isotree 0.3.0 → 0.3.1

data/vendor/isotree/src/c_interface.cpp

@@ -798,13 +798,18 @@ void* isotree_deserialize_from_raw(const char *serialized_model, int nthreads)
         Imputer imputer = Imputer();
         TreesIndexer indexer = TreesIndexer();
 
+        std::unique_ptr<char[]> buffer_metadata;
+        if (size_metadata) {
+            buffer_metadata = std::unique_ptr<char[]>(new char[size_metadata]);
+        }
+
         deserialize_combined(
             serialized_model,
             &model,
             &model_ext,
             &imputer,
             &indexer,
-            (char*)nullptr
+            size_metadata? buffer_metadata.get() : (char*)nullptr
         );
 
         if (!model.trees.size() && !model_ext.hplanes.size())

data/vendor/isotree/src/crit.hpp

@@ -792,6 +792,7 @@ double calc_kurtosis_internal(size_t cnt, int x[], int ncat, size_t buffer_cnt[]
         }
     }
 
+    unreachable();
     return -1; /* this will never be reached, but CRAN complains otherwise */
 }
 
@@ -940,6 +941,7 @@ double calc_kurtosis_weighted_internal(std::vector<ldouble_safe> &buffer_cnt, in
         }
     }
 
+    unreachable();
     return -1; /* this will never be reached, but CRAN complains otherwise */
 }
 
@@ -2858,6 +2860,7 @@ double find_split_dens_longform(int *restrict x, int ncat, size_t *restrict ix_a
     }
 
     /* This will not be reached, but CRAN might complain otherwise */
+    unreachable();
     return -HUGE_VAL;
 }
 
@@ -3052,6 +3055,7 @@ double find_split_dens_longform_weighted(int *restrict x, int ncat, size_t *rest
     }
 
     /* This will not be reached, but CRAN might complain otherwise */
+    unreachable();
     return -HUGE_VAL;
 }
 

data/vendor/isotree/src/digamma.hpp

@@ -0,0 +1,184 @@
+/* Copyright (c) 2001-2002 Enthought, Inc. 2003-2023, SciPy Developers.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions
+are met:
+
+1. Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following
+   disclaimer in the documentation and/or other materials provided
+   with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived
+   from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+Distributed under 3-clause BSD license with permission from the author,
+see https://lists.debian.org/debian-legal/2004/12/msg00295.html
+
+------------------------------------------------------------------
+
+Cephes Math Library Release 2.8:  June, 2000
+Copyright 1984, 1995, 2000 by Stephen L. Moshier
+
+This software is derived from the Cephes Math Library and is
+incorporated herein by permission of the author.
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+  * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+  * Redistributions in binary form must reproduce the above copyright
+    notice, this list of conditions and the following disclaimer in the
+    documentation and/or other materials provided with the distribution.
+  * Neither the name of the <organization> nor the
+    names of its contributors may be used to endorse or promote products
+    derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL <COPYRIGHT HOLDER> BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+static inline void poly6_twice_return_p7
+(
+    const double x,
+    const double *restrict coefs1,
+    const double *restrict coefs2,
+    double &restrict res1,
+    double &restrict res2,
+    double &restrict w7
+)
+{
+    double w[6];
+    w[0] = 1.;
+    w[1] = x;
+    w[2] = x*x;
+    w[3] = x*w[2];
+    w[4] = w[2]*w[2];
+    w[5] = w[2]*w[3];
+    w7 = w[3]*w[3];
+    res1 = 0.;
+    res2 = 0.;
+    for (int ix = 0; ix < 6; ix++)
+    {
+        res1 = std::fma(coefs1[ix], w[ix], res1);
+        res2 = std::fma(coefs2[ix], w[ix], res2);
+    }
+}
+
+static inline double poly7(const double x, const double *restrict coefs)
+{
+    double w[7];
+    w[0] = 1.;
+    w[1] = x;
+    w[2] = x*x;
+    w[3] = x*w[2];
+    w[4] = w[2]*w[2];
+    w[5] = w[2]*w[3];
+    w[6] = w[3]*w[3];
+
+    double out = 0.;
+    for (int ix = 0; ix < 7; ix++)
+        out = std::fma(w[ix], coefs[ix], out);
+    return out;
+}
+
+static const double coefs_12_m[6] = {
+    0.25479851061131551,
+   -0.32555031186804491,
+   -0.65031853770896507,
+   -0.28919126444774784,
+   -0.045251321448739056,
+   -0.0020713321167745952
+};
+
+static const double coefs_12_d[6] = {
+    1.0,
+    2.0767117023730469,
+    1.4606242909763515,
+    0.43593529692665969,
+    0.054151797245674225,
+    0.0021284987017821144
+};
+
+static double coefs_asy[] = {
+    8.33333333333333333333E-2,
+    -8.33333333333333333333E-3,
+    3.96825396825396825397E-3,
+    -4.16666666666666666667E-3,
+    7.57575757575757575758E-3,
+    -2.10927960927960927961E-2,
+    8.33333333333333333333E-2
+};
+
+/* This is implemented only for positive non-integer inputs */
+double digamma(double x)
+{
+    /* check for positive integer up to 64 */
+    if (unlikely((x <= 64) && (x == std::floor(x)))) {
+        return harmonic_recursive(1.0, (double)x) - EULERS_GAMMA;
+    }
+
+    double y = 0.;
+
+    /* use the recurrence relation to move x into [1, 2] */
+    if (x < 1.) {
+        y -= 1. / x;
+        x += 1.;
+    }
+    else if (x < 10.) {
+        while (x > 2.) {
+            x -= 1.;
+            y += 1. / x;
+        }
+    }
+
+    if (x < 1. || x > 2.) {
+        double z = 1. / (x*x);
+        return y + std::log(x) - 0.5/x - z*poly7(z, coefs_asy);
+    }
+
+    const double r1 = 1.46163214463740587234;
+    const double r2 = 0.00000000033095646883;
+    const double r3 = 0.9016312093258695918615325266959189453125e-19;
+    const double Y = 0.99558162689208984;
+    double m, d, p7;
+    poly6_twice_return_p7(
+        x - 1.,
+        coefs_12_m,
+        coefs_12_d,
+        m,
+        d,
+        p7
+    );
+    double r = m / std::fma(p7, -0.55789841321675513e-6, d);
+    double g = x - r1;
+    g -= r2;
+    g -= r3;
+    return y + g*Y + g*r;
+}

data/vendor/isotree/src/dist.hpp

@@ -234,7 +234,7 @@ void calc_similarity(real_t numeric_data[], int categ_data[],
 {
     if (use_long_double && !has_long_double()) {
         use_long_double = false;
-        fprintf(stderr, "Passed 'use_long_double=true', but library was compiled without long double support.\n");
+        print_errmsg("Passed 'use_long_double=true', but library was compiled without long double support.\n");
     }
     #ifndef NO_LONG_DOUBLE
     if (likely(!use_long_double))
@@ -297,7 +297,7 @@ void calc_similarity_internal(
             throw std::runtime_error("Indexer was built without distances. Cannot use references from it.\n");
         else {
             indexer = NULL;
-            fprintf(stderr, "Indexer has no pre-computed distances, will not be used for distance calculations.\n");
+            print_errmsg("Indexer has no pre-computed distances, will not be used for distance calculations.\n");
         }
     }
     if (

data/vendor/isotree/src/fit_model.hpp

@@ -598,7 +598,7 @@ int fit_iforest(IsoForest *model_outputs, ExtIsoForest *model_outputs_ext,
 {
     if (use_long_double && !has_long_double()) {
         use_long_double = false;
-        fprintf(stderr, "Passed 'use_long_double=true', but library was compiled without long double support.\n");
+        print_errmsg("Passed 'use_long_double=true', but library was compiled without long double support.\n");
     }
     #ifndef NO_LONG_DOUBLE
     if (likely(!use_long_double))
@@ -1400,7 +1400,7 @@ int add_tree(IsoForest *model_outputs, ExtIsoForest *model_outputs_ext,
 {
     if (use_long_double && !has_long_double()) {
         use_long_double = false;
-        fprintf(stderr, "Passed 'use_long_double=true', but library was compiled without long double support.\n");
+        print_errmsg("Passed 'use_long_double=true', but library was compiled without long double support.\n");
     }
     #ifndef NO_LONG_DOUBLE
     if (likely(!use_long_double))

data/vendor/isotree/src/helpers_iforest.hpp

@@ -148,6 +148,7 @@ int choose_cat_from_present(WorkerMemory &workspace, InputData &input_data, size
         }
     }
 
+    unreachable();
     return -1; /* this will never be reached, but CRAN complains otherwise */
 }
 

data/vendor/isotree/src/impute.hpp

@@ -145,7 +145,7 @@ void impute_missing_values(real_t numeric_data[], int categ_data[], bool is_col_
 {
     if (use_long_double && !has_long_double()) {
         use_long_double = false;
-        fprintf(stderr, "Passed 'use_long_double=true', but library was compiled without long double support.\n");
+        print_errmsg("Passed 'use_long_double=true', but library was compiled without long double support.\n");
     }
     #ifndef NO_LONG_DOUBLE
     if (likely(!use_long_double))
@@ -313,6 +313,7 @@ void initialize_imputer(Imputer &imputer, InputData &input_data, size_t ntrees,
     imputer.col_modes.resize(input_data.ncols_categ);
     imputer.imputer_tree = std::vector<std::vector<ImputeNode>>(ntrees);
 
+    /* TODO: here should use sample weights if specified as density */
     size_t offset, cnt;
     if (input_data.numeric_data != NULL)
     {
@@ -328,6 +329,7 @@ void initialize_imputer(Imputer &imputer, InputData &input_data, size_t ntrees,
                 cnt -= is_na_or_inf(input_data.numeric_data[row + offset]);
             }
             imputer.col_means[col] /= (ldouble_safe) cnt;
+            if (!cnt) imputer.col_means[col] = NAN;
         }
     }
 
@@ -344,6 +346,7 @@ void initialize_imputer(Imputer &imputer, InputData &input_data, size_t ntrees,
                 cnt -= is_na_or_inf(input_data.Xc[ix]);
             }
             imputer.col_means[col] /= (ldouble_safe) cnt;
+            if (!cnt) imputer.col_means[col] = NAN;
         }
     }
 
@@ -624,7 +627,7 @@ void build_impute_node(ImputeNode &imputer,    WorkerMemory &workspace,
                             imputer.cat_sum[col][cat]
                                 +=
                             (imputer_tree[curr_tree].cat_sum[col][cat] > 0)?
-                                (imputer_tree[curr_tree].cat_sum[col][cat] / imputer.cat_weight[col]) : 0.;
+                                (imputer_tree[curr_tree].cat_sum[col][cat] / imputer_tree[curr_tree].cat_weight[col]) : 0.;
                             imputer.cat_weight[col]    =  wsum / (double)(2 * look_aboves);
                         }
                         break;
@@ -999,6 +1002,11 @@ void apply_imputation_results(std::vector<ImputedData> &impute_vec,
 }
 
 
+/* TODO: investigate why in the case of all-missing numeric columns the node weights still
+   get filled when using extended model, then remove the workaround that was added here that
+   checks if the sum is zero and column is all-nan. Should also modify the earlier code to
+   remove these cases from the imputation tracking list when doing the imputations on-the-fly
+   as the model is being fit. */
 template <class PredictionData, class ImputedData>
 void apply_imputation_results(PredictionData  &prediction_data,
                               ImputedData     &imp,
@@ -1012,7 +1020,7 @@ void apply_imputation_results(PredictionData  &prediction_data,
         for (size_t ix = 0; ix < imp.n_missing_num; ix++)
         {
             col = imp.missing_num[ix];
-            if (imp.num_weight[ix] > 0 && !is_na_or_inf(imp.num_sum[ix]))
+            if (imp.num_weight[ix] > 0 && !is_na_or_inf(imp.num_sum[ix]) && !(imp.num_sum[ix] == 0 && std::isnan(imputer.col_means[col])))
                 prediction_data.numeric_data[row + col * prediction_data.nrows]
                     =
                 imp.num_sum[ix] / imp.num_weight[ix];
@@ -1028,7 +1036,7 @@ void apply_imputation_results(PredictionData  &prediction_data,
         for (size_t ix = 0; ix < imp.n_missing_num; ix++)
         {
             col = imp.missing_num[ix];
-            if (imp.num_weight[ix] > 0 && !is_na_or_inf(imp.num_sum[ix]))
+            if (imp.num_weight[ix] > 0 && !is_na_or_inf(imp.num_sum[ix]) && !(imp.num_sum[ix] == 0 && std::isnan(imputer.col_means[col])))
                 prediction_data.numeric_data[col + row * imputer.ncols_numeric]
                     =
                 imp.num_sum[ix] / imp.num_weight[ix];
@@ -1066,10 +1074,20 @@ void apply_imputation_results(PredictionData  &prediction_data,
             std::distance(imp.cat_sum[col].begin(),
                           std::max_element(imp.cat_sum[col].begin(), imp.cat_sum[col].end()));
 
-            if (prediction_data.categ_data[row + col * prediction_data.nrows] == 0 && imp.cat_sum[col][0] <= 0)
-                prediction_data.categ_data[row + col * prediction_data.nrows]
-                    =
-                imputer.col_modes[col];
+            if (prediction_data.categ_data[row + col * prediction_data.nrows] == 0)
+            {
+                if (imp.cat_sum.empty() || imp.cat_sum[col].empty())
+                {
+                    prediction_data.categ_data[row + col * prediction_data.nrows] = -1;
+                }
+
+                else if (imp.cat_sum[col][0] <= 0)
+                {
+                    prediction_data.categ_data[row + col * prediction_data.nrows]
+                        =
+                    imputer.col_modes[col];
+                }
+            }
         }
     }
 
@@ -1083,10 +1101,20 @@ void apply_imputation_results(PredictionData  &prediction_data,
             std::distance(imp.cat_sum[col].begin(),
                           std::max_element(imp.cat_sum[col].begin(), imp.cat_sum[col].end()));
 
-            if (prediction_data.categ_data[col + row * imputer.ncols_categ] == 0 && imp.cat_sum[col][0] <= 0)
-                prediction_data.categ_data[col + row * imputer.ncols_categ]
-                    =
-                imputer.col_modes[col];
+            if (prediction_data.categ_data[col + row * imputer.ncols_categ] == 0)
+            {
+                if (imp.cat_sum.empty() || imp.cat_sum[col].empty())
+                {
+                    prediction_data.categ_data[col + row * imputer.ncols_categ] = -1;
+                }
+                
+                else if (imp.cat_sum[col][0] <= 0)
+                {
+                    prediction_data.categ_data[col + row * imputer.ncols_categ]
+                        =
+                    imputer.col_modes[col];
+                }
+            }
         }
     }
 }

data/vendor/isotree/src/isotree.hpp

@@ -109,15 +109,13 @@
     extern "C" {
         #include <R_ext/Print.h>
     }
-    #define printf Rprintf
-    #define fprintf(f, message) REprintf(message)
+    #define print_errmsg(msg) REprintf("%s", msg)
 #elif defined(_FOR_PYTHON)
-    extern "C" void cy_warning(const char *msg);
-    #define fprintf(f, message) cy_warning(message)
+    extern "C" int cy_warning(const char *msg);
+    #define print_errmsg cy_warning
 #else
     #include <cstdio>
-    using std::printf;
-    using std::fprintf;
+    #define print_errmsg(msg) std::fprintf(stderr, "%s", msg)
 #endif
 #ifdef _OPENMP
     #include <omp.h>
@@ -285,6 +283,19 @@ using std::memcpy;
     #endif
 #endif
 
+#ifndef unreachable
+#   if (__cplusplus >= 202309L)
+#       include <utility>
+        using std::unreachable;
+#   elif defined(_MSC_VER)
+#       define unreachable() (__assume(false))
+#   elif defined(__GNUC__) || defined(__clang__)
+#       define unreachable() __builtin_unreachable()
+#   else
+#       define unreachable() 
+#   endif
+#endif
+
 
 /*   Apple at some point decided to drop OMP library and headers from its compiler distribution
 *    and to alias 'gcc' to 'clang', which work differently when given flags they cannot interpret,
@@ -1936,8 +1947,7 @@ public:
     {
         if (this->handle) {
             int err = std::fclose(this->handle);
-            if (err)
-                fprintf(stderr, "Error: could not close file.\n");
+            if (err) print_errmsg("Error: could not close file.\n");
         }
         this->handle = NULL;
     }
@@ -1960,8 +1970,7 @@ public:
         {
             if (this->handle) {
                 int err = std::fclose(this->handle);
-                if (err)
-                    fprintf(stderr, "Error: could not close file.\n");
+                if (err) print_errmsg("Error: could not close file.\n");
             }
             this->handle = NULL;
         }

data/vendor/isotree/src/mult.hpp

@@ -598,7 +598,7 @@ void add_linear_comb_weighted(size_t ix_arr[], size_t st, size_t end, double *re
                     w_this = w[ix_arr[row]];
                     res_write[row]     = std::fma(x[ix_arr[row]] - x_mean, coef, res_write[row]);
                     obs_weight[cnt]    = w_this;
-                    buffer_arr[cnt++]  =  x[ix_arr[row]];
+                    buffer_arr[cnt++]  = x[ix_arr[row]];
                     cumw += w_this;
                 }
 

data/vendor/isotree/src/oop_interface.cpp

@@ -628,14 +628,12 @@ void IsolationForest::check_nthreads()
         #endif
     }
     if (nthreads <= 0) {
-        fprintf(stderr, "'isotree' got invalid 'nthreads', will set to 1.\n");
+        print_errmsg("'isotree' got invalid 'nthreads', will set to 1.\n");
         this->nthreads = 1;
     }
     #ifndef _OPENMP
     else if (nthreads > 1) {
-        fprintf(stderr,
-                "Passed nthreads:%d to 'isotree', but library was compiled without multithreading.\n",
-                this->nthreads);
+        print_errmsg("Passed nthreads>1 to 'isotree', but library was compiled without multithreading.\n");
         this->nthreads = 1;
     }
     #endif
@@ -787,6 +785,10 @@ IsolationForest IsolationForest::deserialize_template(itype &inp, int nthreads)
     ExtIsoForest model_ext = ExtIsoForest();
     Imputer imputer = Imputer();
     TreesIndexer indexer = TreesIndexer();
+    std::unique_ptr<char[]> buffer_metadata;
+    if (size_metadata) {
+        buffer_metadata = std::unique_ptr<char[]>(new char[size_metadata]);
+    }
 
     deserialize_combined(
         inp,
@@ -794,7 +796,7 @@ IsolationForest IsolationForest::deserialize_template(itype &inp, int nthreads)
         &model_ext,
         &imputer,
         &indexer,
-        (char*)nullptr
+        size_metadata? buffer_metadata.get() : (char*)nullptr
     );
 
     if (model.trees.empty() && model_ext.hplanes.empty())

data/vendor/isotree/src/python_helpers.hpp

@@ -92,36 +92,6 @@ TreesIndexer get_Indexer()
     return TreesIndexer();
 }
 
-/* Reason behind these functions: Cython (as of v0.29) will not auto-deallocate
-   structs which are part of a cdef'd class, which produces a memory leak
-   but can be force-destructed. Unfortunately, Cython itself doesn't even
-   allow calling destructors for structs, so it has to be done externally.
-   These functions should otherwise have no reason to be.
-
-   This is supposed to be already fixed in newer Cython versions:
-   https://github.com/cython/cython/issues/3226
-   But is not yet available in the relase versions at the time of writing */
-
-void dealloc_IsoForest(IsoForest &model_outputs)
-{
-    model_outputs.~IsoForest();
-}
-
-void dealloc_IsoExtForest(ExtIsoForest &model_outputs_ext)
-{
-    model_outputs_ext.~ExtIsoForest();
-}
-
-void dealloc_Imputer(Imputer &imputer)
-{
-    imputer.~Imputer();
-}
-
-void dealloc_Indexer(TreesIndexer &indexer)
-{
-    indexer.~TreesIndexer();
-}
-
 bool get_has_openmp(void)
 {
     #ifdef _OPENMP
@@ -131,4 +101,14 @@ bool get_has_openmp(void)
     #endif
 }
 
+size_t py_strerrorlen_s()
+{
+    return strlen(strerror(errno));
+}
+
+void copy_errno_msg(char *inp)
+{
+    strcpy(inp, strerror(errno));
+}
+
 #endif

data/vendor/isotree/src/robinmap/README.md

@@ -90,7 +90,7 @@ target_link_libraries(your_target PRIVATE tsl::robin_map)
 
 If the project has been installed through `make install`, you can also use `find_package(tsl-robin-map REQUIRED)` instead of `add_subdirectory`.
 
-The library is available in [vcpkg](https://github.com/Microsoft/vcpkg/tree/master/ports/robin-map) and [conan](https://bintray.com/tessil/tsl/tsl-robin-map%3Atessil). It's also present in [Debian](https://packages.debian.org/buster/robin-map-dev), [Ubuntu](https://packages.ubuntu.com/disco/robin-map-dev) and [Fedora](https://apps.fedoraproject.org/packages/robin-map-devel) package repositories.
+The library is available in [vcpkg](https://github.com/Microsoft/vcpkg/tree/master/ports/robin-map) and [conan](https://conan.io/center/tsl-robin-map). It's also present in [Debian](https://packages.debian.org/buster/robin-map-dev), [Ubuntu](https://packages.ubuntu.com/disco/robin-map-dev) and [Fedora](https://apps.fedoraproject.org/packages/robin-map-devel) package repositories.
 
 The code should work with any C++11 standard-compliant compiler and has been tested with GCC 4.8.4, Clang 3.5.0 and Visual Studio 2015.