FeatureSpace.hpp 17.7 KB
Newer Older
1
2
3
4
5
6
7
8
9
/** @file feature_creation/feature_space/FeatureSpace.hpp
 *  @brief Create a feature space from an initial set of features and algebraic operators
 *
 *  Use an initial set of features and combine them to generate more complicated algebraical features. SIS is also performed here
 *
 *  @author Thomas A. R. Purcell (tpurcell)
 *  @bug No known bugs.
 */

Thomas Purcell's avatar
Thomas Purcell committed
10
11
12
#ifndef FEATURE_SPACE
#define FEATURE_SPACE

Thomas Purcell's avatar
Thomas Purcell committed
13
#include <mpi_interface/MPI_Interface.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
14
#include <feature_creation/node/FeatureNode.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
15
#include <feature_creation/node/ModelNode.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
16
#include <feature_creation/node/operator_nodes/allowed_ops.hpp>
17
#include <feature_creation/node/utils.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
18
#include <feature_creation/node/value_storage/nodes_value_containers.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
19
#include <utils/project.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
20

Thomas Purcell's avatar
Thomas Purcell committed
21
#include <boost/serialization/shared_ptr.hpp>
22
#include <boost/filesystem.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
23

Thomas Purcell's avatar
Thomas Purcell committed
24
#include <iostream>
Thomas Purcell's avatar
Thomas Purcell committed
25
#include <iomanip>
Thomas Purcell's avatar
Thomas Purcell committed
26

27
28
29
30
#ifdef PY_BINDINGS
    namespace np = boost::python::numpy;
    namespace py = boost::python;
#endif
31

32
// DocString: cls_feat_space
33
34
35
36
37
/**
 * @brief Feature Space for SISSO calculations
 * @details Stores and performs all feature calculations for SIS
 *
 */
Thomas Purcell's avatar
Thomas Purcell committed
38
39
class FeatureSpace
{
40
    std::vector<node_ptr> _phi_selected; //!< selected features
41
42
43
44
45
46
47
48
49
50
    std::vector<node_ptr> _phi; //!< all features
    std::vector<node_ptr> _phi_0; //!< initial feature space

    std::vector<std::string> _allowed_ops; //!< list of all allowed operators strings
    std::vector<un_op_node_gen> _un_operators; //!< list of all unary operators
    std::vector<bin_op_node_gen> _com_bin_operators; //!< list of all commutable binary operators
    std::vector<bin_op_node_gen> _bin_operators; //!< list of all binary operators

    std::vector<double> _scores; //!< projection scores for each feature

51
52
53
    std::vector<int> _task_sizes; //!< The number of elements in each task (training data)
    std::vector<int> _start_gen; //!< list of the indexes where each generation starts in _phi
    std::string _feature_space_file; //!< File to store information about the selected features
Thomas Purcell's avatar
Thomas Purcell committed
54
    std::string _feature_space_summary_file; //!< File to store information about the selected features
55

56
    std::function<void(double*, double*, std::vector<node_ptr>&, std::vector<int>&, int)> _project; //!< Function used to calculate the scores for SIS
57
    std::function<void(double*, double*, std::vector<node_ptr>&, std::vector<int>&, int)> _project_no_omp; //!< Function used to calculate the scores for SIS without changing omp environment
58
    std::shared_ptr<MPI_Interface> _mpi_comm; //!< MPI communicator
59

Thomas Purcell's avatar
Thomas Purcell committed
60
    double _cross_cor_max; //!< Maximum cross-correlation used for selecting features
61
62
63
    double _l_bound; //!< lower bound for absolute value of the features
    double _u_bound; //!< upper bound for absolute value of the features

64
65
    int _max_phi; //!< Maximum rung for the feature creation
    int _n_sis_select; //!< Number of features to select for each dimensions
66
    int _n_samp; //!< Number of samples (training data)
67
    int _n_feat; //!< Total number of features
Thomas Purcell's avatar
Thomas Purcell committed
68
    int _n_rung_store; //!< Total rungs stored
69
    int _n_rung_generate; //!< Total number of rungs to generate on the fly
Thomas Purcell's avatar
Thomas Purcell committed
70
public:
Thomas Purcell's avatar
Thomas Purcell committed
71

72
73
    /**
     * @brief Constructor for the feature space
74
     * @details constructs the feature space from an initial set of features and a list of allowed operators
75
76
     *
     * @param mpi_comm MPI communicator for the calculations
77
     * @param phi_0 The initial set of features to combine
78
     * @param allowed_ops list of allowed operators
79
     * @param prop The property to be learned (training data)
80
81
     * @param task_sizes The number of samples per task
     * @param project_type The projection operator to use
82
83
     * @param max_phi highest rung value for the calculation
     * @param n_sis_select number of features to select during each SIS step
84
85
     * @param max_store_rung number of rungs to calculate and store the value of the features for all samples
     * @param n_rung_generate number of rungs to generate on the fly during SIS (this must be 1 or 0 right now, possible to be higher with recursive algorithm)
Thomas Purcell's avatar
Thomas Purcell committed
86
     * @param cross_corr_max Maximum cross-correlation used for selecting features
87
     * @param min_abs_feat_val minimum absolute feature value
88
89
     * @param max_abs_feat_val maximum absolute feature value
     */
Thomas Purcell's avatar
Thomas Purcell committed
90
    FeatureSpace(
Thomas Purcell's avatar
Thomas Purcell committed
91
        std::shared_ptr<MPI_Interface> mpi_comm,
Thomas Purcell's avatar
Thomas Purcell committed
92
93
        std::vector<node_ptr> phi_0,
        std::vector<std::string> allowed_ops,
94
        std::vector<double> prop,
Thomas Purcell's avatar
Thomas Purcell committed
95
        std::vector<int> task_sizes,
96
        std::string project_type="pearson",
Thomas Purcell's avatar
Thomas Purcell committed
97
98
        int max_phi=1,
        int n_sis_select=1,
99
100
        int max_store_rung=-1,
        int n_rung_generate=0,
Thomas Purcell's avatar
Thomas Purcell committed
101
        double cross_corr_max=1.0,
102
103
104
105
        double min_abs_feat_val=1e-50,
        double max_abs_feat_val=1e50
    );

106
107
108
109
110
    /**
     * @brief Initialize the feature set given a property vector
     *
     * @param prop The property trying to be learned
     */
111
    void initialize_fs(std::vector<double> prop, std::string project_type);
112

113
114
115
116
    /**
     * @brief Generate the full feature set from the allowed operators and initial feature set
     * @details populates phi with all features from an initial set and the allowed operators
     */
117
    void generate_feature_space(std::vector<double>& prop);
Thomas Purcell's avatar
Thomas Purcell committed
118

119
    /**
120
     * @brief The selected feature space
121
     */
122
    inline std::vector<node_ptr> phi_selected(){return _phi_selected;};
123
124

    /**
125
     * @brief The full feature space
126
     */
Thomas Purcell's avatar
Thomas Purcell committed
127
    inline std::vector<node_ptr> phi(){return _phi;};
128
129

    /**
130
     * @brief The initial feature space
131
     */
Thomas Purcell's avatar
Thomas Purcell committed
132
    inline std::vector<node_ptr> phi0(){return _phi_0;};
133
134

    /**
135
     * @brief The vector of projection scores for SIS
136
     */
137
138
    inline std::vector<double> scores(){return _scores;}

139
    /**
140
     * @brief The MPI Communicator
141
     */
Thomas Purcell's avatar
Thomas Purcell committed
142
    inline std::shared_ptr<MPI_Interface> mpi_comm(){return _mpi_comm;}
143

144
    /**
145
     * @brief The vector storing the number of samples in each task
146
     */
Thomas Purcell's avatar
Thomas Purcell committed
147
    inline std::vector<int> task_sizes(){return _task_sizes;}
148

149
    // DocString: feat_space_feature_space_file
150
    /**
151
     * @brief The feature space filename
152
     */
153
    inline std::string feature_space_file(){return _feature_space_file;}
154

155
    // DocString: feat_space_l_bound
156
    /**
157
     * @brief The minimum absolute value of the feature
158
     */
159
    inline double l_bound(){return _l_bound;}
160

161
    // DocString: feat_space_u_bound
162
    /**
163
     * @brief The maximum absolute value of the feature
164
     */
165
    inline double u_bound(){return _u_bound;}
166

167
    // DocString: feat_space_max_phi
168
    /**
169
     * @brief The maximum rung of the feature space
170
     */
171
    inline int max_phi(){return _max_phi;}
172

173
    // DocString: feat_space_n_sis_select
174
    /**
175
     * @brief The number of features selected in each SIS step
176
     */
177
    inline int n_sis_select(){return _n_sis_select;}
178

179
    // DocString: feat_space_n_samp
180
    /**
181
     * @brief The number of samples per feature
182
     */
183
    inline int n_samp(){return _n_samp;}
184

185
    // DocString: feat_space_n_feat
186
    /**
187
     * @brief The number of features in the feature space
188
     */
189
    inline int n_feat(){return _n_feat;}
190

191
    // DocString: feat_space_n_rung_store
192
    /**
193
     * @brief The number of rungs whose feature training data is stored in memory
194
     */
195
    inline int n_rung_store(){return _n_rung_store;}
196

197
    // DocString: feat_space_n_rung_generate
198
    /**
199
     * @brief The number of rungs to be generated on the fly during SIS
200
     */
201
    inline int n_rung_generate(){return _n_rung_generate;}
202

203
204
205
206
207
208
209
210
211
212
    /**
     * @brief Generate a new set of features from a single feature
     * @details Take in the feature and perform all valid algebraic operations on it.
     *
     * @param feat The feature to spawn new features from
     * @param feat_set The feature set to pull features from for combinations
     * @param feat_ind starting index for the next feature generated
     * @param l_bound lower bound for the absolute value of the feature
     * @param u_bound upper bound for the abosulte value of the feature
     */
213
214
    void generate_new_feats(std::vector<node_ptr>::iterator& feat, std::vector<node_ptr>& feat_set, int& feat_ind, double l_bound=1e-50, double u_bound=1e50);

215
216
217
218
219
220
221
222
223
224
    /**
     * @brief Calculate the SIS Scores for feature generated on the fly
     * @details Create the next rung of features and calculate their projection scores. Only keep those that can be selected by SIS.
     *
     * @param prop Pointer to the start of the vector storing the data to project the features onto
     * @param size The size of the data to project over
     * @param phi_selected The features that would be selected from the previous rungs
     * @param scores_selected The projection scores of the features that would be selected from the previous rungs
     * @param scores_comp vector to store temporary score comparisons
     */
225
    void project_generated(double* prop, int size, std::vector<node_ptr>& phi_selected, std::vector<double>& scores_selected);
226

Thomas Purcell's avatar
Thomas Purcell committed
227
228
229
230
231
232
233
234
    /**
     * @brief Checks the feature to see if it is still valid against previously selected features
     *
     * @param val_ptr pointer to value array of the current feature
     * @param end_sel index of the feature to stop checking
     *
     * @return True if the feature is still valid
     */
235
    bool valid_feature_against_selected(double* val_ptr, std::vector<double>& scores_sel, double cur_score, int end_sel, int start_sel = 0);
Thomas Purcell's avatar
Thomas Purcell committed
236

237
    bool valid_feature_against_private_selected(double* val_ptr, std::vector<node_ptr>& selected, std::vector<double>& scores_sel, double cur_score);
238

239
240
    /**
     * @brief Perform SIS on a feature set with a specified property
241
     * @details Perform sure-independence screening with either the correct property or the error
242
     *
243
     * @param prop The property to perform SIS over
244
     */
Thomas Purcell's avatar
Thomas Purcell committed
245
    void sis(std::vector<double>& prop);
246

247
    // DocString: feat_space_feat_in_phi
248
249
250
    /**
     * @brief Is a feature in this process' _phi?
     *
251
252
     * @param ind The index of the feature
     * @return True if feature is in this rank's _phi
253
254
255
     */
    inline bool feat_in_phi(int ind){return (ind >= _phi[0]->feat_ind()) && (ind <= _phi.back()->feat_ind());}

256
257
258
259
260
261
262
263
    // DocString: feat_space_remove_feature
    /**
     * @brief Remove a feature from phi
     *
     * @param ind index of feature to remove
     */
    void remove_feature(int ind);

264
265
266
    // Python Interface Functions
    #ifdef PY_BINDINGS
        /**
267
268
         * @brief Constructor for the feature space that takes in python objects
         * @details constructs the feature space from an initial set of features and a list of allowed operators (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
269
         *
270
         * @param phi_0 The initial set of features to combine
271
         * @param allowed_ops list of allowed operators
272
         * @param prop The property to be learned (training data)
273
274
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
275
276
         * @param max_phi highest rung value for the calculation
         * @param n_sis_select number of features to select during each SIS step
277
278
         * @param max_store_rung number of rungs to calculate and store the value of the features for all samples
         * @param n_rung_generate number of rungs to generate on the fly during SIS (this must be 1 or 0 right now, possible to be higher with recursive algorithm)
Thomas Purcell's avatar
Thomas Purcell committed
279
         * @param cross_corr_max Maximum cross-correlation used for selecting features
280
         * @param min_abs_feat_val minimum absolute feature value
281
282
283
284
285
286
287
         * @param max_abs_feat_val maximum absolute feature value
         */
        FeatureSpace(
            py::list phi_0,
            py::list allowed_ops,
            py::list prop,
            py::list task_sizes,
288
            std::string project_type="pearson",
289
290
291
292
            int max_phi=1,
            int n_sis_select=1,
            int max_store_rung=-1,
            int n_rung_generate=0,
Thomas Purcell's avatar
Thomas Purcell committed
293
            double cross_corr_max=1.0,
294
295
296
297
298
            double min_abs_feat_val=1e-50,
            double max_abs_feat_val=1e50
        );

        /**
299
300
         * @brief Constructor for the feature space that takes in python and numpy objects
         * @details constructs the feature space from an initial set of features and a list of allowed operators (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
301
         *
302
         * @param phi_0 The initial set of features to combine
303
         * @param allowed_ops list of allowed operators
304
         * @param prop The property to be learned (training data)
305
306
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
307
308
         * @param max_phi highest rung value for the calculation
         * @param n_sis_select number of features to select during each SIS step
309
310
         * @param max_store_rung number of rungs to calculate and store the value of the features for all samples
         * @param n_rung_generate number of rungs to generate on the fly during SIS (this must be 1 or 0 right now, possible to be higher with recursive algorithm)
Thomas Purcell's avatar
Thomas Purcell committed
311
         * @param cross_corr_max Maximum cross-correlation used for selecting features
312
         * @param min_abs_feat_val minimum absolute feature value
313
314
315
316
317
318
319
         * @param max_abs_feat_val maximum absolute feature value
         */
        FeatureSpace(
            py::list phi_0,
            py::list allowed_ops,
            np::ndarray prop,
            py::list task_sizes,
320
            std::string project_type="pearson",
321
322
323
324
            int max_phi=1,
            int n_sis_select=1,
            int max_store_rung=-1,
            int n_rung_generate=0,
Thomas Purcell's avatar
Thomas Purcell committed
325
            double cross_corr_max=1.0,
326
327
328
329
            double min_abs_feat_val=1e-50,
            double max_abs_feat_val=1e50
        );

330
331
332
333
334
335
        /**
         * @brief Constructor for the feature space that takes in python and numpy objects
         * @details constructs the feature space from an initial set of features and a file containing postfix expressions for the features (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
         *
         * @param feature_file The file with the postfix expressions for the feature space
         * @param phi_0 The initial set of features to combine
336
337
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
338
339
340
341
342
343
344
345
         * @param n_sis_select number of features to select during each SIS step
         * @param max_store_rung number of rungs to calculate and store the value of the features for all samples
         * @param cross_corr_max Maximum cross-correlation used for selecting features
         */
        FeatureSpace(
            std::string feature_file,
            py::list phi_0,
            py::list task_sizes,
346
            std::string project_type="pearson",
347
348
349
350
            int n_sis_select=1,
            double cross_corr_max=1.0
        );

351
        // DocString: feat_space_sis_arr
352
353
354
        /**
         * @brief Wrapper function for SIS using a numpy array
         *
355
         * @param prop(np.ndarray) The property to perform SIS over as a numpy array
356
         */
357
358
359
360
361
        inline void sis(np::ndarray prop)
        {
            std::vector<double> prop_vec = python_conv_utils::from_ndarray<double>(prop);
            sis(prop_vec);
        }
362
363

        // DocString: feat_space_sis_list
364
365
366
        /**
         * @brief Wrapper function for SIS using a python list
         *
367
         * @param prop(list) The property to perform SIS over as a python list
368
         */
369
370
371
372
373
374
        inline void sis(py::list prop)
        {
            std::vector<double> prop_vec = python_conv_utils::from_list<double>(prop);
            sis(prop_vec);
        }

375
        // DocString: feat_space_phi_selected_py
376
        /**
377
         * @brief The selected feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
378
379
         * @return _phi_selected as a python list
         */
380
        py::list phi_selected_py();
381

382
        // DocString: feat_space_phi0_py
383
        /**
384
         * @brief The initial feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
385
386
         * @return _phi0 as a python list
         */
387
        py::list phi0_py();
388

389
390
391
392
393
394
395
        // DocString: feat_space_phi_py
        /**
         * @brief The feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
         * @return _phi as a python list
         */
        py::list phi_py();

396
        // DocString: feat_space_scores_py
397
        /**
398
         * @brief The vector of projection scores for SIS
399
400
         * @return _scores as a numpy array
         */
401
        inline np::ndarray scores_py(){return python_conv_utils::to_ndarray<double>(_scores);};
402

403
        // DocString: feat_space_task_sizes_py
404
        /**
405
         * @brief The vector storing the number of samples in each task
406
407
         * @return _task_sizes as a python list
         */
408
        inline py::list task_sizes_py(){return python_conv_utils::to_list<int>(_task_sizes);};
409

410
        // DocString: feat_space_allowed_ops_py
411
        /**
412
         * @brief The list of allowed operator nodes
413
414
         * @return _allowed_ops as a python list
         */
415
        inline py::list allowed_ops_py(){return python_conv_utils::to_list<std::string>(_allowed_ops);}
416

417
        // DocString: feat_space_start_gen_py
418
        /**
419
         * @brief The index in _phi where each generation starts
420
421
         * @return _start_gen as a python list
         */
422
        inline py::list start_gen_py(){return python_conv_utils::to_list<int>(_start_gen);}
423

424
425
426
427
428
429
430
431
        // DocString: feat_space_get_feature
        /**
         * @brief Return a feature at a specified index
         *
         * @param ind index of the feature to get
         * @return A ModelNode of the feature at index ind
         */
        inline ModelNode get_feature(int ind){return ModelNode(_phi[ind]->d_mat_ind(), _phi[ind]->rung(), _phi[ind]->expr(), _phi[ind]->postfix_expr(), _phi[ind]->value(), _phi[ind]->test_value(), _phi[ind]->unit());}
432
    #endif
Thomas Purcell's avatar
Thomas Purcell committed
433
434
435
};

#endif