Fix attribute name extraction in ArffFiles

Add hyperparameters to discretized

Hyperparameters added:

- min_length: int, default=3: The minimum length of an interval to be considered to be discretized.
- max_depth: int, default=1e6: The maximum depth of the discretization process.
- max_cuts: float, default=0: The maximum number of cut points to be computed for each feature. Compute all cutpoints and select the ones that produce less entropy

.github/workflows/main.yml

@@ -20,14 +20,14 @@ jobs:
         with:
           submodules: recursive
       - name: Set up Python ${{ matrix.python }}
-        uses: actions/setup-python@v2
+        uses: actions/setup-python@v4
         with:
           python-version: ${{ matrix.python }}
       - name: Install dependencies
         run: |
           pip install -q --upgrade pip
           pip install -q scikit-learn cython
-          pip install -q --upgrade codecov coverage black flake8 codacy-coverage
+          pip install -q coverage black flake8 codacy-coverage
       - name: Build and install
         run: |
           make install
@@ -40,7 +40,7 @@ jobs:
           coverage run -m unittest discover -v -s src
           coverage xml
       - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v1
+        uses: codecov/codecov-action@v3
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
           files: ./coverage.xml

.gitignore

@@ -33,8 +33,8 @@ MANIFEST
 *.manifest
 *.spec
 
-# Installer log2s
+# Installer logs
-pip-log2.txt
+pip-log.txt
 pip-delete-this-directory.txt
 
 # Unit test / coverage reports
@@ -56,7 +56,7 @@ coverage.xml
 *.pot
 
 # Django stuff:
-*.log2
+*.log
 local_settings.py
 db.sqlite3
 db.sqlite3-journal
@@ -134,4 +134,6 @@ cmake-build-debug
 cmake-build-debug/**
 **/lcoverage/**
 **/x/*
-**/*.so
+**/*.so
+**/CMakeFiles
+wheelhouse

MANIFEST.in

@@ -0,0 +1,5 @@
+include src/cppmdlp/CPPFImdlp.h
+include src/cppmdlp/typesFImdlp.h
+include src/cppmdlp/Metrics.h
+include src/fimdlp/Factorize.h
+include src/fimdlp/ArffFiles.h

Makefile

@@ -37,6 +37,12 @@ install:  ## Build extension
 audit: ## Audit pip
 	pip-audit
 
+version:
+	@echo "Current Python version .: $(shell python --version)"
+	@echo "Current FImdlp version .: $(shell python -c "from fimdlp import _version; print(_version.__version__)")"
+	@echo "Current mdlp version ...: $(shell python -c "from fimdlp.cppfimdlp import CFImdlp; print(CFImdlp().get_version().decode())")"
+	@echo "Installed FImdlp version: $(shell pip show fimdlp | grep Version | cut -d' ' -f2)"
+
 help: ## Show help message
 	@IFS=$$'\n' ; \
 	help_lines=(`fgrep -h "##" $(MAKEFILE_LIST) | fgrep -v fgrep | sed -e 's/\\$$//' | sed -e 's/##/:/'`); \

README.md

@@ -3,14 +3,14 @@
 [![CodeQL](https://github.com/Doctorado-ML/FImdlp/actions/workflows/codeql.yml/badge.svg)](https://github.com/Doctorado-ML/FImdlp/actions/workflows/codeql.yml)
 [![Codacy Badge](https://app.codacy.com/project/badge/Grade/8b4d784fee13401588aa8c06532a2f6d)](https://www.codacy.com/gh/Doctorado-ML/FImdlp/dashboard?utm_source=github.com&utm_medium=referral&utm_content=Doctorado-ML/FImdlp&utm_campaign=Badge_Grade)
 [![codecov](https://codecov.io/gh/Doctorado-ML/FImdlp/branch/main/graph/badge.svg?token=W8I45B5Z3J)](https://codecov.io/gh/Doctorado-ML/FImdlp)
-[![pypy](https://img.shields.io/pypi/v/FImdlp?color=g)](https://img.shields.io/pypi/v/FImdlp?color=g)
+[![pypy](https://img.shields.io/pypi/v/FImdlp?color=g)](https://pypi.org/project/FImdlp)
 ![https://img.shields.io/badge/python-3.9%2B-blue](https://img.shields.io/badge/python-3.9%2B-brightgreen)
 
 Discretization algorithm based on the paper by Usama M. Fayyad and Keki B. Irani 
 
-```
+
 Multi-Interval Discretization of Continuous-Valued Attributes for Classification Learning. In Proceedings of the 13th International Joint Conference on Artificial Intelligence (IJCAI-95), pages 1022-1027, Montreal, Canada, August 1995.
-```
+
 
 ## Installation
 
@@ -24,8 +24,8 @@ git clone --recurse-submodules https://github.com/doctorado-ml/FImdlp.git
 
 ```bash
 pip install -e .
-python samples/sample.py iris --original 
+python samples/sample.py iris  
-python samples/sample.py iris --proposal
+python samples/sample.py iris -c 2
 python samples/sample.py -h # for more options
 ```
 
@@ -33,9 +33,12 @@ python samples/sample.py -h # for more options
 
 ```bash
 cd samples
-mkdir build
+cmake -B build
 cd build
-cmake ..
 make
-./sample iris
+./sample -f iris -c 2
+./sample -h
 ```
+
+### Based on
+[https://github.com/rmontanana/mdlp](https://github.com/rmontanana/mdlp)

pyproject.toml

@@ -18,10 +18,10 @@ authors = [
     { name = "Ricardo Montañana", email = "ricardo.montanana@alu.uclm.es" },
 ]
 dynamic = ['version']
-dependencies = ["numpy", "joblib"]
+dependencies = ["numpy", "joblib", "scikit-learn"]
 requires-python = ">=3.9"
 classifiers = [
-    "Development Status :: 3 - Alpha",
+    "Development Status :: 4 - Beta",
     "Intended Audience :: Science/Research",
     "Intended Audience :: Developers",
     "Topic :: Software Development",
@@ -33,14 +33,16 @@ classifiers = [
     "Programming Language :: Python",
     "Programming Language :: Python :: 3.9",
     "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
 ]
 
 [project.urls]
 Home = "https://github.com/doctorado-ml/FImdlp"
+Base = "https://github.com/rmontanana/mdlp"
 
 [tool.black]
 line-length = 79
-target_version = ['py39', 'py310']
+target_version = ['py39', 'py310', 'py311']
 include = '\.pyi?$'
 exclude = '''
 /(

samples/ArffFiles.cpp

@@ -1,117 +0,0 @@
-#include "ArffFiles.h"
-
-#include <fstream>
-#include <sstream>
-#include <map>
-#include <iostream>
-
-using namespace std;
-
-ArffFiles::ArffFiles()
-{
-}
-vector<string> ArffFiles::getLines()
-{
-    return lines;
-}
-unsigned long int ArffFiles::getSize()
-{
-    return lines.size();
-}
-vector<tuple<string, string>> ArffFiles::getAttributes()
-{
-    return attributes;
-}
-string ArffFiles::getClassName()
-{
-    return className;
-}
-string ArffFiles::getClassType()
-{
-    return classType;
-}
-vector<vector<float>>& ArffFiles::getX()
-{
-    return X;
-}
-vector<int>& ArffFiles::getY()
-{
-    return y;
-}
-void ArffFiles::load(string fileName, bool classLast)
-{
-    ifstream file(fileName);
-    string keyword, attribute, type;
-    if (file.is_open()) {
-        string line;
-        while (getline(file, line)) {
-            if (line[0] == '%' || line.empty() || line == "\r" || line == " ") {
-                continue;
-            }
-            if (line.find("@attribute") != string::npos || line.find("@ATTRIBUTE") != string::npos) {
-                stringstream ss(line);
-                ss >> keyword >> attribute >> type;
-                attributes.push_back(make_tuple(attribute, type));
-                continue;
-            }
-            if (line[0] == '@') {
-                continue;
-            }
-            lines.push_back(line);
-        }
-        file.close();
-        if (attributes.empty())
-            throw invalid_argument("No attributes found");
-        if (classLast) {
-            className = get<0>(attributes.back());
-            classType = get<1>(attributes.back());
-            attributes.pop_back();
-        } else {
-            className = get<0>(attributes.front());
-            classType = get<1>(attributes.front());
-            attributes.erase(attributes.begin());
-        }
-        generateDataset(classLast);
-    } else
-        throw invalid_argument("Unable to open file");
-}
-void ArffFiles::generateDataset(bool classLast)
-{
-    X = vector<vector<float>>(attributes.size(), vector<float>(lines.size()));
-    vector<string> yy = vector<string>(lines.size(), "");
-    int labelIndex = classLast ? attributes.size() : 0;
-    for (int i = 0; i < lines.size(); i++) {
-        stringstream ss(lines[i]);
-        string value;
-        int pos = 0, xIndex = 0;
-        while (getline(ss, value, ',')) {
-            if (pos++ == labelIndex) {
-                yy[i] = value;
-            } else {
-                X[xIndex++][i] = stof(value);
-            }
-        }
-    }
-    y = factorize(yy);
-}
-string ArffFiles::trim(const string& source)
-{
-    string s(source);
-    s.erase(0, s.find_first_not_of(" \n\r\t"));
-    s.erase(s.find_last_not_of(" \n\r\t") + 1);
-    return s;
-}
-vector<int> ArffFiles::factorize(const vector<string>& labels_t)
-{
-    vector<int> yy;
-    yy.reserve(labels_t.size());
-    map<string, int> labelMap;
-    int i = 0;
-    for (string label : labels_t) {
-        if (labelMap.find(label) == labelMap.end()) {
-            labelMap[label] = i++;
-        }
-        yy.push_back(labelMap[label]);
-    }
-    return yy;
-}

samples/ArffFiles.h

@@ -1,28 +0,0 @@
-#ifndef ARFFFILES_H
-#define ARFFFILES_H
-#include <string>
-#include <vector>
-#include <tuple>
-using namespace std;
-class ArffFiles {
-private:
-    vector<string> lines;
-    vector<tuple<string, string>> attributes;
-    string className, classType;
-    vector<vector<float>> X;
-    vector<int> y;
-    void generateDataset(bool);
-public:
-    ArffFiles();
-    void load(string, bool = true);
-    vector<string> getLines();
-    unsigned long int getSize();
-    string getClassName();
-    string getClassType();
-    string trim(const string&);
-    vector<vector<float>>& getX();
-    vector<int>& getY();
-    vector<tuple<string, string>> getAttributes();
-    vector<int> factorize(const vector<string>& labels_t);
-};
-#endif

samples/CMakeLists.txt

@@ -1,6 +1,7 @@
 cmake_minimum_required(VERSION 3.20)
-project(main)
+project(sample)
 
-set(CMAKE_CXX_STANDARD 14)
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_BUILD_TYPE Debug)
 
-add_executable(sample sample.cpp ArffFiles.cpp ../src/cppmdlp/Metrics.cpp ../src/cppmdlp/CPPFImdlp.cpp)
+add_executable(sample sample.cpp ../src/cppmdlp/tests/ArffFiles.cpp ../src/cppmdlp/Metrics.cpp ../src/cppmdlp/CPPFImdlp.cpp)

samples/sample.cpp

@@ -1,30 +1,101 @@
-#include "ArffFiles.h"
 #include <iostream>
 #include <vector>
 #include <iomanip>
+#include <chrono>
+#include <algorithm>
+#include <cstring>
+#include <getopt.h>
 #include "../src/cppmdlp/CPPFImdlp.h"
+#include "../src/cppmdlp/tests/ArffFiles.h"
 
 using namespace std;
+using namespace mdlp;
 
-int main(int argc, char** argv)
+const string PATH = "../../src/cppmdlp/tests/datasets/";
+
+/* print a description of all supported options */
+void usage(const char* path)
+{
+    /* take only the last portion of the path */
+    const char* basename = strrchr(path, '/');
+    basename = basename ? basename + 1 : path;
+
+    cout << "usage: " << basename << "[OPTION]" << endl;
+    cout << "  -h, --help\t\t Print this help and exit." << endl;
+    cout
+        << "  -f, --file[=FILENAME]\t {all, glass, iris, kdd_JapaneseVowels, letter, liver-disorders, mfeat-factors, test}."
+        << endl;
+    cout << "  -p, --path[=FILENAME]\t folder where the arff dataset is located, default " << PATH << endl;
+    cout << "  -m, --max_depth=INT\t max_depth pased to discretizer. Default = MAX_INT" << endl;
+    cout
+        << "  -c, --max_cutpoints=FLOAT\t percentage of lines expressed in decimal or integer number or cut points. Default = 0 -> any"
+        << endl;
+    cout << "  -n, --min_length=INT\t interval min_length pased to discretizer. Default = 3" << endl;
+}
+
+tuple<string, string, int, int, float> parse_arguments(int argc, char** argv)
+{
+    string file_name;
+    string path = PATH;
+    int max_depth = numeric_limits<int>::max();
+    int min_length = 3;
+    float max_cutpoints = 0;
+    const vector<struct option> long_options = {
+            {"help",          no_argument,       nullptr, 'h'},
+            {"file",          required_argument, nullptr, 'f'},
+            {"path",          required_argument, nullptr, 'p'},
+            {"max_depth",     required_argument, nullptr, 'm'},
+            {"max_cutpoints", required_argument, nullptr, 'c'},
+            {"min_length",    required_argument, nullptr, 'n'},
+            {nullptr,         no_argument,       nullptr, 0}
+    };
+    while (true) {
+        const auto c = getopt_long(argc, argv, "hf:p:m:c:n:", long_options.data(), nullptr);
+        if (c == -1)
+            break;
+        switch (c) {
+            case 'h':
+                usage(argv[0]);
+                exit(0);
+            case 'f':
+                file_name = string(optarg);
+                break;
+            case 'm':
+                max_depth = stoi(optarg);
+                break;
+            case 'n':
+                min_length = stoi(optarg);
+                break;
+            case 'c':
+                max_cutpoints = stof(optarg);
+                break;
+            case 'p':
+                path = optarg;
+                if (path.back() != '/')
+                    path += '/';
+                break;
+            case '?':
+                usage(argv[0]);
+                exit(1);
+            default:
+                abort();
+        }
+    }
+    if (file_name.empty()) {
+        usage(argv[0]);
+        exit(1);
+    }
+    return make_tuple(file_name, path, max_depth, min_length, max_cutpoints);
+}
+
+void process_file(const string& path, const string& file_name, bool class_last, int max_depth, int min_length,
+    float max_cutpoints)
 {
     ArffFiles file;
-    vector<string> lines;
-    string path = "../../src/cppmdlp/tests/datasets/";
-    map<string, bool > datasets = {
-        {"mfeat-factors", true},
-        {"iris", true},
-        {"letter", true},
-        {"kdd_JapaneseVowels", false}
-    };
-    if (argc != 2 || datasets.find(argv[1]) == datasets.end()) {
-        cout << "Usage: " << argv[0] << " {mfeat-factors, iris, letter, kdd_JapaneseVowels}" << endl;
-        return 1;
-    }
 
-    file.load(path + argv[1] + ".arff", datasets[argv[1]]);
+    file.load(path + file_name + ".arff", class_last);
-    auto attributes = file.getAttributes();
+    const auto attributes = file.getAttributes();
-    int items = file.getSize();
+    const auto items = file.getSize();
     cout << "Number of lines: " << items << endl;
     cout << "Attributes: " << endl;
     for (auto attribute : attributes) {
@@ -33,22 +104,93 @@ int main(int argc, char** argv)
     cout << "Class name: " << file.getClassName() << endl;
     cout << "Class type: " << file.getClassType() << endl;
     cout << "Data: " << endl;
-    vector<vector<float>>& X = file.getX();
+    vector<samples_t>& X = file.getX();
-    vector<int>& y = file.getY();
+    labels_t& y = file.getY();
-    for (int i = 0; i < 50; i++) {
+    for (int i = 0; i < 5; i++) {
         for (auto feature : X) {
             cout << fixed << setprecision(1) << feature[i] << " ";
         }
         cout << y[i] << endl;
     }
-    mdlp::CPPFImdlp test = mdlp::CPPFImdlp(false);
+    auto test = mdlp::CPPFImdlp(min_length, max_depth, max_cutpoints);
+    size_t total = 0;
     for (auto i = 0; i < attributes.size(); i++) {
-        cout << "Cut points for " << get<0>(attributes[i]) << endl;
+        auto min_max = minmax_element(X[i].begin(), X[i].end());
-        cout << "--------------------------" << setprecision(3) << endl;
+        cout << "Cut points for feature " << get<0>(attributes[i]) << ": [" << setprecision(3);
         test.fit(X[i], y);
-        for (auto item : test.getCutPoints()) {
+        auto cut_points = test.getCutPoints();
-            cout << item << endl;
+        for (auto item : cut_points) {
+            cout << item;
+            if (item != cut_points.back())
+                cout << ", ";
         }
+        total += test.getCutPoints().size();
+        cout << "]" << endl;
+        cout << "Min: " << *min_max.first << " Max: " << *min_max.second << endl;
+        cout << "--------------------------" << endl;
+    }
+    cout << "Total cut points ...: " << total << endl;
+    cout << "Total feature states: " << total + attributes.size() << endl;
+}
+
+void process_all_files(const map<string, bool>& datasets, const string& path, int max_depth, int min_length,
+    float max_cutpoints)
+{
+    cout << "Results: " << "Max_depth: " << max_depth << "  Min_length: " << min_length << "  Max_cutpoints: "
+        << max_cutpoints << endl << endl;
+    printf("%-20s %4s %4s\n", "Dataset", "Feat", "Cuts Time(ms)");
+    printf("==================== ==== ==== ========\n");
+    for (const auto& dataset : datasets) {
+        ArffFiles file;
+        file.load(path + dataset.first + ".arff", dataset.second);
+        auto attributes = file.getAttributes();
+        vector<samples_t>& X = file.getX();
+        labels_t& y = file.getY();
+        size_t timing = 0;
+        size_t cut_points = 0;
+        for (auto i = 0; i < attributes.size(); i++) {
+            auto test = mdlp::CPPFImdlp(min_length, max_depth, max_cutpoints);
+            std::chrono::steady_clock::time_point begin = std::chrono::steady_clock::now();
+            test.fit(X[i], y);
+            std::chrono::steady_clock::time_point end = std::chrono::steady_clock::now();
+            timing += std::chrono::duration_cast<std::chrono::milliseconds>(end - begin).count();
+            cut_points += test.getCutPoints().size();
+        }
+        printf("%-20s %4lu %4zu %8zu\n", dataset.first.c_str(), attributes.size(), cut_points, timing);
+    }
+}
+
+
+int main(int argc, char** argv)
+{
+    map<string, bool> datasets = {
+            {"glass",              true},
+            {"iris",               true},
+            {"kdd_JapaneseVowels", false},
+            {"letter",             true},
+            {"liver-disorders",    true},
+            {"mfeat-factors",      true},
+            {"test",               true}
+    };
+    string file_name;
+    string path;
+    int max_depth;
+    int min_length;
+    float max_cutpoints;
+    tie(file_name, path, max_depth, min_length, max_cutpoints) = parse_arguments(argc, argv);
+    if (datasets.find(file_name) == datasets.end() && file_name != "all") {
+        cout << "Invalid file name: " << file_name << endl;
+        usage(argv[0]);
+        exit(1);
+    }
+    if (file_name == "all")
+        process_all_files(datasets, path, max_depth, min_length, max_cutpoints);
+    else {
+        process_file(path, file_name, datasets[file_name], max_depth, min_length, max_cutpoints);
+        cout << "File name ....: " << file_name << endl;
+        cout << "Max depth ....: " << max_depth << endl;
+        cout << "Min length ...: " << min_length << endl;
+        cout << "Max cutpoints : " << max_cutpoints << endl;
     }
     return 0;
-}
+}

samples/sample.py

@@ -1,43 +1,72 @@
 import time
 import argparse
 import os
-from scipy.io import arff
 import pandas as pd
 from sklearn.ensemble import RandomForestClassifier
 from fimdlp.mdlp import FImdlp
+from fimdlp.cppfimdlp import CArffFiles
 
 datasets = {
     "mfeat-factors": True,
     "iris": True,
+    "glass": True,
+    "liver-disorders": True,
     "letter": True,
     "kdd_JapaneseVowels": False,
 }
 
 ap = argparse.ArgumentParser()
-ap.add_argument("--proposal", action="store_true")
+ap.add_argument(
-ap.add_argument("--original", dest="proposal", action="store_false")
+    "-n",
+    "--min_length",
+    type=int,
+    default=3,
+    help="Minimum length of interval",
+)
+ap.add_argument(
+    "-m", "--max_depth", type=int, default=9999, help="Maximum depth"
+)
+ap.add_argument(
+    "-c",
+    "--max_cuts",
+    type=float,
+    default=0,
+    help="Maximum number of cut points",
+)
 ap.add_argument("dataset", type=str, choices=datasets.keys())
 args = ap.parse_args()
 relative = "" if os.path.isdir("src") else ".."
 file_name = os.path.join(
     relative, "src", "cppmdlp", "tests", "datasets", args.dataset
 )
-data = arff.loadarff(file_name + ".arff")
+arff = CArffFiles()
-df = pd.DataFrame(data[0])
+arff.load(bytes(f"{file_name}.arff", "utf-8"))
-class_column = -1 if datasets[args.dataset] else 0
+X = arff.get_X()
-class_name = df.columns.to_list()[class_column]
+y = arff.get_y()
-X = df.drop(class_name, axis=1)
+attributes = arff.get_attributes()
-y, _ = pd.factorize(df[class_name])
+attributes = [x[0].decode() for x in attributes]
-X = X.to_numpy()
+df = pd.DataFrame(X, columns=attributes)
-test = FImdlp(proposal=args.proposal)
+class_name = arff.get_class_name().decode()
+df[class_name] = y
+test = FImdlp(
+    min_length=args.min_length,
+    max_depth=args.max_depth,
+    max_cuts=args.max_cuts,
+)
 now = time.time()
 test.fit(X, y)
 fit_time = time.time()
-print("Fitting: ", fit_time - now)
+print(f"Fitting ....: {fit_time - now:7.5f} seconds")
 now = time.time()
 Xt = test.transform(X)
-print("Transforming: ", time.time() - now)
+print(f"Transforming: {time.time() - now:7.5f} seconds")
-print(test.get_cut_points())
+cut_points = test.get_cut_points()
+for i, cuts in enumerate(cut_points):
+    print(f"Cut points for feature {attributes[i]}: {cuts}")
+    print(f"Min: {min(X[:, i]):6.4f} Max: {max(X[:, i]):6.4f}")
+num_cuts = sum([len(x) for x in cut_points])
+print(f"Total cut points ...: {num_cuts}")
+print(f"Total feature states: {num_cuts + len(attributes)}")
 clf = RandomForestClassifier(random_state=0)
 print(
     "Random Forest score with discretized data: ", clf.fit(Xt, y).score(Xt, y)

setup.py

@@ -14,10 +14,14 @@ setup(
                 "src/fimdlp/cfimdlp.pyx",
                 "src/cppmdlp/CPPFImdlp.cpp",
                 "src/cppmdlp/Metrics.cpp",
+                "src/fimdlp/Factorize.cpp",
+                "src/fimdlp/ArffFiles.cpp",
             ],
             language="c++",
             include_dirs=["fimdlp"],
-            extra_compile_args=["-std=c++2a"],
+            extra_compile_args=[
+                "-std=c++11",
+            ],
         ),
     ]
 )

src/fimdlp/ArffFiles.cpp

@@ -0,0 +1,132 @@
+#include "ArffFiles.h"
+#include <fstream>
+#include <sstream>
+#include <map>
+
+using namespace std;
+
+ArffFiles::ArffFiles() = default;
+
+vector<string> ArffFiles::getLines() const
+{
+    return lines;
+}
+
+unsigned long int ArffFiles::getSize() const
+{
+    return lines.size();
+}
+
+vector<pair<string, string>> ArffFiles::getAttributes() const
+{
+    return attributes;
+}
+
+string ArffFiles::getClassName() const
+{
+    return className;
+}
+
+string ArffFiles::getClassType() const
+{
+    return classType;
+}
+
+vector<vector<float>>& ArffFiles::getX()
+{
+    return X;
+}
+
+vector<int>& ArffFiles::getY()
+{
+    return y;
+}
+
+void ArffFiles::load(const string& fileName, bool classLast)
+{
+    ifstream file(fileName);
+    if (!file.is_open()) {
+        throw invalid_argument("Unable to open file");
+    }
+    string line;
+    string keyword;
+    string attribute;
+    string type;
+    string type_w;
+    while (getline(file, line)) {
+        if (line.empty() || line[0] == '%' || line == "\r" || line == " ") {
+            continue;
+        }
+        if (line.find("@attribute") != string::npos || line.find("@ATTRIBUTE") != string::npos) {
+            stringstream ss(line);
+            ss >> keyword >> attribute;
+            type = "";
+            while (ss >> type_w)
+                type += type_w + " ";
+            attributes.emplace_back(trim(attribute), trim(type));
+            continue;
+        }
+        if (line[0] == '@') {
+            continue;
+        }
+        lines.push_back(line);
+    }
+    file.close();
+    if (attributes.empty())
+        throw invalid_argument("No attributes found");
+    if (classLast) {
+        className = get<0>(attributes.back());
+        classType = get<1>(attributes.back());
+        attributes.pop_back();
+    } else {
+        className = get<0>(attributes.front());
+        classType = get<1>(attributes.front());
+        attributes.erase(attributes.begin());
+    }
+    generateDataset(classLast);
+
+}
+
+void ArffFiles::generateDataset(bool classLast)
+{
+    X = vector<vector<float>>(attributes.size(), vector<float>(lines.size()));
+    auto yy = vector<string>(lines.size(), "");
+    int labelIndex = classLast ? static_cast<int>(attributes.size()) : 0;
+    for (size_t i = 0; i < lines.size(); i++) {
+        stringstream ss(lines[i]);
+        string value;
+        int pos = 0;
+        int xIndex = 0;
+        while (getline(ss, value, ',')) {
+            if (pos++ == labelIndex) {
+                yy[i] = value;
+            } else {
+                X[xIndex++][i] = stof(value);
+            }
+        }
+    }
+    y = factorize(yy);
+}
+
+string ArffFiles::trim(const string& source)
+{
+    string s(source);
+    s.erase(0, s.find_first_not_of(" '\n\r\t"));
+    s.erase(s.find_last_not_of(" '\n\r\t") + 1);
+    return s;
+}
+
+vector<int> ArffFiles::factorize(const vector<string>& labels_t)
+{
+    vector<int> yy;
+    yy.reserve(labels_t.size());
+    map<string, int> labelMap;
+    int i = 0;
+    for (const string& label : labels_t) {
+        if (labelMap.find(label) == labelMap.end()) {
+            labelMap[label] = i++;
+        }
+        yy.push_back(labelMap[label]);
+    }
+    return yy;
+}

src/fimdlp/ArffFiles.h

@@ -0,0 +1,34 @@
+#ifndef ARFFFILES_H
+#define ARFFFILES_H
+
+#include <string>
+#include <vector>
+
+using namespace std;
+
+class ArffFiles {
+private:
+    vector<string> lines;
+    vector<pair<string, string>> attributes;
+    string className;
+    string classType;
+    vector<vector<float>> X;
+    vector<int> y;
+
+    void generateDataset(bool);
+
+public:
+    ArffFiles();
+    void load(const string&, bool = true);
+    vector<string> getLines() const;
+    unsigned long int getSize() const;
+    string getClassName() const;
+    string getClassType() const;
+    static string trim(const string&);
+    vector<vector<float>>& getX();
+    vector<int>& getY();
+    vector<pair<string, string>> getAttributes() const;
+    static vector<int> factorize(const vector<string>& labels_t);
+};
+
+#endif

src/fimdlp/Factorize.cpp

@@ -0,0 +1,18 @@
+#include "Factorize.h"
+
+namespace utils {
+    vector<int> cppFactorize(const vector<string>& labels_t)
+    {
+        vector<int> yy;
+        yy.reserve(labels_t.size());
+        map<string, int> labelMap;
+        int i = 0;
+        for (const string& label : labels_t) {
+            if (labelMap.find(label) == labelMap.end()) {
+                labelMap[label] = i++;
+            }
+            yy.push_back(labelMap[label]);
+        }
+        return yy;
+    }
+}

src/fimdlp/Factorize.h

@@ -0,0 +1,10 @@
+#ifndef FACTORIZE_H
+#define FACTORIZE_H
+#include <vector>
+#include <map>
+#include <string>
+namespace utils {
+    using namespace std;
+    vector<int> cppFactorize(const vector<string>&);
+}
+#endif

src/fimdlp/__init__.py

@@ -1,8 +1,4 @@
 from ._version import __version__
 
 
-def version():
-    return __version__
-
-
 all = ["FImdlp", "__version__"]

src/fimdlp/_version.py

@@ -1 +1 @@
-__version__ = "0.9.1"
+__version__ = "0.9.4"

src/fimdlp/cfimdlp.pyx

@@ -1,20 +1,27 @@
 # distutils: language = c++
 # cython: language_level = 3
 from libcpp.vector cimport vector
+from libcpp.pair cimport pair
+from libcpp.string cimport string
 from libcpp cimport bool
+import numpy as np
 
+cdef extern from "limits.h":
+    cdef int INT_MAX
 cdef extern from "../cppmdlp/CPPFImdlp.h" namespace "mdlp":
     ctypedef float precision_t
     cdef cppclass CPPFImdlp:
-        CPPFImdlp(bool) except + 
+        CPPFImdlp() except + 
+        CPPFImdlp(size_t, int, float) except + 
         CPPFImdlp& fit(vector[precision_t]&, vector[int]&)
+        int get_depth()
         vector[precision_t] getCutPoints()
+        string version()
         
-
 cdef class CFImdlp:
     cdef CPPFImdlp *thisptr
-    def __cinit__(self, proposal):
+    def __cinit__(self, size_t min_length=3, int max_depth=INT_MAX, float max_cuts=0):
-        self.thisptr = new CPPFImdlp(proposal)
+        self.thisptr = new CPPFImdlp(min_length, max_depth, max_cuts)
     def __dealloc__(self):
         del self.thisptr
     def fit(self, X, y):
@@ -22,4 +29,53 @@ cdef class CFImdlp:
         return self
     def get_cut_points(self):
         return self.thisptr.getCutPoints()
- 
+    def get_version(self):
+        return self.thisptr.version()
+    def get_depth(self):
+        return self.thisptr.get_depth()
+    def __reduce__(self):
+        return (CFImdlp, ())
+
+cdef extern from "Factorize.h" namespace "utils":
+    vector[int] cppFactorize(vector[string] &input_vector)
+def factorize(input_vector):
+    return cppFactorize(input_vector)
+
+cdef extern from "ArffFiles.h":
+    cdef cppclass ArffFiles:
+        ArffFiles() except +
+        void load(string, bool)
+        unsigned long int getSize()
+        string getClassName()
+        string getClassType()
+        string trim(const string&)
+        vector[vector[float]]& getX()
+        vector[int]& getY()
+        vector[string] getLines()
+        vector[pair[string, string]] getAttributes()
+
+cdef class CArffFiles:
+    cdef ArffFiles *thisptr
+    def __cinit__(self):
+        self.thisptr = new ArffFiles()
+    def __dealloc__(self):
+        del self.thisptr
+    def load(self, string filename, bool verbose = True):
+        self.thisptr.load(filename, verbose)
+    def get_size(self):
+        return self.thisptr.getSize()
+    def get_class_name(self):
+        return self.thisptr.getClassName()
+    def get_class_type(self):
+        return self.thisptr.getClassType()
+    def get_X(self):
+        return np.array(self.thisptr.getX()).T
+    def get_y(self):
+        return self.thisptr.getY()
+    def get_lines(self):
+        return self.thisptr.getLines()
+    def get_attributes(self):
+        return self.thisptr.getAttributes()
+    def __reduce__(self):
+        return (CArffFiles, ())
+   

src/fimdlp/mdlp.py

@@ -1,15 +1,18 @@
 import numpy as np
-from .cppfimdlp import CFImdlp
+from .cppfimdlp import CFImdlp, factorize
 from sklearn.base import BaseEstimator, TransformerMixin
 from sklearn.utils.multiclass import unique_labels
 from sklearn.utils.validation import check_X_y, check_array, check_is_fitted
 from joblib import Parallel, delayed
+from ._version import __version__
 
 
 class FImdlp(TransformerMixin, BaseEstimator):
-    def __init__(self, n_jobs=-1, proposal=False):
+    def __init__(self, n_jobs=-1, min_length=3, max_depth=1e6, max_cuts=0):
         self.n_jobs = n_jobs
-        self.proposal = proposal
+        self.min_length = min_length
+        self.max_depth = max_depth
+        self.max_cuts = max_cuts
 
     """Fayyad - Irani MDLP discretization algorithm based implementation.
 
@@ -19,30 +22,35 @@ class FImdlp(TransformerMixin, BaseEstimator):
         The number of jobs to run in parallel. :meth:`fit` and
         :meth:`transform`, are parallelized over the features. ``-1`` means
         using all cores available.
+    min_length: int, default=3
+        The minimum length of an interval to be considered to be discretized.
+    max_depth: int, default=1e6
+        The maximum depth of the discretization process.
+    max_cuts: float, default=0
+        The maximum number of cut points to be computed for each feature.
 
     Attributes
     ----------
-    n_features_ : int
+    n_features_in_ : int
         The number of features of the data passed to :meth:`fit`.
     discretizer_ : list
         The list of discretizers, one for each feature.
     cut_points_ : list
         The list of cut points for each feature.
-    X_ : array
+    X_ : array, shape (n_samples, n_features)
-        the samples used to fit, shape (n_samples, n_features)
+        the samples used to fit
-    y_ : array
+    y_ : array, shape(n_samples,)
-        the labels used to fit, shape (n_samples,)
+        the labels used to fit
     features_ : list
         the list of features to be discretized
     """
 
-    def _check_params_fit(self, X, y, expected_args, kwargs):
+    def _more_tags(self):
-        """Check the common parameters passed to fit"""
+        return {"preserves_dtype": [np.int32], "requires_y": True}
+
+    def _check_args(self, X, y, expected_args, kwargs):
         # Check that X and y have correct shape
         X, y = check_X_y(X, y)
-        # Store the classes seen during fit
-        self.classes_ = unique_labels(y)
-        self.n_classes_ = self.classes_.shape[0]
         # Default values
         self.features_ = [i for i in range(X.shape[1])]
         for key, value in kwargs.items():
@@ -63,15 +71,24 @@ class FImdlp(TransformerMixin, BaseEstimator):
             raise ValueError("Feature index out of range")
         return X, y
 
+    def _update_params(self, X, y):
+        # Store the classes seen during fit
+        self.classes_ = unique_labels(y)
+        self.n_classes_ = self.classes_.shape[0]
+        self.n_features_in_ = X.shape[1]
+
+    @staticmethod
+    def get_version():
+        return f"{__version__}({CFImdlp().get_version().decode()})"
+
     def fit(self, X, y, **kwargs):
         """A reference implementation of a fitting function for a transformer.
         Parameters
         ----------
-        X : {array-like, sparse matrix}, shape (n_samples, n_features)
+        X : array, shape (n_samples, n_features)
             The training input samples.
-        y : None
+        y : array, shape (n_samples,)
-            There is no need of a target in a transformer, yet the pipeline API
+            the labels used to fit
-            requires this parameter.
         features : list, default=[i for i in range(n_features)]
             The list of features to be discretized.
         Returns
@@ -79,24 +96,41 @@ class FImdlp(TransformerMixin, BaseEstimator):
         self : object
             Returns self.
         """
-        X, y = self._check_params_fit(
+        X, y = self._check_args(
             X, y, expected_args=["features"], kwargs=kwargs
         )
-        self.n_features_ = X.shape[1]
+        self._update_params(X, y)
         self.X_ = X
         self.y_ = y
-        self.discretizer_ = [None] * self.n_features_
+        self.efective_min_length_ = (
-        self.cut_points_ = [None] * self.n_features_
+            self.min_length
+            if self.min_length > 1
+            else int(self.min_length * X.shape[0])
+        )
+        self.discretizer_ = [None] * self.n_features_in_
+        self.cut_points_ = [None] * self.n_features_in_
         Parallel(n_jobs=self.n_jobs, prefer="threads")(
             delayed(self._fit_discretizer)(feature)
-            for feature in range(self.n_features_)
+            for feature in range(self.n_features_in_)
         )
+        # target of every feature. Start with -1 => y (see join_fit)
+        self.target_ = [-1] * self.n_features_in_
         return self
 
     def _fit_discretizer(self, feature):
-        self.discretizer_[feature] = CFImdlp(proposal=self.proposal)
+        if feature in self.features_:
-        self.discretizer_[feature].fit(self.X_[:, feature], self.y_)
+            self.discretizer_[feature] = CFImdlp(
-        self.cut_points_[feature] = self.discretizer_[feature].get_cut_points()
+                min_length=self.efective_min_length_,
+                max_depth=self.max_depth,
+                max_cuts=self.max_cuts,
+            )
+            self.discretizer_[feature].fit(self.X_[:, feature], self.y_)
+            self.cut_points_[feature] = self.discretizer_[
+                feature
+            ].get_cut_points()
+        else:
+            self.discretizer_[feature] = None
+            self.cut_points_[feature] = []
 
     def _discretize_feature(self, feature, X, result):
         if feature in self.features_:
@@ -108,7 +142,7 @@ class FImdlp(TransformerMixin, BaseEstimator):
         """Discretize X values.
         Parameters
         ----------
-        X : {array-like}, shape (n_samples, n_features)
+        X : array, shape (n_samples, n_features)
             The input samples.
         Returns
         -------
@@ -116,22 +150,41 @@ class FImdlp(TransformerMixin, BaseEstimator):
             The array containing the discretized values of ``X``.
         """
         # Check is fit had been called
-        check_is_fitted(self, "n_features_")
+        check_is_fitted(self, "n_features_in_")
         # Input validation
         X = check_array(X)
         # Check that the input is of the same shape as the one passed
         # during fit.
-        if X.shape[1] != self.n_features_:
+        if X.shape[1] != self.n_features_in_:
             raise ValueError(
                 "Shape of input is different from what was seen in `fit`"
             )
-        result = np.zeros_like(X, dtype=np.int32) - 1
+        if len(self.features_) == self.n_features_in_:
+            result = np.zeros_like(X, dtype=np.int32) - 1
+        else:
+            result = np.zeros_like(X) - 1
         Parallel(n_jobs=self.n_jobs, prefer="threads")(
             delayed(self._discretize_feature)(feature, X[:, feature], result)
-            for feature in range(self.n_features_)
+            for feature in range(self.n_features_in_)
         )
         return result
 
+    @staticmethod
+    def factorize(yy):
+        """Factorize the input labels
+
+        Parameters
+        ----------
+        yy : array, shape (n_samples,)
+            Labels to be factorized, MUST be bytes, i.e. b"0", b"1", ...
+
+        Returns
+        -------
+        array, shape (n_samples,)
+            Factorized labels
+        """
+        return factorize(yy)
+
     def get_cut_points(self):
         """Get the cut points for each feature.
         Returns
@@ -140,6 +193,78 @@ class FImdlp(TransformerMixin, BaseEstimator):
             The list of cut points for each feature.
         """
         result = []
-        for feature in range(self.n_features_):
+        for feature in range(self.n_features_in_):
             result.append(self.cut_points_[feature])
         return result
+
+    def get_states_feature(self, feature):
+        """Return the states a feature can take
+
+        Parameters
+        ----------
+        feature : int
+            feature to get the states
+
+        Returns
+        -------
+        list
+            states of the feature
+        """
+        if feature in self.features_:
+            return list(range(len(self.cut_points_[feature]) + 1))
+        return None
+
+    def join_fit(self, features, target, data):
+        """Join the selected features with the labels and fit the discretizer
+        of the target variable
+        join - fit - transform
+
+        Parameters
+        ----------
+        features : [list]
+            index of the features to join with the labels
+        target : [int]
+            index of the target variable to discretize
+        data: [array] shape (n_samples, n_features)
+            dataset that contains the features to join
+
+        Returns
+        -------
+        result: np.array
+            The target variable newly discretized
+        """
+        check_is_fitted(self, "n_features_in_")
+        if len(features) < 1 or len(features) > self.n_features_in_:
+            raise ValueError(
+                "Number of features must be in range [1, "
+                f"{self.n_features_in_}]"
+            )
+        for feature in features:
+            if feature < 0 or feature >= self.n_features_in_:
+                raise ValueError(
+                    f"Feature {feature} not in range [0, "
+                    f"{self.n_features_in_})"
+                )
+        if target < 0 or target >= self.n_features_in_:
+            raise ValueError(
+                f"Target {target} not in range [0, {self.n_features_in_})"
+            )
+        if target in features:
+            raise ValueError("Target cannot be in features to join")
+        y_join = [
+            f"{str(item_y)}{''.join([str(x) for x in items_x])}".encode()
+            for item_y, items_x in zip(self.y_, data[:, features])
+        ]
+        # Store in target_ the features used with class to discretize target
+        self.target_[target] = features + [-1]
+        self.y_join_ = y_join
+        self.discretizer_[target].fit(self.X_[:, target], factorize(y_join))
+        self.cut_points_[target] = self.discretizer_[target].get_cut_points()
+        # return the discretized target variable with the new cut points
+        return np.searchsorted(self.cut_points_[target], self.X_[:, target])
+
+    def get_depths(self):
+        res = [0] * self.n_features_in_
+        for feature in self.features_:
+            res[feature] = self.discretizer_[feature].get_depth()
+        return res

src/fimdlp/tests/FImdlp_test.py

@@ -1,73 +1,57 @@
 import unittest
 import sklearn
-from sklearn.datasets import load_iris
 import numpy as np
+from sklearn.datasets import load_iris
+from sklearn.utils.estimator_checks import check_estimator
+from ..cppfimdlp import CFImdlp, factorize, CArffFiles
 from ..mdlp import FImdlp
-from .. import version
+from .. import __version__
-from .._version import __version__
 
 
 class FImdlpTest(unittest.TestCase):
+    delta = 1e-6  # same tolerance as in C++ code
+
     def test_version(self):
-        self.assertEqual(version(), __version__)
+        clf = FImdlp()
+        self.assertEqual(
+            clf.get_version(),
+            f"{__version__}({CFImdlp().get_version().decode()})",
+        )
+
+    def test_minimum_mdlp_version(self):
+        mdlp_version = tuple(
+            int(c) for c in CFImdlp().get_version().decode().split(".")[0:3]
+        )
+        minimum_mdlp_version = (1, 1, 2)
+        self.assertTrue(mdlp_version >= minimum_mdlp_version)
 
     def test_init(self):
         clf = FImdlp()
         self.assertEqual(-1, clf.n_jobs)
-        self.assertFalse(clf.proposal)
+        self.assertEqual(3, clf.min_length)
-        clf = FImdlp(proposal=True, n_jobs=7)
+        self.assertEqual(1e6, clf.max_depth)
-        self.assertTrue(clf.proposal)
+        clf = FImdlp(n_jobs=7, min_length=24, max_depth=17)
         self.assertEqual(7, clf.n_jobs)
+        self.assertEqual(24, clf.min_length)
+        self.assertEqual(17, clf.max_depth)
 
-    def test_fit_proposal(self):
+    def test_fit_definitive(self):
-        clf = FImdlp(proposal=True)
+        clf = FImdlp()
-        clf.fit([[1, 2], [3, 4]], [1, 2])
-        self.assertEqual(clf.n_features_, 2)
-        self.assertListEqual(clf.X_.tolist(), [[1, 2], [3, 4]])
-        self.assertListEqual(clf.y_.tolist(), [1, 2])
-        self.assertListEqual([[], []], clf.get_cut_points())
         X, y = load_iris(return_X_y=True)
         clf.fit(X, y)
-        self.assertEqual(clf.n_features_, 4)
+        self.assertEqual(clf.n_features_in_, 4)
         self.assertTrue(np.array_equal(X, clf.X_))
         self.assertTrue(np.array_equal(y, clf.y_))
         expected = [
-            [
+            [5.45, 5.75],
-                4.900000095367432,
+            [2.75, 2.85, 2.95, 3.05, 3.35],
-                5.0,
+            [2.45, 4.75, 5.05],
-                5.099999904632568,
+            [0.8, 1.75],
-                5.400000095367432,
-                5.699999809265137,
-            ],
-            [2.6999998092651367, 2.9000000953674316, 3.1999998092651367],
-            [2.3499999046325684, 4.5, 4.800000190734863],
-            [0.75, 1.399999976158142, 1.5, 1.7000000476837158],
         ]
-        self.assertListEqual(expected, clf.get_cut_points())
+        computed = clf.get_cut_points()
-        self.assertListEqual([0, 1, 2, 3], clf.features_)
+        for item_computed, item_expected in zip(computed, expected):
-        clf.fit(X, y, features=[0, 2, 3])
+            for x_, y_ in zip(item_computed, item_expected):
-        self.assertListEqual([0, 2, 3], clf.features_)
+                self.assertAlmostEqual(x_, y_, delta=self.delta)
-
-    def test_fit_original(self):
-        clf = FImdlp(proposal=False)
-        clf.fit([[1, 2], [3, 4]], [1, 2])
-        self.assertEqual(clf.n_features_, 2)
-        self.assertListEqual(clf.X_.tolist(), [[1, 2], [3, 4]])
-        self.assertListEqual(clf.y_.tolist(), [1, 2])
-        self.assertListEqual([[], []], clf.get_cut_points())
-        X, y = load_iris(return_X_y=True)
-        clf.fit(X, y)
-        self.assertEqual(clf.n_features_, 4)
-        self.assertTrue(np.array_equal(X, clf.X_))
-        self.assertTrue(np.array_equal(y, clf.y_))
-
-        expected = [
-            [5.5, 5.800000190734863],
-            [2.9000000953674316, 3.3499999046325684],
-            [2.450000047683716, 4.800000190734863],
-            [0.800000011920929, 1.7999999523162842],
-        ]
-        self.assertListEqual(expected, clf.get_cut_points())
         self.assertListEqual([0, 1, 2, 3], clf.features_)
         clf.fit(X, y, features=[0, 2, 3])
         self.assertListEqual([0, 2, 3], clf.features_)
@@ -88,67 +72,296 @@ class FImdlpTest(unittest.TestCase):
             clf.fit([[1, 2], [3, 4]], [1, 2], features=[0, 2])
 
     def test_fit_features(self):
+        clf = FImdlp(n_jobs=-1)
+        # Two samples doesn't have enough information to split
+        clf.fit([[1, -2], [3, 4]], [1, 2], features=[0])
+        self.assertListEqual(clf.get_cut_points(), [[], []])
+        clf.fit([[1, -2], [3, 4], [5, 6]], [1, 2, 2], features=[0])
+        self.assertListEqual(clf.get_cut_points(), [[2], []])
+        res = clf.transform([[1, -2], [3, 4]])
+        self.assertListEqual(res.tolist(), [[0, -2], [1, 4]])
+        X, y = load_iris(return_X_y=True)
+        X_expected = X[:, [0, 2]].copy()
+        clf.fit(X, y, features=[1, 3])
+        X_computed = clf.transform(X)
+        self.assertListEqual(
+            X_expected[:, 0].tolist(), X_computed[:, 0].tolist()
+        )
+        self.assertListEqual(
+            X_expected[:, 1].tolist(), X_computed[:, 2].tolist()
+        )
+        self.assertEqual(X_computed.dtype, np.float64)
+
+    def test_transform(self):
         clf = FImdlp()
-        clf.fit([[1, 2], [3, 4]], [1, 2], features=[0])
+        clf.fit([[1, 2], [3, 4], [5, 6]], [1, 2, 2])
-        res = clf.transform([[1, 2], [3, 4]])
-        self.assertListEqual(res.tolist(), [[0, 2], [0, 4]])
-
-    def test_transform_original(self):
-        clf = FImdlp(proposal=False)
-        clf.fit([[1, 2], [3, 4]], [1, 2])
         self.assertEqual(
-            clf.transform([[1, 2], [3, 4]]).tolist(), [[0, 0], [0, 0]]
+            clf.transform([[1, 2], [3, 4]]).tolist(), [[0, 0], [1, 1]]
         )
         X, y = load_iris(return_X_y=True)
         clf.fit(X, y)
-        self.assertEqual(clf.n_features_, 4)
+        self.assertEqual(clf.n_features_in_, 4)
         self.assertTrue(np.array_equal(X, clf.X_))
         self.assertTrue(np.array_equal(y, clf.y_))
+        X_transformed = clf.transform(X)
         self.assertListEqual(
-            clf.transform(X).tolist(), clf.fit(X, y).transform(X).tolist()
+            X_transformed.tolist(), clf.fit(X, y).transform(X).tolist()
         )
+        self.assertEqual(X_transformed.dtype, np.int32)
         expected = [
-            [0, 0, 1, 1],
-            [2, 1, 1, 1],
             [1, 0, 1, 1],
+            [2, 3, 1, 1],
+            [2, 0, 1, 1],
             [0, 0, 1, 1],
             [1, 0, 1, 1],
-            [1, 1, 1, 1],
+            [1, 3, 1, 1],
-            [1, 0, 1, 1],
+            [1, 2, 1, 1],
         ]
         self.assertTrue(np.array_equal(clf.transform(X[90:97]), expected))
         with self.assertRaises(ValueError):
             clf.transform([[1, 2, 3], [4, 5, 6]])
         with self.assertRaises(sklearn.exceptions.NotFittedError):
-            clf = FImdlp(proposal=False)
+            clf = FImdlp()
             clf.transform([[1, 2], [3, 4]])
 
-    def test_transform_proposal(self):
+    def test_cppfactorize(self):
-        clf = FImdlp(proposal=True)
+        source = [
-        clf.fit([[1, 2], [3, 4]], [1, 2])
+            b"f0",
-        self.assertEqual(
+            b"f1",
-            clf.transform([[1, 2], [3, 4]]).tolist(), [[0, 0], [0, 0]]
+            b"f2",
+            b"f3",
+            b"f4",
+            b"f5",
+            b"f6",
+            b"f1",
+            b"f1",
+            b"f7",
+            b"f8",
+        ]
+        expected = [0, 1, 2, 3, 4, 5, 6, 1, 1, 7, 8]
+        computed = factorize(source)
+        self.assertListEqual(expected, computed)
+
+    def test_join_fit(self):
+        y = np.array([b"f0", b"f0", b"f2", b"f3", b"f3", b"f4", b"f4"])
+        x = np.array(
+            [
+                [0, 1, 2, 3, 4, 5],
+                [0, 2, 2, 3, 4, 5],
+                [1, 2, 3, 4, 5, 5],
+                [2, 3, 4, 5, 6, 6],
+                [3, 4, 5, 6, 7, 7],
+                [1, 2, 2, 3, 5, 7],
+                [1, 3, 4, 4, 4, 7],
+            ]
         )
+        expected = [0, 1, 1, 2, 2, 1, 2]
+        clf = FImdlp()
+        clf.fit(x, factorize(y))
+        computed = clf.join_fit([0, 2, 3, 4], 1, x)
+        self.assertListEqual(computed.tolist(), expected)
+        expected_y = [
+            b"00234",
+            b"00234",
+            b"11345",
+            b"22456",
+            b"23567",
+            b"31235",
+            b"31444",
+        ]
+        self.assertListEqual(expected_y, clf.y_join_)
+
+    def test_join_fit_error(self):
+        y = np.array([b"f0", b"f0", b"f2", b"f3", b"f4"])
+        x = np.array(
+            [
+                [0, 1, 2, 3, 4],
+                [0, 1, 2, 3, 4],
+                [1, 2, 3, 4, 5],
+                [2, 3, 4, 5, 6],
+                [3, 4, 5, 6, 7],
+            ]
+        )
+        clf = FImdlp()
+        clf.fit(x, factorize(y))
+        with self.assertRaises(ValueError) as exception:
+            clf.join_fit([], 1, x)
+        self.assertEqual(
+            str(exception.exception),
+            "Number of features must be in range [1, 5]",
+        )
+        with self.assertRaises(ValueError) as exception:
+            FImdlp().join_fit([0, 4], 1, x)
+        self.assertTrue(
+            str(exception.exception).startswith(
+                "This FImdlp instance is not fitted yet."
+            )
+        )
+        with self.assertRaises(ValueError) as exception:
+            clf.join_fit([0, 5], 1, x)
+        self.assertEqual(
+            str(exception.exception),
+            "Feature 5 not in range [0, 5)",
+        )
+        with self.assertRaises(ValueError) as exception:
+            clf.join_fit([0, 2], 5, x)
+        self.assertEqual(
+            str(exception.exception),
+            "Target 5 not in range [0, 5)",
+        )
+        with self.assertRaises(ValueError) as exception:
+            clf.join_fit([0, 2], 2, x)
+        self.assertEqual(
+            str(exception.exception),
+            "Target cannot be in features to join",
+        )
+
+    def test_factorize(self):
+        y = np.array([b"f0", b"f0", b"f2", b"f3", b"f4"])
+        clf = FImdlp()
+        computed = clf.factorize(y)
+        self.assertListEqual([0, 0, 1, 2, 3], computed)
+        y = [b"f4", b"f0", b"f0", b"f2", b"f3"]
+        clf = FImdlp()
+        computed = clf.factorize(y)
+        self.assertListEqual([0, 1, 1, 2, 3], computed)
+
+    def test_join_fit_info(self):
+        clf = FImdlp()
         X, y = load_iris(return_X_y=True)
         clf.fit(X, y)
-        self.assertEqual(clf.n_features_, 4)
+        clf.join_fit([0, 2], 1, X)
-        self.assertTrue(np.array_equal(X, clf.X_))
+        clf.join_fit([0, 3], 2, X)
-        self.assertTrue(np.array_equal(y, clf.y_))
+        clf.join_fit([1, 2], 3, X)
-        self.assertListEqual(
+        expected = [-1, [0, 2, -1], [0, 3, -1], [1, 2, -1]]
-            clf.transform(X).tolist(), clf.fit(X, y).transform(X).tolist()
+        self.assertListEqual(expected, clf.target_)
-        )
+
-        expected = [
+    @staticmethod
-            [4, 0, 1, 1],
+    def test_sklearn_transformer():
-            [5, 2, 2, 2],
+        for check, test in check_estimator(FImdlp(), generate_only=True):
-            [5, 0, 1, 1],
+            test(check)
-            [1, 0, 1, 1],
+
-            [4, 1, 1, 1],
+    def test_states_feature(self):
-            [5, 2, 1, 1],
+        clf = FImdlp()
-            [5, 1, 1, 1],
+        X, y = load_iris(return_X_y=True)
+        clf.fit(X, y)
+        expected = []
+        for i in [3, 6, 4, 3]:
+            expected.append(list(range(i)))
+        for feature in range(X.shape[1]):
+            self.assertListEqual(
+                expected[feature], clf.get_states_feature(feature)
+            )
+
+    def test_states_no_feature(self):
+        clf = FImdlp()
+        X, y = load_iris(return_X_y=True)
+        clf.fit(X, y)
+        self.assertIsNone(clf.get_states_feature(4))
+
+    def test_MaxDepth(self):
+        clf = FImdlp(max_depth=1)
+        X, y = load_iris(return_X_y=True)
+        clf.fit(X, y)
+        expected_cutpoints = [
+            [5.45],
+            [3.35],
+            [2.45],
+            [0.8],
         ]
-        self.assertTrue(np.array_equal(clf.transform(X[90:97]), expected))
+        expected_depths = [1] * 4
-        with self.assertRaises(ValueError):
+        self.assertListEqual(expected_depths, clf.get_depths())
-            clf.transform([[1, 2, 3], [4, 5, 6]])
+        for expected, computed in zip(
-        with self.assertRaises(sklearn.exceptions.NotFittedError):
+            expected_cutpoints, clf.get_cut_points()
-            clf = FImdlp(proposal=True)
+        ):
-            clf.transform([[1, 2], [3, 4]])
+            for e, c in zip(expected, computed):
+                self.assertAlmostEqual(e, c, delta=self.delta)
+
+    def test_MinLength(self):
+        clf = FImdlp(min_length=75)
+        X, y = load_iris(return_X_y=True)
+        clf.fit(X, y)
+        expected_cutpoints = [
+            [5.45, 5.75],
+            [2.85, 3.35],
+            [2.45, 4.75],
+            [0.8, 1.75],
+        ]
+        expected_depths = [3, 2, 2, 2]
+        self.assertListEqual(expected_depths, clf.get_depths())
+        for expected, computed in zip(
+            expected_cutpoints, clf.get_cut_points()
+        ):
+            for e, c in zip(expected, computed):
+                self.assertAlmostEqual(e, c, delta=self.delta)
+
+    def test_MinLengthMaxDepth(self):
+        clf = FImdlp(min_length=75, max_depth=2)
+        X, y = load_iris(return_X_y=True)
+        clf.fit(X, y)
+        expected_cutpoints = [
+            [5.45, 5.75],
+            [2.85, 3.35],
+            [2.45, 4.75],
+            [0.8, 1.75],
+        ]
+        expected_depths = [2, 2, 2, 2]
+        self.assertListEqual(expected_depths, clf.get_depths())
+        for expected, computed in zip(
+            expected_cutpoints, clf.get_cut_points()
+        ):
+            for e, c in zip(expected, computed):
+                self.assertAlmostEqual(e, c, delta=self.delta)
+
+    def test_max_cuts(self):
+        clf = FImdlp(max_cuts=1)
+        X, y = load_iris(return_X_y=True)
+        clf.fit(X, y)
+        expected_cutpoints = [
+            [5.45],
+            [2.85],
+            [2.45],
+            [0.8],
+        ]
+        expected_depths = [3, 5, 4, 3]
+        self.assertListEqual(expected_depths, clf.get_depths())
+        for expected, computed in zip(
+            expected_cutpoints, clf.get_cut_points()
+        ):
+            for e, c in zip(expected, computed):
+                self.assertAlmostEqual(e, c, delta=self.delta)
+
+    def test_ArffFiles(self):
+        loader = CArffFiles()
+        loader.load(b"src/cppmdlp/tests/datasets/iris.arff")
+        X = loader.get_X()
+        y = loader.get_y()
+        expected = [
+            (b"sepallength", b"REAL"),
+            (b"sepalwidth", b"REAL"),
+            (b"petallength", b"REAL"),
+            (b"petalwidth", b"REAL"),
+        ]
+        self.assertListEqual(loader.get_attributes(), expected)
+        self.assertListEqual(y[:10], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0])
+        expected = [
+            b"5.1,3.5,1.4,0.2,Iris-setosa",
+            b"4.9,3.0,1.4,0.2,Iris-setosa",
+            b"4.7,3.2,1.3,0.2,Iris-setosa",
+            b"4.6,3.1,1.5,0.2,Iris-setosa",
+            b"5.0,3.6,1.4,0.2,Iris-setosa",
+            b"5.4,3.9,1.7,0.4,Iris-setosa",
+            b"4.6,3.4,1.4,0.3,Iris-setosa",
+            b"5.0,3.4,1.5,0.2,Iris-setosa",
+            b"4.4,2.9,1.4,0.2,Iris-setosa",
+            b"4.9,3.1,1.5,0.1,Iris-setosa",
+        ]
+        self.assertListEqual(loader.get_lines()[:10], expected)
+        expected_X = [
+            [5.0999999, 3.5, 1.39999998, 0.2],
+            [4.9000001, 3, 1.39999998, 0.2],
+            [4.69999981, 3.20000005, 1.29999995, 0.2],
+        ]
+        for computed, expected in zip(X[:3].tolist(), expected_X):
+            for c, e in zip(computed, expected):
+                self.assertAlmostEqual(c, e, delta=self.delta)