Problem with "show_train_ninternals"

[Giulio Tani]

Hi,

I want to display the internal states of svhn network.

I use all the default parameters but i have this error:

Error: cannot loop on idx with order 0. idx is: <scalar>, in idxlooper at /home/giulio/eblearn/core/libidx/include/idxiter.hpp:498

Stack:

train(_ZN3ebl9idxlooperIdEC2ERNS_3idxIdEEi+0x1d5)[0x5f7895]

train(_ZN3ebl14module_1_1_gui14display_fprop2IdEEvRNS_10module_1_1IT_EERNS_5stateIS3_EES8_RjS9_dS3_S3_biPKc+0x8c7)[0x62f6a7]

train(_ZN3ebl10layers_gui13display_fpropIdEEvRNS_14module_1_1_guiERNS_6layersIT_EERNS_5stateIS5_EESA_RjSB_dS5_S5_b+0xf9)[0x62f9d9]

train(_ZN3ebl14module_1_1_gui14display_fprop2IdEEvRNS_10module_1_1IT_EERNS_5stateIS3_EES8_RjS9_dS3_S3_biPKc+0x129)[0x62ef09]

train(_ZN3ebl14module_1_1_gui14display_fprop2IdEEvRNS_10module_1_1IT_EERNS_5stateIS3_EES8_RjS9_dS3_S3_biPKc+0x294)[0x62f074]

train(_ZN3ebl10layers_gui13display_fpropIdEEvRNS_14module_1_1_guiERNS_6layersIT_EERNS_5stateIS5_EESA_RjSB_dS5_S5_b+0xf9)[0x62f9d9]

train(_ZN3ebl14module_1_1_gui14display_fprop2IdEEvRNS_10module_1_1IT_EERNS_5stateIS3_EES8_RjS9_dS3_S3_biPKc+0x129)[0x62ef09]

train(_ZN3ebl22supervised_trainer_guiIdfiE17display_internalsERNS_18supervised_trainerIdfiEERNS_18labeled_datasourceIdfiEERNS_11infer_paramERNS_8gd_paramEjjjdiPKc+0x7da)[0x6304fa]

train(_ZN3ebl4testIdfiEEvjRNS_13configurationERSsRNS_9parameterIT_EERNS_18supervised_trainerIS5_T0_T1_EERNS_18labeled_datasourceIS5_S9_SA_EESF_RNS_16classifier_meterESH_RNS_11infer_paramERNS_8gd_paramES3_RKSs+0x7c5)[0x687245]

train[0x4a00dc]

Aborted (core dumped)

This is all the the svhn.conf file:

# important training parameters (others are described later below)

reg=.00001       # regularization factor

reg_time=0     # time (in samples) after which to start regularizing

eta=.00001     # learning rate

ndiaghessian=400     # number of sample for 2nd derivatives estimation

hessian_period=20000 # calculates 2nd derivatives after every n samples

epoch_mode=0     # 0: fixed number of samples/iter 1: show all at least once

epoch_size=10000     # number of samples to show per iteration

iterations=1000      # number of training iterations

# training data ################################################################

root=${HOME}/eblearn/demos/svhn/svhn/ToUse # root directory of the dataset

train_dsname=svhn_yuv7_train   # name of train dataset

val_dsname=svhn_yuv7_val # name of validation dataset

train=${root}/${train_dsname}_data.mat

train_labels=${root}/${train_dsname}_labels.mat

train_classes=${root}/${train_dsname}_classes.mat

# train_size = 10000            # limit number of samples

val=${root}/${val_dsname}_data.mat

val_labels=${root}/${val_dsname}_labels.mat

val_classes=${root}/${val_dsname}_classes.mat

# val_size = 200                # limit number of samples

# network high level parameters ################################################

classification=1                # 1 is classification, 0 is regression

shared=1     # all modules with the same name share weights

input_thickness=3               # input has 3-channels

# put norm before/after subsampling

aft=1

bef=0

pooling=l4pool # type of pooling to use

nonlin=tanh     # type of nonlinearity to use

shrink=lshrink0     # type of shrink to use

learn_norm=0

learn_mean_norm=0

gaussian_coeff=4

norm=snorm # type of normalization

norm_div=0                   # divisive norm or not

norm_split=1

snorm_div=${norm_div}

snorm_split=${norm_split}

l1=1 # l1 penalty

penalty=${penalty${l1}}

penalty0=

penalty1=l1penalty

l1penalty_threshold=0

l1penalty_coeff=.00001

# network architecture #########################################################

# We describe the following architecture in the section below

############################################################

#                                                                           

#                                                                           

#                                                                           

#    I         *******    *******    *******                                

#    M         *     *    *     *    *     * 

#    A         *  Y  *    *  U  *    *  V  *                                

#    G         *     *    *     *    *     *                                

#    E         ***.***    ***.***    ***.***                                

#                  .         .         .                                       

#                   .        .        .                                        

#  _______      *****************************************   

#   F     |     *     Convolution layer5x5 kernel       *   16 feature maps                            

#   E     |     *****************************************                               

#   A     |     *****************************************                               

#   T     |     *l4pooling  layer 2x2 kernel, 2x2 stride*   16 feature maps                            

#   U     |     ********.***********.********************                               

#   R     |             .            .                                         

#   E     |     ********.*******      .                                        

#         |     *conv layer 7x7*   **********************                               

#   E     |     ********.*******   * l4pool 5x5 kernel  *   512 feature maps                            

#   X     |     ********.*******   *        3x3 stride  *                               

#   T     |     *  l4pool 2x2  *   ***********.**********                               

#   R     |     ********.*******             .                                

#   A     |     ********.*******************.************                              

#   C     |     *     merge layer(merges the two inputs)*                               

#   T     |     *****************************************                               

#   O     |                  .                                          

#   R     |                  .                                          

#   ______|      $$$$$$$$$$$$.$$$$$$$$$$$                               

#         |      $    20 hidden units   $                               

#    C    |      $                      $                               

#    L    |      $$$$$$$$$$$$.$$$$$$$$$$$                   

#    A    |                  .                                          

#    S    |                  .                                          

#    S    |          $$$$$$$$.$$$$$$$                                   

#    I    |          $  10 outputs  $                                   

#    F    |          $              $                                   

#    I    |          $$$$$$$$$$$$$$$$ 

#    E    |                                                                 

#    R    |                                                                 

#   ______|                                                                        

#                                                                           

############################################################

#arch = conv051,addc0,tanh,l4pool22,snorm5,ms2,merge2,linear6,addc6,tanh,linear7,addc7,tanh

arch= ${features},${classifier}

# feature extractors

features=${c05sh},${s022_5},ms2,merge2

# ms2 is a pipe from a sequential layer to two parallel layers 

# which would be joined back at the end of the pipe

ms2_pipe0=${c17sh},${s122_7}

ms2_pipe1=${s153_5}

# classifier

classifier=${f6},${f7}

f6=linear6,addc6,${nonlin}

f7=linear7,addc7,${nonlin}

# features parameters ##########################################################

# convolution layers

c05sh=conv051,addc0,${nonlin}

c17sh=conv171,addc2,${nonlin}

# subsampling layers with norm

s022_5=${norm${bef}5},${pooling}22,${norm${aft}5}

s153_5=${norm${bef}5},${pooling}53,${norm${aft}5}

s122_7=${norm${bef}7},${pooling}22,${norm${aft}7}

norm03=

norm05=

norm06=

norm09=

norm15=${norm}5

norm17=${norm}7

# l4 poolings

l4pool22_kernel=2x2

l4pool22_stride=2x2

l4pool22_energy=${penalty}

l4pool53_kernel=5x5

l4pool53_stride=3x3

l4pool53_energy=${penalty}

# contrast normalizations

snorm5_kernel=5x5

snorm5_learn=${learn_mean_norm}

snorm5_gaussian_coeff=${gaussian_coeff}

snorm5_fsum_div=${norm_div}

snorm5_fsum_split=${norm_split}

snorm7_kernel=7x7

snorm7_learn=${learn_mean_norm}

snorm7_gaussian_coeff=${gaussian_coeff}

snorm7_fsum_div=${norm_div}

snorm7_fsum_split=${norm_split}

# convolutions

conv051_kernel=5x5

conv051_shared=${shared}

conv051_stride=1x1

conv051_table=${table0}

table0=table_3_16_connect_32_fanin_density_0.67_yuv_y8_u0_v0_yuv16_uv0.mat

conv171_kernel=7x7

conv171_shared=${shared}

conv171_stride=1x1

conv171_table=${table1}

table1=table_16_512_connect_1792_fanin_2_3_4_5_density_0.22_random.mat

#16 input nodes

#512 output nodes

#1792 connected nodes

#fanin = Number of inputs connected to each output

#If use fanin don't use density!

# merging

merge_padding=0

merge2_type=mflat

merge2_ins=4x4,4x4

merge2_strides=1x1,1x1

# classifiers parameters #######################################################

linear6_in=8448

linear6_out=20

linear7_in=${linear6_out}

linear7_out=noutputs

# energies & answers ###########################################################

trainer=trainable_module1

trainable_module1_energy=l2_energy

answer=class_answer

# training params ##############################################################

reg_l1=${reg} # L1 regularization

reg_l2=${reg} # L2 regularization

inertia = 0.0 # gradient inertia

anneal_value = 0.0 # learning rate decay value

annea_period = 0 # period (in samples) at which to decay learning rate

gradient_threshold = 0.0

epoch_show_modulo=1000    # print message every n training samples

no_testing_test=0  # do not test on test set if 1

no_training_test=1  # do not test on training set if 1

test_only=0  # if 1, just test the data and return

sample_probabilities=0  # use probabilities to pick samples

hardest_focus=1           # 0: focus on easiest samples 1: focus on hardest ones

ignore_correct=1  # If 1, do not train on correctly classified samples

min_sample_weight=0  # minimum probability of each sample

per_class_norm=1  # normalize probability by class (1) or globally (0)

shuffle_passes=1  # shuffle samples between passes

balanced_training=1  # show each class the same amount of samples or not

random_class_order=1  # class order is randomized or not when balanced

target_factor=1

save_pickings=0  # save sample picking statistics

binary_target=0           # use only 1 output, -1 or +1

save_weights=1    # if 0, do not save weights after each iteration

save_confusion=0  # if 0, do not save confusion matrix after each iter

keep_outputs=0  # keep the predicted outputs in this iteration

fixed_randomization=1  # if 1, uses a fixed randomization seed

training_precision=double # training precision

# training display #############################################################

show_conf=1

show_train=1  # enable/disable all training display

show_train_ninternals=1   # number of internal examples to display

show_train_errors=0  # show worst errors on training set

show_val_errors=1  # show worst errors on validation set

show_val_correct=1  # show worst correct errors on validation set

show_hsample=5  # number of samples to show on height axis

show_wsample=18  # number of samples to show on width axis

Do you know how to fit it?