示例

我正在尝试创建一个系统，它根据订单号告诉我这是哪种类型的订单。

我已经创建了训练数据，对其进行了训练和测试，但无法得到我期望的结果。我现在处于随机更改参数不再有帮助的地步，我不确定我的假设一开始是否正确。

一些训练数据：我得到了 60k 行的间隔分割的二进制序号

60000 32 1
0 0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 1 1 1 0 1 1 1 0 0 0 1 1 0 0 1 0
0.01
0 0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 1 1 1 0 1 1 1 0 0 0 1 1 0 1 0 0
0.01
0 0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 1 1 1 0 1 1 1 0 0 0 1 1 0 1 1 0
0.01
0 0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 1 1 1 0 1 1 1 0 0 0 1 1 1 0 0 0
0.01
0 0 0 1 1 1 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 1 1 0 0 1 0 0 1 0 1 0
0.07
0 0 0 1 1 1 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 1 1 0 0 1 0 0 1 1 0 0
0.07
0 0 0 1 1 1 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 1 1 0 0 1 0 0 1 1 1 0
0.07
0 0 0 1 1 1 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 1 1 0 0 1 0 1 0 0 0 0
0.07

培训文件：

FANN_FLO_2.1
num_layers=3
learning_rate=0.700000
connection_rate=1.000000
network_type=0
learning_momentum=0.000000
training_algorithm=2
train_error_function=1
train_stop_function=0
cascade_output_change_fraction=0.010000
quickprop_decay=-0.000100
quickprop_mu=1.750000
rprop_increase_factor=1.200000
rprop_decrease_factor=0.500000
rprop_delta_min=0.000000
rprop_delta_max=50.000000
rprop_delta_zero=0.100000
cascade_output_stagnation_epochs=12
cascade_candidate_change_fraction=0.010000
cascade_candidate_stagnation_epochs=12
cascade_max_out_epochs=150
cascade_min_out_epochs=50
cascade_max_cand_epochs=150
cascade_min_cand_epochs=50
cascade_num_candidate_groups=2
bit_fail_limit=3.49999994039535522461e-01
cascade_candidate_limit=1.00000000000000000000e+03
cascade_weight_multiplier=4.00000005960464477539e-01
cascade_activation_functions_count=10
cascade_activation_functions=3 5 7 8 10 11 14 15 16 17 
cascade_activation_steepnesses_count=4
cascade_activation_steepnesses=2.50000000000000000000e-01 5.00000000000000000000e-01 7.50000000000000000000e-01 1.00000000000000000000e+00 
layer_sizes=33 17 2 
scale_included=0
neurons (num_inputs, activation_function, activation_steepness)=(0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (0, 0, 0.00000000000000000000e+00) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (33, 5, 5.00000000000000000000e-01) (0, 5, 0.00000000000000000000e+00) (17, 5, 5.00000000000000000000e-01) (0, 5, 0.00000000000000000000e+00) 
connections (connected_to_neuron, weight)=(0, -4.61362116038799285889e-02) (1, -7.24165216088294982910e-02) (2, -1.54439583420753479004e-02) (3, 8.89342501759529113770e-02) (4, -1.17050260305404663086e-02) (5, 2.18402743339538574219e-02) (6, 3.76827046275138854980e-02) (7, -4.71979975700378417969e-02) (8, 9.12376716732978820801e-02) (9, -4.86264117062091827393e-02) (10, -8.81998762488365173340e-02) (11, -4.78897392749786376953e-02) (12, 9.77639481425285339355e-02) (13, 2.96645238995552062988e-02) (14, 6.46188631653785705566e-02) (15, 7.25518167018890380859e-03) (16, -9.11594703793525695801e-02) (17, 2.28227004408836364746e-02) (18, 5.24043217301368713379e-02) (19, -4.13042865693569183350e-02) (20, 6.29015043377876281738e-02) (21, 7.06591978669166564941e-02) (22, 5.67197278141975402832e-02) (23, 5.40713146328926086426e-02) (24, 1.12115144729614257812e-02) (25, 1.84408575296401977539e-02) (26, 8.76630619168281555176e-02) (27, -9.43159908056259155273e-02) (28, -2.85221189260482788086e-02) (29, -2.38240733742713928223e-02) (30, -5.08805401623249053955e-02) (31, 2.53416672348976135254e-02) (32, 3.75940650701522827148e-03) (0, 3.36754992604255676270e-02) (1, 1.42759233713150024414e-02) (2, 9.20543894171714782715e-02) (3, -4.44842278957366943359e-02) (4, -4.80413846671581268311e-02) (5, -5.51436059176921844482e-02) (6, -5.32465577125549316406e-02) (7, 3.33221256732940673828e-03) (8, -4.33434806764125823975e-02) (9, -1.13629549741744995117e-03) (10, 1.09615176916122436523e-03) (11, 8.63210633397102355957e-02) (12, -3.65174412727355957031e-02) (13, -9.16486680507659912109e-02) (14, 9.51615795493125915527e-02) (15, 8.63052681088447570801e-02) (16, 6.07556626200675964355e-02) (17, -4.61427047848701477051e-02) (18, 4.92067709565162658691e-02) (19, 3.14148589968681335449e-02) (20, -8.94229784607887268066e-02) (21, 3.27809154987335205078e-03) (22, -5.73736317455768585205e-02) (23, 2.90178731083869934082e-02) (24, -9.05884802341461181641e-03) (25, -5.16896173357963562012e-02) (26, -9.95042547583580017090e-02) (27, 6.71170875430107116699e-02) (28, -2.57015973329544067383e-03) (29, 2.58374139666557312012e-02) (30, -2.91235074400901794434e-02) (31, -6.88946545124053955078e-02) (32, -5.98866716027259826660e-02) (0, -3.70691195130348205566e-02) (1, -1.33788734674453735352e-02) (2, -7.92805850505828857422e-03) (3, 7.78727233409881591797e-03) (4, 3.33745554089546203613e-02) (5, 9.54041555523872375488e-02) (6, 6.44438043236732482910e-02) (7, -6.77617341279983520508e-02) (8, -3.49969416856765747070e-03) (9, 5.07648512721061706543e-02) (10, -4.27917391061782836914e-03) (11, 4.85165417194366455078e-03) (12, 4.59264293313026428223e-02) (13, -1.79739147424697875977e-02) (14, -3.43926995992660522461e-02) (15, 9.97837260365486145020e-02) (16, -6.87671378254890441895e-02) (17, 9.70221534371376037598e-02) (18, -8.96392464637756347656e-02) (19, 3.45109626650810241699e-02) (20, -6.03514760732650756836e-02) (21, 3.93786355853080749512e-02) (22, -7.45478942990303039551e-02) (23, -1.20410919189453125000e-02) (24, 3.98743823170661926270e-02) (25, 9.25691798329353332520e-02) (26, 8.53887572884559631348e-02) (27, -3.42882126569747924805e-02) (28, -3.65543216466903686523e-02) (29, -8.35058987140655517578e-02) (30, 5.82511723041534423828e-03) (31, 2.63765677809715270996e-02) (32, 3.11522185802459716797e-03) (0, 9.78970602154731750488e-02) (1, -6.58361613750457763672e-02) (2, -6.35102093219757080078e-02) (3, 9.33012291789054870605e-02) (4, 9.86076369881629943848e-02) (5, -3.12719494104385375977e-02) (6, -1.01984664797782897949e-02) (7, 4.93725016713142395020e-02) (8, 6.44488856196403503418e-02) (9, 9.46531817317008972168e-02) (10, -4.70107048749923706055e-03) (11, -5.35250306129455566406e-02) (12, -3.97395193576812744141e-02) (13, -4.91733849048614501953e-03) (14, -2.22921743988990783691e-02) (15, -4.27173636853694915771e-02) (16, 5.44340908527374267578e-03) (17, -8.77812206745147705078e-02) (18, -3.06884199380874633789e-03) (19, -5.51779642701148986816e-02) (20, -6.23291134834289550781e-02) (21, 8.48900750279426574707e-02) (22, 8.46964195370674133301e-02) (23, -6.97599276900291442871e-02) (24, 7.02788308262825012207e-02) (25, -4.95917983353137969971e-02) (26, -6.31424784660339355469e-03) (27, 8.67729261517524719238e-02) (28, 5.62333241105079650879e-02) (29, -7.99376815557479858398e-02) (30, -1.01118534803390502930e-02) (31, 5.41303828358650207520e-02) (32, -4.57738414406776428223e-02) (0, 2.63779237866401672363e-02) (1, 4.74315956234931945801e-02) (2, -4.71661984920501708984e-02) (3, 9.51059833168983459473e-02) (4, -6.27668648958206176758e-02) (5, -9.77937132120132446289e-02) (6, 5.95548674464225769043e-02) (7, -6.81136846542358398438e-02) (8, -2.49478220939636230469e-03) (9, -9.39701646566390991211e-02) (10, -7.85320997238159179688e-03) (11, 9.25878807902336120605e-02) (12, -1.62623375654220581055e-02) (13, 4.94294241070747375488e-02) (14, -1.96871906518936157227e-03) (15, -4.04354929924011230469e-03) (16, -5.36394119262695312500e-02) (17, 4.28533181548118591309e-02) (18, 3.36273387074470520020e-02) (19, -6.87493458390235900879e-02) (20, 2.75497362017631530762e-02) (21, 6.38674125075340270996e-02) (22, -9.84705314040184020996e-02) (23, 7.79579356312751770020e-02) (24, -4.24468331038951873779e-02) (25, 8.83023813366889953613e-02) (26, 3.41912582516670227051e-02) (27, -2.23845094442367553711e-02) (28, -2.18094661831855773926e-02) (29, -1.16783604025840759277e-02) (30, 3.18416431546211242676e-02) (31, -9.54315364360809326172e-02) (32, -6.42467588186264038086e-02) (0, 8.46754387021064758301e-02) (1, 9.96744558215141296387e-02) (2, -2.70136222243309020996e-02) (3, 8.68817344307899475098e-02) (4, 5.92293217778205871582e-02) (5, 4.87269461154937744141e-03) (6, -1.56130492687225341797e-02) (7, 6.52591660618782043457e-02) (8, 9.70194861292839050293e-02) (9, -2.30251699686050415039e-02) (10, -5.10031804442405700684e-02) (11, 4.64489087462425231934e-02) (12, 7.50061199069023132324e-02) (13, 4.49532791972160339355e-02) (14, 9.28095057606697082520e-02) (15, 1.78594365715980529785e-02) (16, -2.14193910360336303711e-02) (17, -7.59398490190505981445e-02) (18, -5.45908398926258087158e-02) (19, -5.75519762933254241943e-02) (20, -7.44103714823722839355e-02) (21, -7.66329094767570495605e-02) (22, 1.19209289550781250000e-06) (23, -8.61079841852188110352e-02) (24, 5.75583502650260925293e-02) (25, 7.76166692376136779785e-02) (26, -7.91744887828826904297e-03) (27, -5.41200228035449981689e-02) (28, 9.45831835269927978516e-03) (29, -3.34898382425308227539e-03) (30, -1.83667764067649841309e-02) (31, -5.86624443531036376953e-03) (32, -3.67452949285507202148e-03) (0, 5.46196028590202331543e-02) (1, -1.89845040440559387207e-02) (2, -4.44452166557312011719e-02) (3, -4.05077114701271057129e-02) (4, 6.54024556279182434082e-02) (5, -7.91860669851303100586e-02) (6, -4.34882305562496185303e-02) (7, -5.76227270066738128662e-02) (8, -3.01892384886741638184e-02) (9, -9.70393195748329162598e-02) (10, -8.26166123151779174805e-02) (11, -8.52359682321548461914e-02) (12, 9.57701876759529113770e-02) (13, 3.52428182959556579590e-02) (14, -6.65535777807235717773e-03) (15, -8.01696628332138061523e-02) (16, 8.06519761681556701660e-02) (17, 3.57926562428474426270e-02) (18, -5.45800328254699707031e-02) (19, -9.59809273481369018555e-02) (20, -6.42061531543731689453e-02) (21, -4.06880155205726623535e-02) (22, 6.15774169564247131348e-02) (23, -8.65894779562950134277e-02) (24, 5.13945445418357849121e-02) (25, -9.25426110625267028809e-02) (26, 2.28688344359397888184e-02) (27, -5.19544407725334167480e-02) (28, -1.09093859791755676270e-02) (29, -8.29973965883255004883e-02) (30, 4.43710312247276306152e-02) (31, -5.62897883355617523193e-02) (32, -1.98189914226531982422e-03) (0, 9.99258235096931457520e-02) (1, 3.20249795913696289062e-03) (2, -3.65794524550437927246e-02) (3, -7.92602524161338806152e-02) (4, 5.97142651677131652832e-02) (5, 5.79782575368881225586e-03) (6, -9.44948941469192504883e-03) (7, 6.26749470829963684082e-02) (8, 2.31812149286270141602e-02) (9, 5.31454384326934814453e-03) (10, 5.84451481699943542480e-02) (11, -4.15759757161140441895e-02) (12, 9.86591801047325134277e-02) (13, 7.82754793763160705566e-02) (14, -6.09239935874938964844e-02) (15, 3.44518497586250305176e-02) (16, -7.63045549392700195312e-02) (17, -5.69049231708049774170e-02) (18, 7.02456906437873840332e-02) (19, -1.69925615191459655762e-02) (20, -9.53275039792060852051e-02) (21, 8.36562141776084899902e-02) (22, -6.55980259180068969727e-02) (23, -8.78701135516166687012e-02) (24, 6.52505457401275634766e-03) (25, -1.75524652004241943359e-02) (26, 1.22050195932388305664e-03) (27, 2.35276594758033752441e-02) (28, -7.31814354658126831055e-02) (29, 4.49307188391685485840e-02) (30, -7.84542486071586608887e-02) (31, -7.32556283473968505859e-02) (32, -5.18667846918106079102e-02) (0, -1.50336995720863342285e-02) (1, -5.25158755481243133545e-02) (2, -9.21525135636329650879e-02) (3, 9.07641127705574035645e-02) (4, 3.80346253514289855957e-02) (5, 7.05224350094795227051e-02) (6, 1.39453262090682983398e-02) (7, -5.66508285701274871826e-02) (8, 2.89675816893577575684e-02) (9, 7.23693594336509704590e-02) (10, -5.79916499555110931396e-02) (11, 7.24305957555770874023e-03) (12, -8.85546356439590454102e-02) (13, 7.64601901173591613770e-02) (14, 3.09385135769844055176e-02) (15, -4.54595573246479034424e-02) (16, 4.67058941721916198730e-02) (17, -8.60540568828582763672e-02) (18, -4.07870598137378692627e-02) (19, 3.03620919585227966309e-02) (20, -5.16520775854587554932e-02) (21, -2.86571756005287170410e-02) (22, -6.31128549575805664062e-02) (23, 3.07954624295234680176e-02) (24, 7.25633278489112854004e-02) (25, 6.04147985577583312988e-02) (26, 5.76140210032463073730e-02) (27, 1.74940451979637145996e-02) (28, 8.19605663418769836426e-02) (29, 8.43584015965461730957e-02) (30, 6.56272694468498229980e-02) (31, -3.30731421709060668945e-02) (32, -6.81574791669845581055e-02) (0, 7.34747573733329772949e-02) (1, -4.23090159893035888672e-02) (2, 6.98771551251411437988e-02) (3, 4.39971908926963806152e-02) (4, 7.16363266110420227051e-02) (5, -8.67736712098121643066e-02) (6, -2.70352214574813842773e-02) (7, 4.40056845545768737793e-02) (8, -4.47653122246265411377e-02) (9, 8.02078470587730407715e-02) (10, 5.54510429501533508301e-02) (11, -6.83051198720932006836e-02) (12, 1.11463516950607299805e-02) (13, -9.00085121393203735352e-02) (14, 7.84007683396339416504e-02) (15, 2.50923112034797668457e-02) (16, -3.07955741882324218750e-02) (17, 8.76285880804061889648e-03) (18, 7.34402164816856384277e-02) (19, 4.05472591519355773926e-02) (20, 4.56500127911567687988e-02) (21, 4.23568487167358398438e-03) (22, 1.31105929613113403320e-02) (23, 6.06481730937957763672e-03) (24, -3.81502993404865264893e-02) (25, -6.93953707814216613770e-02) (26, -1.19746178388595581055e-02) (27, -5.37918992340564727783e-02) (28, 9.62318852543830871582e-02) (29, 5.49522563815116882324e-02) (30, -2.19493731856346130371e-02) (31, 6.97066411375999450684e-02) (32, -8.73567685484886169434e-02) (0, -5.20722158253192901611e-02) (1, 1.37038379907608032227e-02) (2, 8.42795446515083312988e-02) (3, -3.88458780944347381592e-02) (4, 8.66686180233955383301e-02) (5, 2.82852128148078918457e-02) (6, 1.63888111710548400879e-02) (7, 6.68764635920524597168e-02) (8, -1.62637382745742797852e-02) (9, 4.80836853384971618652e-02) (10, -2.19771862030029296875e-02) (11, -6.27224892377853393555e-03) (12, 2.64844521880149841309e-02) (13, -9.68848913908004760742e-02) (14, 6.29321858286857604980e-02) (15, -6.47526830434799194336e-02) (16, 7.65553340315818786621e-02) (17, 3.47943603992462158203e-03) (18, 8.08973386883735656738e-02) (19, -1.92089825868606567383e-02) (20, -8.34099799394607543945e-02) (21, -1.30378454923629760742e-02) (22, 4.26407232880592346191e-02) (23, -5.28053492307662963867e-02) (24, 7.49875381588935852051e-02) (25, 8.88488367199897766113e-02) (26, -5.65734580159187316895e-02) (27, 2.99397930502891540527e-02) (28, -3.31005528569221496582e-02) (29, -8.68668183684349060059e-02) (30, 4.25830259919166564941e-02) (31, 1.48272365331649780273e-02) (32, 2.68370136618614196777e-02) (0, 2.68625691533088684082e-02) (1, 7.59813562035560607910e-02) (2, 1.35056376457214355469e-02) (3, -4.48522083461284637451e-02) (4, -7.62983411550521850586e-03) (5, -1.96179077029228210449e-02) (6, 3.88840511441230773926e-02) (7, -5.95461502671241760254e-02) (8, 5.84049001336097717285e-02) (9, -6.73882067203521728516e-02) (10, 6.69383034110069274902e-02) (11, 6.15200176835060119629e-02) (12, 9.55439880490303039551e-02) (13, -9.78143736720085144043e-02) (14, 3.80753502249717712402e-02) (15, -9.76592302322387695312e-04) (16, 8.30829665064811706543e-02) (17, -8.11336338520050048828e-02) (18, 1.56134217977523803711e-02) (19, -2.99548804759979248047e-02) (20, 6.15070834755897521973e-02) (21, 6.28080740571022033691e-02) (22, -5.49673400819301605225e-02) (23, 5.03559187054634094238e-02) (24, -9.37653779983520507812e-02) (25, 7.49724581837654113770e-02) (26, -8.27446356415748596191e-02) (27, -8.06321948766708374023e-02) (28, 1.75554752349853515625e-02) (29, 3.20826098322868347168e-02) (30, 4.62048277258872985840e-02) (31, -5.55819571018218994141e-02) (32, 8.06395709514617919922e-03) (0, -4.02895472943782806396e-02) (1, -4.34167683124542236328e-04) (2, -9.95658785104751586914e-02) (3, 4.00925502181053161621e-02) (4, -6.15501180291175842285e-02) (5, -5.91120272874832153320e-02) (6, -1.50255113840103149414e-03) (7, -2.89383158087730407715e-02) (8, -9.21737253665924072266e-02) (9, -3.99825386703014373779e-02) (10, -3.33943367004394531250e-02) (11, -8.99880975484848022461e-02) (12, 9.80928018689155578613e-02) (13, 6.56290724873542785645e-02) (14, 9.30948629975318908691e-02) (15, -8.30408260226249694824e-02) (16, -1.87574997544288635254e-02) (17, -3.68600189685821533203e-02) (18, 7.84662589430809020996e-02) (19, -5.59494234621524810791e-02) (20, 8.17264616489410400391e-03) (21, 2.88221761584281921387e-02) (22, -4.97148036956787109375e-02) (23, -1.68548971414566040039e-02) (24, 4.60775420069694519043e-02) (25, -3.03469970822334289551e-02) (26, -9.92994233965873718262e-02) (27, -2.18398571014404296875e-02) (28, -8.41421782970428466797e-02) (29, -5.48813790082931518555e-02) (30, 8.62241014838218688965e-02) (31, -2.44317203760147094727e-02) (32, 4.46844622492790222168e-02) (0, 8.66582170128822326660e-02) (1, -8.43391716480255126953e-02) (2, 8.31343457102775573730e-02) (3, -7.24538117647171020508e-02) (4, 1.41582712531089782715e-02) (5, -4.58039753139019012451e-02) (6, -6.46275281906127929688e-02) (7, 7.41757377982139587402e-02) (8, 2.08016857504844665527e-02) (9, -5.46156279742717742920e-02) (10, 7.22685530781745910645e-02) (11, -1.35692507028579711914e-02) (12, -6.15207627415657043457e-02) (13, 8.92277285456657409668e-02) (14, 6.76732584834098815918e-02) (15, 1.61921977996826171875e-03) (16, 6.76939859986305236816e-02) (17, -8.82761701941490173340e-02) (18, -9.02081355452537536621e-02) (19, -3.48383188247680664062e-03) (20, -3.79909761250019073486e-02) (21, -7.06303864717483520508e-03) (22, -5.74062950909137725830e-02) (23, 3.16620245575904846191e-02) (24, -6.36245310306549072266e-03) (25, 2.07538455724716186523e-02) (26, 4.75198552012443542480e-02) (27, 3.87561544775962829590e-02) (28, 6.97793811559677124023e-03) (29, -7.69118666648864746094e-02) (30, -1.65593847632408142090e-02) (31, -6.36383891105651855469e-03) (32, -6.12510368227958679199e-02) (0, -3.34250479936599731445e-02) (1, 2.11823582649230957031e-02) (2, 5.29072359204292297363e-02) (3, 2.07709670066833496094e-02) (4, 5.65548315644264221191e-02) (5, 2.70829871296882629395e-02) (6, -5.84273450076580047607e-02) (7, -9.80608016252517700195e-02) (8, -6.48468732833862304688e-04) (9, 2.80034020543098449707e-02) (10, -5.95815591514110565186e-02) (11, -1.14207416772842407227e-02) (12, -4.32334095239639282227e-03) (13, 4.20376583933830261230e-02) (14, -4.37267534434795379639e-02) (15, 7.40049034357070922852e-03) (16, 5.18295243382453918457e-02) (17, 5.27894124388694763184e-02) (18, 6.94095119833946228027e-02) (19, -5.52335083484649658203e-02) (20, 9.53831151127815246582e-02) (21, 1.07154995203018188477e-03) (22, 3.84040400385856628418e-02) (23, 1.61369666457176208496e-02) (24, -5.14086000621318817139e-02) (25, -2.28398069739341735840e-02) (26, -7.68850892782211303711e-02) (27, -2.83204615116119384766e-02) (28, 6.06008097529411315918e-02) (29, 1.67510733008384704590e-02) (30, 1.04285031557083129883e-02) (31, -7.28242397308349609375e-02) (32, -6.20665699243545532227e-02) (0, -3.66642549633979797363e-02) (1, 4.79467287659645080566e-02) (2, 9.44882556796073913574e-02) (3, 9.04187336564064025879e-02) (4, 8.95193889737129211426e-02) (5, 9.64274480938911437988e-02) (6, -1.02297365665435791016e-02) (7, 1.75227895379066467285e-02) (8, -6.31541088223457336426e-02) (9, 7.83495232462882995605e-02) (10, -8.68005454540252685547e-02) (11, 7.88835510611534118652e-02) (12, -6.53772354125976562500e-02) (13, 2.05999463796615600586e-02) (14, 3.07130888104438781738e-02) (15, 8.74121859669685363770e-02) (16, -9.99053567647933959961e-03) (17, 7.54795745015144348145e-02) (18, 8.27952995896339416504e-02) (19, 9.10810157656669616699e-02) (20, 1.38836055994033813477e-02) (21, -1.06773525476455688477e-03) (22, -6.03275895118713378906e-02) (23, 9.10437926650047302246e-02) (24, 2.20471844077110290527e-02) (25, 1.13519430160522460938e-02) (26, 5.16446009278297424316e-02) (27, -6.12017475068569183350e-02) (28, -7.82195478677749633789e-02) (29, 7.88203552365303039551e-02) (30, -2.32683122158050537109e-02) (31, -1.48838013410568237305e-02) (32, 2.67670825123786926270e-02) (33, -2.87800580263137817383e-02) (34, -2.44650691747665405273e-02) (35, 1.62864699959754943848e-02) (36, -3.23526039719581604004e-02) (37, 6.53051808476448059082e-02) (38, -6.61907345056533813477e-02) (39, 4.49328124523162841797e-03) (40, 4.36547026038169860840e-02) (41, -5.29912821948528289795e-02) (42, -1.66231542825698852539e-02) (43, 7.82774761319160461426e-02) (44, 6.76086619496345520020e-02) (45, -8.59100818634033203125e-02) (46, 6.56896606087684631348e-02) (47, -4.23818789422512054443e-02) (48, 8.95694866776466369629e-02) (49, 4.84849587082862854004e-02) 

我的训练脚本：

$filenameLoad = dirname(__FILE__) . "/data/order.data";
$filenameSave = dirname(__FILE__) . "/data/ordernumbers_float.net";

$num_input = 32;
$num_output = 1;
$num_layers = 3;
$num_neurons_hidden = ($num_input + $num_output) / 2;
//$num_neurons_hidden = 20;

$desired_error = 0.00001;
$max_epochs = 5000000;
$epochs_between_reports = 10;

$ann = fann_create_standard($num_layers, $num_input, $num_neurons_hidden, $num_output);

if ($ann) {
    fann_set_activation_function_hidden($ann, FANN_SIGMOID_SYMMETRIC);
    fann_set_activation_function_output($ann, FANN_SIGMOID_SYMMETRIC);

    if (fann_train_on_file($ann, $filenameLoad, $max_epochs, $epochs_between_reports, $desired_error)) {
        print('ordernumbers trained' . PHP_EOL);
    }

    if (fann_save($ann, $filenameSave)) {
        print('ordernumbers_float.net saved' . PHP_EOL);
    }

    fann_destroy($ann);
}

我的测试脚本：

<?php

//include_once 'Classes/Helper.php';

$train_file = (dirname(__FILE__) . "/data/ordernumbers_float.net");

if (!is_file($train_file)) {
    die("The file ordernumbers_float.net has not been created!" . PHP_EOL);
}

//$helper = new Helper();

$ann = fann_create_from_file($train_file);

if ($ann) {
    $orderNumber = 108643364;
    //$binaryOrderNumber = $helper->getBinaryFromOrdernumber($orderNumber);
    $binaryOrderNumber = '00000110011110011100010000100100';
//    $input = $helper->getSplittedBinary($binaryOrderNumber);
    $input = array("0", "0", "0", "0", "0", "1", "1", "0", "0", "1", "1", "1", "1", "0", "0", "1", "1", "1", "0", "0", "0", "1", "0", "0", "0", "0", "1", "0", "0", "1", "0", "0");
//    $inputString = $helper->getSplittetBinaryOutput($input);
    $inputString = "0 0 0 0 0 1 1 0 0 1 1 1 1 0 0 1 1 1 0 0 0 1 0 0 0 0 1 0 0 1 0 0";

    $calc_out = fann_run($ann, $input);
    printf("ordernumber %s -> %s -> test raw: %f trimmed: %f expected: %f\n", $orderNumber, $inputString, $calc_out[0], floor($calc_out[0] * 100) / 100, 0.01);

    fann_destroy($ann);
} else {
    die("Invalid file format" . PHP_EOL);
}

标签： phpfann

Answer 1

长话短说，对于这样一个小而简单的网络，您的数据集可能太复杂了。

当我编写 OCR 示例时，我通过将所有 94 个字符“压缩”成单个输出神经元来有点炫耀。通常不会以这种方式完成，当然也不会使用复杂的数据集。

通常，您希望为网络需要识别的每个“类”指定一个输出神经元。

简而言之，与学习将专用输出神经元/模式与特定类相关联相比，网络更难学会在单个神经元上正确地将输出值增加或减少 0.01（就像我的 OCR ANN 的情况）。

您可以在我的神经网络 OCR“家族”的仓库中的 MNIST 子文件夹中找到更典型分类器实现的更好示例：https://github.com/geekgirljoy/OCR_Neural_Network

我的建议是重新设计您的 ANN。

根据您的代码，您的网络如下所示：

L0：

IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII L1：HHHHHHHHHHHHHHH

L2：O


如果您像这样重新设计它可能会更好地操作（分类）您的数据：

首先，在示例中确定不同类类型的数量你给了我看到 0.07 列出，所以我假设有七种不同的订单类型。

因此，ANN 应该如下所示：

L0：IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII

L1：足够数量的“隐藏”神经元

L2: OOOOOOO

其中 O1 代表第 1 类，O2 代表第 2 类等...

这意味着您的训练数据将更改为：


60000 32 7
0 0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 1 1 1 0 1 1 1 0 0 0 1 1 0 0 1 0
1 0 0 0 0 0 0 0
0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 1 1 1 0 1 1 1 0 0 0 1 1 0 1 0 0
1 0 0 0 0 0 0 0
0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 1 1 1 0 1 1 1 0 0 0 1 1 0 1 1 0
1 0 0 0 0 0 0
0 0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 1 1 1 0 1 1 1 0 0 0 1 1 1 0 0 0
1 0 0 0 0 0 0
0 0 0 1 1 1 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 1 1 0 0 1 0 0 1 0 1 0
0 0 0 0 0 0 1
0 0 0 1 1 1 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 1 1 0 0 1 0 0 1 1 0 0
0 0 0 0 0 0 1
0 0 0 1 1 1 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 1 1 0 0 1 0 0 1 1 1 0
0 0 0 0 0 0 1
0 0 0 1 1 1 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 1 1 0 0 1 0 1 0 0 0 0
0 0 0 0 0 0 1

类输出示例：

1 类：1 0 0 0 0 0 0
2 类：0 1 0 0 0 0 0
第 3 类：0 0 1 0 0 0 0
第 4 类：0 0 0 1 0 0 0
第 5 类：0 0 0 0 1 0 0
第 6 类：0 0 0 0 0 1 0
第 7 类：0 0 0 0 0 0 1


此外，根据您的方法，您可能会使用更硬的负值（如 -1 而不是 0）获得更好的结果，如下所示：

60000 32 7
-1 -1 -1 -1 -1 1 1 -1 -1 1 1 -1 1 1 -1 -1 1 1 1 -1 1 1 1 -1 -1 -1 1 1 -1 -1 1 -1
1 -1 -1 -1 -1 -1 -1
-1 -1 -1 -1 -1 1 1 -1 -1 1 1 -1 1 1 -1 -1 1 1 1 -1 1 1 1 -1 -1 -1 1 1 -1 1 -1 -1
1 -1 -1 -1 -1 -1 -1 -1
-1 -1 -1 -1 1 1 -1 -1 1 1 -1 1 1 -1 -1 1 1 1 -1 1 1 1 -1 -1 -1 1 1 -1 1 1 -1
1 -1 -1 -1 -1 -1 -1 -1
-1 -1 -1 -1 1 1 -1 -1 1 1 -1 1 1 -1 -1 1 1 1 -1 1 1 1 -1 -1 -1 1 1 1 -1 -1 -1
1 -1 -1 -1 -1 -1 -1
-1 -1 -1 1 1 1 -1 1 1 1 1 -1 -1 1 -1 -1 -1 1 -1 -1 1 1 1 -1 -1 1 -1 -1 1 -1 1 -1
-1 -1 -1 -1 -1 -1 1
-1 - 1 -1 1 1 1 -1 1 1 1 1 -1 -1 1 -1 -1 -1 1 -1 -1 1 1 1 -1 -1 1 -1 -1 1 1 -1 -1
-1 -1 -1 -1 -1 -1 1
-1 -1 -1 1 1 1 -1 1 1 1 1 -1 -1 1 -1 -1 -1 1 -1 -1 1 1 1 -1 -1 1 -1 -1 1 1 1 -1
-1 -1 -1 -1 -1 -1 1
-1 -1 -1 1 1 1 -1 1 1 1 1 -1 -1 1 -1 -1 -1 1 -1 - 1 1 1 1 -1 -1 1 -1 1 -1 -1 -1 -1
-1 -1 -1 -1 -1 -1 1


这是因为您使用的是“对称”隐藏/输出函数，例如 FANN_SIGMOID_SYMMETRIC，它是一个 sigmoid，因此 -1 到 0 以及从 0 到 1 之间的关系不是线性的，因此您应该通过更强烈地对比这样的输入/输出，在分类和可能更快的训练/更少的训练时期之间获得更好/更硬的区别。

无论如何，一旦您训练了网络并运行了测试，您只需将 max() 输出神经元作为您的答案。

示例：

// ANN 计算输入并将输出存储在结果数组中
$result = fann_run($ann, $input);

// 假设 ANN 响应如下：
// [-0.9,0.1,-0.2,0.4,0.1,0.5,0.6,0.99,-0.6,0.4]

// 假设有 10 个输出代表那么多类
// 0 - 9
// [0,1,2,3,4,5,6,7,8,9]
//
// 找出哪个输出包含最高值（预测/分类）
$highest = max($result); // $highest 现在包含值 0.99

// 所以要将最高值转换为一个类，我们在 $result 数组中找到键/位置
$class = array_search($highest, $result);

var_dump($class);
// int(7)

为什么？因为第 7 个键（第 7 个/第 8 个（取决于你怎么看）是高值）：

array(0=>0.9,
1=>0.1,
2=>-0.2,
3=>0.4,
4=>0.1 ,
5=>0.5,
6=>0.6,
7=>0.99,
8=>-0.6,
0=>0.4
);

在可能同时存在多种类类型的情况下，您可以改为“softmax”。

希望这可以帮助！:-)