FeatureSpace.hpp 19.8 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>
14
15
#include <mpi_interface/MPI_ops.hpp>
#include <mpi_interface/serialize_tuple.h>
Thomas Purcell's avatar
Thomas Purcell committed
16
#include <feature_creation/node/ModelNode.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
17
#include <feature_creation/node/operator_nodes/allowed_ops.hpp>
18
#include <feature_creation/node/utils.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
19
#include <feature_creation/node/value_storage/nodes_value_containers.hpp>
20
#include <utils/compare_features.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
21
#include <utils/project.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
22

Thomas Purcell's avatar
Thomas Purcell committed
23
#include <boost/serialization/shared_ptr.hpp>
24
#include <boost/filesystem.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
25

Thomas Purcell's avatar
Thomas Purcell committed
26
#include <iostream>
Thomas Purcell's avatar
Thomas Purcell committed
27
#include <iomanip>
28
#include <utility>
Thomas Purcell's avatar
Thomas Purcell committed
29

30
31
32
33
#ifdef PY_BINDINGS
    namespace np = boost::python::numpy;
    namespace py = boost::python;
#endif
34

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

47
    #ifdef PARAMETERIZE
Thomas Purcell's avatar
Thomas Purcell committed
48
49
50
        std::vector<un_param_op_node_gen> _un_param_operators; //!< list of all parameterized unary operators with free parameters
        std::vector<bin_param_op_node_gen> _com_bin_param_operators; //!< list of all parameterized commutable binary operators with free parameters
        std::vector<bin_param_op_node_gen> _bin_param_operators; //!< list of all parameterized binary operators with free parameters
51
    #endif
52

Thomas Purcell's avatar
Thomas Purcell committed
53
    std::vector<std::string> _allowed_param_ops; //!< Map of parameterization operator set (set of operators and non-linear parameters used for a non-linear least squares fit to property)
54
55
56
57
58
    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

59
    std::vector<double> _prop; //!< The property to fit
60
61
    std::vector<double> _scores; //!< projection scores for each feature

62
    const std::vector<int> _task_sizes; //!< The number of elements in each task (training data)
63
    std::vector<int> _start_gen; //!< list of the indexes where each generation starts in _phi
64
65
    const std::string _feature_space_file; //!< File to store information about the selected features
    const std::string _feature_space_summary_file; //!< File to store information about the selected features
66

67
68
    std::function<void(double*, double*, std::vector<node_ptr>&, const std::vector<int>&, int)> _project; //!< Function used to calculate the scores for SIS
    std::function<void(double*, double*, std::vector<node_ptr>&, const std::vector<int>&, int)> _project_no_omp; //!< Function used to calculate the scores for SIS without changing omp environment
69
    std::function<bool(double*, int, double, std::vector<double>&, double, int, int)> _is_valid; //!< Function used to calculate the scores for SIS
Thomas Purcell's avatar
Bug fix    
Thomas Purcell committed
70
    std::function<bool(double*, int, double, std::vector<node_ptr>&, std::vector<double>&, double)> _is_valid_feat_list; //!< Function used to calculate the scores for SIS without changing omp environment
71

72
    std::shared_ptr<MPI_Interface> _mpi_comm; //!< MPI communicator
73

74
75
76
    const double _cross_cor_max; //!< Maximum cross-correlation used for selecting features
    const double _l_bound; //!< lower bound for absolute value of the features
    const double _u_bound; //!< upper bound for absolute value of the features
77

Thomas Purcell's avatar
Thomas Purcell committed
78
    int _n_rung_store; //!< Total rungs stored
79
80
    int _n_feat; //!< Total number of features
    int _max_phi; //!< Maximum rung for the feature creation
81

82
83
84
    const int _n_sis_select; //!< Number of features to select for each dimensions
    const int _n_samp; //!< Number of samples (training data)
    const int _n_rung_generate; //!< Total number of rungs to generate on the fly
85

86
87
    int _max_param_depth; //!< Max depth to parameterize a feature (default=_max_rung)

Thomas Purcell's avatar
Thomas Purcell committed
88
public:
Thomas Purcell's avatar
Thomas Purcell committed
89

90
91
    /**
     * @brief Constructor for the feature space
92
     * @details constructs the feature space from an initial set of features and a list of allowed operators
93
94
     *
     * @param mpi_comm MPI communicator for the calculations
95
     * @param phi_0 The initial set of features to combine
96
     * @param allowed_ops list of allowed operators
97
     * @param allowed_param_ops dictionary of the parameterizable operators and their associated free parameters
98
     * @param prop The property to be learned (training data)
99
100
     * @param task_sizes The number of samples per task
     * @param project_type The projection operator to use
101
102
     * @param max_phi highest rung value for the calculation
     * @param n_sis_select number of features to select during each SIS step
103
104
     * @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
105
     * @param cross_corr_max Maximum cross-correlation used for selecting features
106
     * @param min_abs_feat_val minimum absolute feature value
107
108
     * @param max_abs_feat_val maximum absolute feature value
     */
Thomas Purcell's avatar
Thomas Purcell committed
109
    FeatureSpace(
Thomas Purcell's avatar
Thomas Purcell committed
110
        std::shared_ptr<MPI_Interface> mpi_comm,
Thomas Purcell's avatar
Thomas Purcell committed
111
112
        std::vector<node_ptr> phi_0,
        std::vector<std::string> allowed_ops,
Thomas Purcell's avatar
Thomas Purcell committed
113
        std::vector<std::string> allowed_param_ops,
114
        std::vector<double> prop,
Thomas Purcell's avatar
Thomas Purcell committed
115
        std::vector<int> task_sizes,
116
        std::string project_type="regression",
Thomas Purcell's avatar
Thomas Purcell committed
117
118
        int max_phi=1,
        int n_sis_select=1,
119
120
        int max_store_rung=-1,
        int n_rung_generate=0,
Thomas Purcell's avatar
Thomas Purcell committed
121
        double cross_corr_max=1.0,
122
        double min_abs_feat_val=1e-50,
123
        double max_abs_feat_val=1e50,
Thomas Purcell's avatar
Thomas Purcell committed
124
        int max_param_depth = -1
125
126
    );

127
128
129
130
131
    /**
     * @brief Initialize the feature set given a property vector
     *
     * @param prop The property trying to be learned
     */
Thomas Purcell's avatar
Thomas Purcell committed
132
    void initialize_fs(std::string project_type);
133

134
135
136
137
    /**
     * @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
     */
138
    void generate_feature_space();
Thomas Purcell's avatar
Thomas Purcell committed
139

140
    /**
141
     * @brief The selected feature space
142
     */
143
    inline std::vector<node_ptr> phi_selected(){return _phi_selected;};
144
145

    /**
146
     * @brief The full feature space
147
     */
Thomas Purcell's avatar
Thomas Purcell committed
148
    inline std::vector<node_ptr> phi(){return _phi;};
149
150

    /**
151
     * @brief The initial feature space
152
     */
Thomas Purcell's avatar
Thomas Purcell committed
153
    inline std::vector<node_ptr> phi0(){return _phi_0;};
154
155

    /**
156
     * @brief The vector of projection scores for SIS
157
     */
158
159
    inline std::vector<double> scores(){return _scores;}

160
    /**
161
     * @brief The MPI Communicator
162
     */
Thomas Purcell's avatar
Thomas Purcell committed
163
    inline std::shared_ptr<MPI_Interface> mpi_comm(){return _mpi_comm;}
164

165
    /**
166
     * @brief The vector storing the number of samples in each task
167
     */
Thomas Purcell's avatar
Thomas Purcell committed
168
    inline std::vector<int> task_sizes(){return _task_sizes;}
169

170
    // DocString: feat_space_feature_space_file
171
    /**
172
     * @brief The feature space filename
173
     */
174
    inline std::string feature_space_file(){return _feature_space_file;}
175

176
    // DocString: feat_space_l_bound
177
    /**
178
     * @brief The minimum absolute value of the feature
179
     */
180
    inline double l_bound(){return _l_bound;}
181

182
    // DocString: feat_space_u_bound
183
    /**
184
     * @brief The maximum absolute value of the feature
185
     */
186
    inline double u_bound(){return _u_bound;}
187

188
    // DocString: feat_space_max_phi
189
    /**
190
     * @brief The maximum rung of the feature space
191
     */
192
    inline int max_phi(){return _max_phi;}
193

194
    // DocString: feat_space_n_sis_select
195
    /**
196
     * @brief The number of features selected in each SIS step
197
     */
198
    inline int n_sis_select(){return _n_sis_select;}
199

200
    // DocString: feat_space_n_samp
201
    /**
202
     * @brief The number of samples per feature
203
     */
204
    inline int n_samp(){return _n_samp;}
205

206
    // DocString: feat_space_n_feat
207
    /**
208
     * @brief The number of features in the feature space
209
     */
210
    inline int n_feat(){return _n_feat;}
211

212
    // DocString: feat_space_n_rung_store
213
    /**
214
     * @brief The number of rungs whose feature training data is stored in memory
215
     */
216
    inline int n_rung_store(){return _n_rung_store;}
217

218
    // DocString: feat_space_n_rung_generate
219
    /**
220
     * @brief The number of rungs to be generated on the fly during SIS
221
     */
222
    inline int n_rung_generate(){return _n_rung_generate;}
223

224
225
226
227
228
229
230
231
232
233
    /**
     * @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
     */
234
235
    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);

236
237
238
239
240
241
242
243
244
245
    /**
     * @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
     */
246
    void project_generated(double* prop, int size, std::vector<node_ptr>& phi_selected, std::vector<double>& scores_selected);
247

248
249
    /**
     * @brief Perform SIS on a feature set with a specified property
250
     * @details Perform sure-independence screening with either the correct property or the error
251
     *
252
     * @param prop The property to perform SIS over
253
     */
Thomas Purcell's avatar
Thomas Purcell committed
254
    void sis(std::vector<double>& prop);
255

256
    // DocString: feat_space_feat_in_phi
257
258
259
    /**
     * @brief Is a feature in this process' _phi?
     *
260
261
     * @param ind The index of the feature
     * @return True if feature is in this rank's _phi
262
263
264
     */
    inline bool feat_in_phi(int ind){return (ind >= _phi[0]->feat_ind()) && (ind <= _phi.back()->feat_ind());}

265
266
267
268
269
270
271
272
    // DocString: feat_space_remove_feature
    /**
     * @brief Remove a feature from phi
     *
     * @param ind index of feature to remove
     */
    void remove_feature(int ind);

273
274
275
    // Python Interface Functions
    #ifdef PY_BINDINGS
        /**
276
277
         * @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>)
278
         *
279
         * @param phi_0 The initial set of features to combine
280
         * @param allowed_ops list of allowed operators
281
         * @param allowed_param_ops dictionary of the parameterizable operators and their associated free parameters
282
         * @param prop The property to be learned (training data)
283
284
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
285
286
         * @param max_phi highest rung value for the calculation
         * @param n_sis_select number of features to select during each SIS step
287
288
         * @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
289
         * @param cross_corr_max Maximum cross-correlation used for selecting features
290
         * @param min_abs_feat_val minimum absolute feature value
291
292
293
294
295
         * @param max_abs_feat_val maximum absolute feature value
         */
        FeatureSpace(
            py::list phi_0,
            py::list allowed_ops,
Thomas Purcell's avatar
Thomas Purcell committed
296
            py::list allowed_param_ops,
297
298
            py::list prop,
            py::list task_sizes,
Thomas Purcell's avatar
Thomas Purcell committed
299
            std::string project_type="regression",
300
301
302
303
            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
304
            double cross_corr_max=1.0,
305
            double min_abs_feat_val=1e-50,
306
            double max_abs_feat_val=1e50,
Thomas Purcell's avatar
Thomas Purcell committed
307
            int max_param_depth = -1
308
309
310
        );

        /**
311
312
         * @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>)
313
         *
314
         * @param phi_0 The initial set of features to combine
315
         * @param allowed_ops list of allowed operators
316
         * @param allowed_param_ops dictionary of the parameterizable operators and their associated free parameters
317
         * @param prop The property to be learned (training data)
318
319
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
320
321
         * @param max_phi highest rung value for the calculation
         * @param n_sis_select number of features to select during each SIS step
322
323
         * @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
324
         * @param cross_corr_max Maximum cross-correlation used for selecting features
325
         * @param min_abs_feat_val minimum absolute feature value
326
327
328
329
330
         * @param max_abs_feat_val maximum absolute feature value
         */
        FeatureSpace(
            py::list phi_0,
            py::list allowed_ops,
Thomas Purcell's avatar
Thomas Purcell committed
331
            py::list allowed_param_ops,
332
333
            np::ndarray prop,
            py::list task_sizes,
Thomas Purcell's avatar
Thomas Purcell committed
334
            std::string project_type="regression",
335
336
337
338
            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
339
            double cross_corr_max=1.0,
340
            double min_abs_feat_val=1e-50,
341
            double max_abs_feat_val=1e50,
Thomas Purcell's avatar
Thomas Purcell committed
342
            int max_param_depth = -1
343
344
        );

345
346
347
348
349
350
        /**
         * @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
351
         * @param prop The property to be learned (training data)
352
353
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
354
355
356
357
358
359
         * @param n_sis_select number of features to select during each SIS step
         * @param cross_corr_max Maximum cross-correlation used for selecting features
         */
        FeatureSpace(
            std::string feature_file,
            py::list phi_0,
360
361
362
363
364
365
366
367
368
369
370
371
372
373
            np::ndarray prop,
            py::list task_sizes,
            std::string project_type="pearson",
            int n_sis_select=1,
            double cross_corr_max=1.0
        );

        /**
         * @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 prop The property to be learned (training data)
         * @param phi_0 The initial set of features to combine
374
375
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
376
377
378
379
380
381
         * @param n_sis_select number of features to select during each SIS step
         * @param cross_corr_max Maximum cross-correlation used for selecting features
         */
        FeatureSpace(
            std::string feature_file,
            py::list phi_0,
382
            py::list prop,
383
            py::list task_sizes,
384
            std::string project_type="pearson",
385
386
387
388
            int n_sis_select=1,
            double cross_corr_max=1.0
        );

389
        // DocString: feat_space_sis_arr
390
391
392
        /**
         * @brief Wrapper function for SIS using a numpy array
         *
393
         * @param prop(np.ndarray) The property to perform SIS over as a numpy array
394
         */
395
396
397
398
399
        inline void sis(np::ndarray prop)
        {
            std::vector<double> prop_vec = python_conv_utils::from_ndarray<double>(prop);
            sis(prop_vec);
        }
400
401

        // DocString: feat_space_sis_list
402
403
404
        /**
         * @brief Wrapper function for SIS using a python list
         *
405
         * @param prop(list) The property to perform SIS over as a python list
406
         */
407
408
409
410
411
412
        inline void sis(py::list prop)
        {
            std::vector<double> prop_vec = python_conv_utils::from_list<double>(prop);
            sis(prop_vec);
        }

413
        // DocString: feat_space_phi_selected_py
414
        /**
415
         * @brief The selected feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
416
417
         * @return _phi_selected as a python list
         */
418
        py::list phi_selected_py();
419

420
        // DocString: feat_space_phi0_py
421
        /**
422
         * @brief The initial feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
423
424
         * @return _phi0 as a python list
         */
425
        py::list phi0_py();
426

427
428
429
430
431
432
433
        // 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();

434
        // DocString: feat_space_scores_py
435
        /**
436
         * @brief The vector of projection scores for SIS
437
438
         * @return _scores as a numpy array
         */
439
        inline np::ndarray scores_py(){return python_conv_utils::to_ndarray<double>(_scores);};
440

441
        // DocString: feat_space_task_sizes_py
442
        /**
443
         * @brief The vector storing the number of samples in each task
444
445
         * @return _task_sizes as a python list
         */
446
        inline py::list task_sizes_py(){return python_conv_utils::to_list<int>(_task_sizes);};
447

448
        // DocString: feat_space_allowed_ops_py
449
        /**
450
         * @brief The list of allowed operator nodes
451
452
         * @return _allowed_ops as a python list
         */
453
        inline py::list allowed_ops_py(){return python_conv_utils::to_list<std::string>(_allowed_ops);}
454

455
        // DocString: feat_space_start_gen_py
456
        /**
457
         * @brief The index in _phi where each generation starts
458
459
         * @return _start_gen as a python list
         */
460
        inline py::list start_gen_py(){return python_conv_utils::to_list<int>(_start_gen);}
461

462
463
464
465
466
467
468
        // 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
         */
469
        inline ModelNode get_feature(int ind){return ModelNode(_phi[ind]);}
470
    #endif
Thomas Purcell's avatar
Thomas Purcell committed
471
472
};

473
#endif