naomatch / fortlearner Goto Github PK

• mod_kmeans
• mod_breathinig_kmeans
• mod_minibatch_kmeans
• mod_dbscan
• mod_pca
• mod_nipals
• mod_balltree
• mod_brute_force_search
• mod_exact_duplicate_search
• mod_kdtree
• mod_local_outlier_factor
• build_lsh
• mod_product_quantization

ファイルが複数ファイルに分散しているところや、フォルダを分割すべきところができていない

対象モデルのリストアップ中

現在は決定木のHyperparameterのデフォルト値がすべてのアルゴリズムで共通になってしまっている。
これを各アルゴリズムごとに設定できるように変更する。
しない場合max_samplesなどの制御が面倒。
・isolation_treeでは256で設定したい一方、ほかのtreeではhuge or -1で設定しておいたほうが安全

Thanks for your project. It's great to have machine learning algorithms in modern Fortran. On Windows with gfortran installed, I cannot use the make file to create executables. I get errors such as this:

c:\fortran\public_domain\github\FortLearner\sample_scripts>make run_utils
make main_cholesky_decomposition
make[1]: Entering directory 'c:/fortran/public_domain/github/FortLearner/sample_scripts'
gfortran -cpp -D_x86_64 -w -unroll -O3 -ftree-vectorize \
../common/mod_const.f90 ../common/mod_common.f90 ../common/mod_timer.f90 \
../common/mod_random.f90 ../common/mod_sort.f90 \
../math/mod_stats.f90 ../linalg/mod_linalg.f90 \
./inc_sum_up_c.out ./inc_sum_up_left_c.out ./inc_get_minmax.out  \
main_cholesky_decomposition.f90 -o main_cholesky_decomposition.out
c:/equation/bin/../lib/gcc/x86_64-w64-mingw32/12.0.0/../../../../x86_64-w64-mingw32/bin/ld.exe: cannot find ./inc_sum_up_c.out: No such file or directory
c:/equation/bin/../lib/gcc/x86_64-w64-mingw32/12.0.0/../../../../x86_64-w64-mingw32/bin/ld.exe: cannot find ./inc_sum_up_left_c.out: No such file or directory
c:/equation/bin/../lib/gcc/x86_64-w64-mingw32/12.0.0/../../../../x86_64-w64-mingw32/bin/ld.exe: cannot find ./inc_get_minmax.out: No such file or directory
collect2.exe: error: ld returned 1 exit status
make[1]: *** [makefile:71: main_cholesky_decomposition] Error 1
make[1]: Leaving directory 'c:/fortran/public_domain/github/FortLearner/sample_scripts'
make: *** [makefile:25: run_utils] Error 2

rand_uniformなど

pointerにするのは内部で変換して渡せばいい

In the file inc_covariance_value_of_vectors.C I think

#include <stdint.h>

should appear. Without that line I get errors when compiling with gcc such as

inc_covariance_value_of_vectors.C:9:56: error: 'int64_t' has not been declared
    9 | double covariance_loop_C(double vec1[], double vec2[], int64_t n){

• mod_k_nearest_neighbor_classifier
• mod_k_nearest_neighbor_regressor
• mod_kernel_svm
• mod_linear_svm

Hello,
I see you use "Real(kind=8)" and similar a lot. A real kind with value 8 might not exist on a compiler/machine, or might mean different things than what you seem to expect (an 8-byte real). The portable way to use KIND numbers is to ask the compiler, either via SELECTED_REAL_KIND/SELECTED_INT_KIND where you state your desired precision and range, or via the defined constants INT32/INT64/REAL32/REAL64 in intrinsic module ISO_FORTRAN_ENV where you state the storage size in bits needed.

I suggest adding the topic machine-learning in the About section.

Could you include the data files read by the test programs in the repo, for example '../../uci_data/97_make_regression/make_regression_x_0000010000x00100.bin' . Then users would be able to run the test programs.

sklearnでは一致するが、FortLearnerでは一致しない。
似たような値はとる。
アルゴリズムを再度見直す

今
dholder = data_holder(x_train, y_train)

修正した後
dholder = data_holder(x_train)

Thanks for your ongoing work on this project. In the About section at https://github.com/NaoMatch/FortLearner I suggest adding a description such as "machine learning algorithms in Fortran".