FeatureSpace.hpp 18.8 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

    bool _param_internal; //!< True if parameterize all scale and shift parameters in a feature
Thomas Purcell's avatar
Thomas Purcell committed
87
public:
Thomas Purcell's avatar
Thomas Purcell committed
88

89
90
    /**
     * @brief Constructor for the feature space
91
     * @details constructs the feature space from an initial set of features and a list of allowed operators
92
93
     *
     * @param mpi_comm MPI communicator for the calculations
94
     * @param phi_0 The initial set of features to combine
95
     * @param allowed_ops list of allowed operators
96
     * @param allowed_param_ops dictionary of the parameterizable operators and their associated free parameters
97
     * @param prop The property to be learned (training data)
98
99
     * @param task_sizes The number of samples per task
     * @param project_type The projection operator to use
100
101
     * @param max_phi highest rung value for the calculation
     * @param n_sis_select number of features to select during each SIS step
102
103
     * @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
104
     * @param cross_corr_max Maximum cross-correlation used for selecting features
105
     * @param min_abs_feat_val minimum absolute feature value
106
107
     * @param max_abs_feat_val maximum absolute feature value
     */
Thomas Purcell's avatar
Thomas Purcell committed
108
    FeatureSpace(
Thomas Purcell's avatar
Thomas Purcell committed
109
        std::shared_ptr<MPI_Interface> mpi_comm,
Thomas Purcell's avatar
Thomas Purcell committed
110
111
        std::vector<node_ptr> phi_0,
        std::vector<std::string> allowed_ops,
Thomas Purcell's avatar
Thomas Purcell committed
112
        std::vector<std::string> allowed_param_ops,
113
        std::vector<double> prop,
Thomas Purcell's avatar
Thomas Purcell committed
114
        std::vector<int> task_sizes,
115
        std::string project_type="regression",
Thomas Purcell's avatar
Thomas Purcell committed
116
117
        int max_phi=1,
        int n_sis_select=1,
118
119
        int max_store_rung=-1,
        int n_rung_generate=0,
Thomas Purcell's avatar
Thomas Purcell committed
120
        double cross_corr_max=1.0,
121
        double min_abs_feat_val=1e-50,
122
123
        double max_abs_feat_val=1e50,
        bool param_internal=true
124
125
    );

126
127
128
129
130
    /**
     * @brief Initialize the feature set given a property vector
     *
     * @param prop The property trying to be learned
     */
Thomas Purcell's avatar
Thomas Purcell committed
131
    void initialize_fs(std::string project_type);
132

133
134
135
136
    /**
     * @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
     */
137
    void generate_feature_space();
Thomas Purcell's avatar
Thomas Purcell committed
138

139
    /**
140
     * @brief The selected feature space
141
     */
142
    inline std::vector<node_ptr> phi_selected(){return _phi_selected;};
143
144

    /**
145
     * @brief The full feature space
146
     */
Thomas Purcell's avatar
Thomas Purcell committed
147
    inline std::vector<node_ptr> phi(){return _phi;};
148
149

    /**
150
     * @brief The initial feature space
151
     */
Thomas Purcell's avatar
Thomas Purcell committed
152
    inline std::vector<node_ptr> phi0(){return _phi_0;};
153
154

    /**
155
     * @brief The vector of projection scores for SIS
156
     */
157
158
    inline std::vector<double> scores(){return _scores;}

159
    /**
160
     * @brief The MPI Communicator
161
     */
Thomas Purcell's avatar
Thomas Purcell committed
162
    inline std::shared_ptr<MPI_Interface> mpi_comm(){return _mpi_comm;}
163

164
    /**
165
     * @brief The vector storing the number of samples in each task
166
     */
Thomas Purcell's avatar
Thomas Purcell committed
167
    inline std::vector<int> task_sizes(){return _task_sizes;}
168

169
    // DocString: feat_space_feature_space_file
170
    /**
171
     * @brief The feature space filename
172
     */
173
    inline std::string feature_space_file(){return _feature_space_file;}
174

175
    // DocString: feat_space_l_bound
176
    /**
177
     * @brief The minimum absolute value of the feature
178
     */
179
    inline double l_bound(){return _l_bound;}
180

181
    // DocString: feat_space_u_bound
182
    /**
183
     * @brief The maximum absolute value of the feature
184
     */
185
    inline double u_bound(){return _u_bound;}
186

187
    // DocString: feat_space_max_phi
188
    /**
189
     * @brief The maximum rung of the feature space
190
     */
191
    inline int max_phi(){return _max_phi;}
192

193
    // DocString: feat_space_n_sis_select
194
    /**
195
     * @brief The number of features selected in each SIS step
196
     */
197
    inline int n_sis_select(){return _n_sis_select;}
198

199
    // DocString: feat_space_n_samp
200
    /**
201
     * @brief The number of samples per feature
202
     */
203
    inline int n_samp(){return _n_samp;}
204

205
    // DocString: feat_space_n_feat
206
    /**
207
     * @brief The number of features in the feature space
208
     */
209
    inline int n_feat(){return _n_feat;}
210

211
    // DocString: feat_space_n_rung_store
212
    /**
213
     * @brief The number of rungs whose feature training data is stored in memory
214
     */
215
    inline int n_rung_store(){return _n_rung_store;}
216

217
    // DocString: feat_space_n_rung_generate
218
    /**
219
     * @brief The number of rungs to be generated on the fly during SIS
220
     */
221
    inline int n_rung_generate(){return _n_rung_generate;}
222

223
224
225
226
227
228
229
230
231
232
    /**
     * @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
     */
233
234
    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);

235
236
237
238
239
240
241
242
243
244
    /**
     * @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
     */
245
    void project_generated(double* prop, int size, std::vector<node_ptr>& phi_selected, std::vector<double>& scores_selected);
246

247
248
    /**
     * @brief Perform SIS on a feature set with a specified property
249
     * @details Perform sure-independence screening with either the correct property or the error
250
     *
251
     * @param prop The property to perform SIS over
252
     */
Thomas Purcell's avatar
Thomas Purcell committed
253
    void sis(std::vector<double>& prop);
254

255
    // DocString: feat_space_feat_in_phi
256
257
258
    /**
     * @brief Is a feature in this process' _phi?
     *
259
260
     * @param ind The index of the feature
     * @return True if feature is in this rank's _phi
261
262
263
     */
    inline bool feat_in_phi(int ind){return (ind >= _phi[0]->feat_ind()) && (ind <= _phi.back()->feat_ind());}

264
265
266
267
268
269
270
271
    // DocString: feat_space_remove_feature
    /**
     * @brief Remove a feature from phi
     *
     * @param ind index of feature to remove
     */
    void remove_feature(int ind);

272
273
274
    // Python Interface Functions
    #ifdef PY_BINDINGS
        /**
275
276
         * @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>)
277
         *
278
         * @param phi_0 The initial set of features to combine
279
         * @param allowed_ops list of allowed operators
280
         * @param allowed_param_ops dictionary of the parameterizable operators and their associated free parameters
281
         * @param prop The property to be learned (training data)
282
283
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
284
285
         * @param max_phi highest rung value for the calculation
         * @param n_sis_select number of features to select during each SIS step
286
287
         * @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
288
         * @param cross_corr_max Maximum cross-correlation used for selecting features
289
         * @param min_abs_feat_val minimum absolute feature value
290
291
292
293
294
         * @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
295
            py::list allowed_param_ops,
296
297
            py::list prop,
            py::list task_sizes,
Thomas Purcell's avatar
Thomas Purcell committed
298
            std::string project_type="regression",
299
300
301
302
            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
303
            double cross_corr_max=1.0,
304
            double min_abs_feat_val=1e-50,
305
306
            double max_abs_feat_val=1e50,
            bool param_internal=true
307
308
309
        );

        /**
310
311
         * @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>)
312
         *
313
         * @param phi_0 The initial set of features to combine
314
         * @param allowed_ops list of allowed operators
315
         * @param allowed_param_ops dictionary of the parameterizable operators and their associated free parameters
316
         * @param prop The property to be learned (training data)
317
318
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
319
320
         * @param max_phi highest rung value for the calculation
         * @param n_sis_select number of features to select during each SIS step
321
322
         * @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
323
         * @param cross_corr_max Maximum cross-correlation used for selecting features
324
         * @param min_abs_feat_val minimum absolute feature value
325
326
327
328
329
         * @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
330
            py::list allowed_param_ops,
331
332
            np::ndarray prop,
            py::list task_sizes,
Thomas Purcell's avatar
Thomas Purcell committed
333
            std::string project_type="regression",
334
335
336
337
            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
338
            double cross_corr_max=1.0,
339
            double min_abs_feat_val=1e-50,
340
341
            double max_abs_feat_val=1e50,
            bool param_internal=true
342
343
        );

344
345
346
347
348
349
        /**
         * @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
350
351
         * @param task_sizes The number of samples per task
         * @param project_type The projection operator to use
352
353
354
355
356
357
358
         * @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,
            py::list task_sizes,
359
            std::string project_type="pearson",
360
361
362
363
            int n_sis_select=1,
            double cross_corr_max=1.0
        );

364
        // DocString: feat_space_sis_arr
365
366
367
        /**
         * @brief Wrapper function for SIS using a numpy array
         *
368
         * @param prop(np.ndarray) The property to perform SIS over as a numpy array
369
         */
370
371
372
373
374
        inline void sis(np::ndarray prop)
        {
            std::vector<double> prop_vec = python_conv_utils::from_ndarray<double>(prop);
            sis(prop_vec);
        }
375
376

        // DocString: feat_space_sis_list
377
378
379
        /**
         * @brief Wrapper function for SIS using a python list
         *
380
         * @param prop(list) The property to perform SIS over as a python list
381
         */
382
383
384
385
386
387
        inline void sis(py::list prop)
        {
            std::vector<double> prop_vec = python_conv_utils::from_list<double>(prop);
            sis(prop_vec);
        }

388
        // DocString: feat_space_phi_selected_py
389
        /**
390
         * @brief The selected feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
391
392
         * @return _phi_selected as a python list
         */
393
        py::list phi_selected_py();
394

395
        // DocString: feat_space_phi0_py
396
        /**
397
         * @brief The initial feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
398
399
         * @return _phi0 as a python list
         */
400
        py::list phi0_py();
401

402
403
404
405
406
407
408
        // 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();

409
        // DocString: feat_space_scores_py
410
        /**
411
         * @brief The vector of projection scores for SIS
412
413
         * @return _scores as a numpy array
         */
414
        inline np::ndarray scores_py(){return python_conv_utils::to_ndarray<double>(_scores);};
415

416
        // DocString: feat_space_task_sizes_py
417
        /**
418
         * @brief The vector storing the number of samples in each task
419
420
         * @return _task_sizes as a python list
         */
421
        inline py::list task_sizes_py(){return python_conv_utils::to_list<int>(_task_sizes);};
422

423
        // DocString: feat_space_allowed_ops_py
424
        /**
425
         * @brief The list of allowed operator nodes
426
427
         * @return _allowed_ops as a python list
         */
428
        inline py::list allowed_ops_py(){return python_conv_utils::to_list<std::string>(_allowed_ops);}
429

430
        // DocString: feat_space_start_gen_py
431
        /**
432
         * @brief The index in _phi where each generation starts
433
434
         * @return _start_gen as a python list
         */
435
        inline py::list start_gen_py(){return python_conv_utils::to_list<int>(_start_gen);}
436

437
438
439
440
441
442
443
        // 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
         */
Thomas Purcell's avatar
Thomas Purcell committed
444
        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]->domain(), _phi[ind]->unit());}
445
    #endif
Thomas Purcell's avatar
Thomas Purcell committed
446
447
};

448
#endif