FeatureSpace.hpp 17.2 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>
Thomas Purcell's avatar
Thomas Purcell committed
17
#include <feature_creation/node/value_storage/nodes_value_containers.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
18
#include <utils/project.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
19

Thomas Purcell's avatar
Thomas Purcell committed
20
#include <boost/serialization/shared_ptr.hpp>
21
#include <boost/filesystem.hpp>
Thomas Purcell's avatar
Thomas Purcell committed
22

Thomas Purcell's avatar
Thomas Purcell committed
23
#include <iostream>
Thomas Purcell's avatar
Thomas Purcell committed
24
#include <iomanip>
25
#include <utility>
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
    std::vector<node_ptr> _phi; //!< all features
    std::vector<node_ptr> _phi_0; //!< initial feature space

44
45
46
47
48
    std::vector<std::pair<un_param_op_node_gen, std::vector<std::string>>> _un_param_operators; //!< list of all parameterized unary operators with free parameters
    std::vector<std::pair<bin_param_op_node_gen, std::vector<std::string>>> _com_bin_param_operators; //!< list of all parameterized commutable binary operators with free parameters
    std::vector<std::pair<bin_param_op_node_gen, std::vector<std::string>>> _bin_param_operators; //!< list of all parameterized binary operators with free parameters

    std::map<std::string, 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)
49
50
51
52
53
    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

54
    std::vector<double> _prop; //!< The property to fit
55
56
    std::vector<double> _scores; //!< projection scores for each feature

57
58
59
    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
60
    std::string _feature_space_hr_file; //!< File to store information about the selected features
61

62
63
    std::function<void(double*, double*, std::vector<node_ptr>&, std::vector<int>&, int)> _project; //!< Function used to calculate the scores for SIS
    std::shared_ptr<MPI_Interface> _mpi_comm; //!< MPI communicator
64
65
66
67

    double _l_bound; //!< lower bound for absolute value of the features
    double _u_bound; //!< upper bound for absolute value of the features

68
69
    int _max_phi; //!< Maximum rung for the feature creation
    int _n_sis_select; //!< Number of features to select for each dimensions
70
    int _n_samp; //!< Number of samples (training data)
71
    int _n_feat; //!< Total number of features
Thomas Purcell's avatar
Thomas Purcell committed
72
    int _n_rung_store; //!< Total rungs stored
73
    int _n_rung_generate; //!< Total number of rungs to generate on the fly
Thomas Purcell's avatar
Thomas Purcell committed
74
public:
Thomas Purcell's avatar
Thomas Purcell committed
75

76
77
    /**
     * @brief Constructor for the feature space
78
     * @details constructs the feature space from an initial set of features and a list of allowed operators
79
80
     *
     * @param mpi_comm MPI communicator for the calculations
81
     * @param phi_0 The initial set of features to combine
82
     * @param allowed_ops list of allowed operators
83
     * @param allowed_param_ops dictionary of the parameterizable operators and their associated free parameters
84
     * @param prop The property to be learned (training data)
85
86
     * @param max_phi highest rung value for the calculation
     * @param n_sis_select number of features to select during each SIS step
87
88
89
     * @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)
     * @param min_abs_feat_val minimum absolute feature value
90
91
     * @param max_abs_feat_val maximum absolute feature value
     */
Thomas Purcell's avatar
Thomas Purcell committed
92
    FeatureSpace(
Thomas Purcell's avatar
Thomas Purcell committed
93
        std::shared_ptr<MPI_Interface> mpi_comm,
Thomas Purcell's avatar
Thomas Purcell committed
94
95
        std::vector<node_ptr> phi_0,
        std::vector<std::string> allowed_ops,
96
        std::map<std::string, std::vector<std::string>> allowed_param_ops,
97
        std::vector<double> prop,
Thomas Purcell's avatar
Thomas Purcell committed
98
        std::vector<int> task_sizes,
Thomas Purcell's avatar
Thomas Purcell committed
99
100
        int max_phi=1,
        int n_sis_select=1,
101
102
103
104
105
106
        int max_store_rung=-1,
        int n_rung_generate=0,
        double min_abs_feat_val=1e-50,
        double max_abs_feat_val=1e50
    );

107
108
109
110
111
    /**
     * @brief Initialize the feature set given a property vector
     *
     * @param prop The property trying to be learned
     */
112
    void initialize_fs();
113

114
115
116
117
    /**
     * @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
     */
118
    void generate_feature_space();
Thomas Purcell's avatar
Thomas Purcell committed
119

120
    /**
121
     * @brief The selected feature space
122
     */
123
    inline std::vector<node_ptr> phi_selected(){return _phi_selected;};
124
125

    /**
126
     * @brief The full feature space
127
     */
Thomas Purcell's avatar
Thomas Purcell committed
128
    inline std::vector<node_ptr> phi(){return _phi;};
129
130

    /**
131
     * @brief The initial feature space
132
     */
Thomas Purcell's avatar
Thomas Purcell committed
133
    inline std::vector<node_ptr> phi0(){return _phi_0;};
134
135

    /**
136
     * @brief The vector of projection scores for SIS
137
     */
138
139
    inline std::vector<double> scores(){return _scores;}

140
    /**
141
     * @brief The MPI Communicator
142
     */
Thomas Purcell's avatar
Thomas Purcell committed
143
    inline std::shared_ptr<MPI_Interface> mpi_comm(){return _mpi_comm;}
144

145
    /**
146
     * @brief The vector storing the number of samples in each task
147
     */
Thomas Purcell's avatar
Thomas Purcell committed
148
    inline std::vector<int> task_sizes(){return _task_sizes;}
149

150
    // DocString: feat_space_feature_space_file
151
    /**
152
     * @brief The feature space filename
153
     */
154
    inline std::string feature_space_file(){return _feature_space_file;}
155

156
    // DocString: feat_space_l_bound
157
    /**
158
     * @brief The minimum absolute value of the feature
159
     */
160
    inline double l_bound(){return _l_bound;}
161

162
    // DocString: feat_space_u_bound
163
    /**
164
     * @brief The maximum absolute value of the feature
165
     */
166
    inline double u_bound(){return _u_bound;}
167

168
    // DocString: feat_space_max_phi
169
    /**
170
     * @brief The maximum rung of the feature space
171
     */
172
    inline int max_phi(){return _max_phi;}
173

174
    // DocString: feat_space_n_sis_select
175
    /**
176
     * @brief The number of features selected in each SIS step
177
     */
178
    inline int n_sis_select(){return _n_sis_select;}
179

180
    // DocString: feat_space_n_samp
181
    /**
182
     * @brief The number of samples per feature
183
     */
184
    inline int n_samp(){return _n_samp;}
185

186
    // DocString: feat_space_n_feat
187
    /**
188
     * @brief The number of features in the feature space
189
     */
190
    inline int n_feat(){return _n_feat;}
191

192
    // DocString: feat_space_n_rung_store
193
    /**
194
     * @brief The number of rungs whose feature training data is stored in memory
195
     */
196
    inline int n_rung_store(){return _n_rung_store;}
197

198
    // DocString: feat_space_n_rung_generate
199
    /**
200
     * @brief The number of rungs to be generated on the fly during SIS
201
     */
202
    inline int n_rung_generate(){return _n_rung_generate;}
203

204
205
206
207
208
209
210
211
212
213
    /**
     * @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
     */
214
215
    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);

216
217
218
219
220
221
222
223
224
225
    /**
     * @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
     */
226
    void project_generated(double* prop, int size, std::vector<node_ptr>& phi_selected, std::vector<double>& scores_selected, std::vector<double>& scores_comp);
227

228
229
230
231
232
233
234
235
236
237
238
    /**
     * @brief Check if a feature overlaps with a feature previously selected in earlier SIS iterations
     * @details Compares the projection score of the current candidate feature with all those of previously selected features (using the current prop) and
     *          if they are within 1e-10, then check the correlation between the features themselves
     *
     * @param val_ptr pointer to the candidate feature's data
     * @param cur_score the projection score of the candidate feature
     * @param scores_past The projection scores of the previous features
     * @param scores_comp vector to temporarily store the comparison of projection scores
     * @return True if the feature does not overlap with any previously selected
     */
239
    bool valid_score_against_past(double* val_ptr, double cur_score, std::vector<double> scores_past, std::vector<double>& scores_comp);
240

241
242
243
244
245
246
247
248
249
250
251
252
    /**
     * @brief Check if a feature overlaps with a feature previously selected in this SIS iterations
     * @details CCompares the projection score of the current candidate feature with all those of previously selected features in this iteration and
     *          if they are within 1e-10, then check the correlation between the features themselves
     *
     * @param end_check the end point to stop the comparison (the same as the current number of selected features)
     * @param val_ptr pointer to the candidate feature's data
     * @param cur_score the projection score of the candidate feature
     * @param scores_selected The projection scores of the previous features
     * @param scores_comp vector to temporarily store the comparison of projection scores
     * @return True if the feature does not overlap with any previously selected
     */
253
    bool valid_score_against_current(int end_check, double* val_ptr, double cur_score, std::vector<double>& scores_selected, std::vector<double>& scores_comp);
254
255
    /**
     * @brief Perform SIS on a feature set with a specified property
256
     * @details Perform sure-independence screening with either the correct property or the error
257
     *
258
     * @param prop The property to perform SIS over
259
     */
Thomas Purcell's avatar
Thomas Purcell committed
260
    void sis(std::vector<double>& prop);
261

262
    // DocString: feat_space_feat_in_phi
263
264
265
    /**
     * @brief Is a feature in this process' _phi?
     *
266
267
     * @param ind The index of the feature
     * @return True if feature is in this rank's _phi
268
269
270
     */
    inline bool feat_in_phi(int ind){return (ind >= _phi[0]->feat_ind()) && (ind <= _phi.back()->feat_ind());}

271
272
273
    // Python Interface Functions
    #ifdef PY_BINDINGS
        /**
274
275
         * @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>)
276
277
         *
         * @param mpi_comm MPI communicator for the calculations
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 max_phi highest rung value for the calculation
         * @param n_sis_select number of features to select during each SIS step
284
285
286
         * @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)
         * @param min_abs_feat_val minimum absolute feature value
287
288
289
290
291
         * @param max_abs_feat_val maximum absolute feature value
         */
        FeatureSpace(
            py::list phi_0,
            py::list allowed_ops,
292
            py::dict allowed_param_ops,
293
294
295
296
297
298
299
300
301
302
303
            py::list prop,
            py::list task_sizes,
            int max_phi=1,
            int n_sis_select=1,
            int max_store_rung=-1,
            int n_rung_generate=0,
            double min_abs_feat_val=1e-50,
            double max_abs_feat_val=1e50
        );

        /**
304
305
         * @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>)
306
307
         *
         * @param mpi_comm MPI communicator for the calculations
308
         * @param phi_0 The initial set of features to combine
309
         * @param allowed_ops list of allowed operators
310
         * @param allowed_param_ops dictionary of the parameterizable operators and their associated free parameters
311
         * @param prop The property to be learned (training data)
312
313
         * @param max_phi highest rung value for the calculation
         * @param n_sis_select number of features to select during each SIS step
314
315
316
         * @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)
         * @param min_abs_feat_val minimum absolute feature value
317
318
319
320
321
         * @param max_abs_feat_val maximum absolute feature value
         */
        FeatureSpace(
            py::list phi_0,
            py::list allowed_ops,
322
            py::dict allowed_param_ops,
323
324
325
326
327
328
329
330
331
332
            np::ndarray prop,
            py::list task_sizes,
            int max_phi=1,
            int n_sis_select=1,
            int max_store_rung=-1,
            int n_rung_generate=0,
            double min_abs_feat_val=1e-50,
            double max_abs_feat_val=1e50
        );

333
        // DocString: feat_space_sis_arr
334
335
336
        /**
         * @brief Wrapper function for SIS using a numpy array
         *
337
         * @param prop(np.ndarray) The property to perform SIS over as a numpy array
338
         */
339
340
341
342
343
        inline void sis(np::ndarray prop)
        {
            std::vector<double> prop_vec = python_conv_utils::from_ndarray<double>(prop);
            sis(prop_vec);
        }
344
345

        // DocString: feat_space_sis_list
346
347
348
        /**
         * @brief Wrapper function for SIS using a python list
         *
349
         * @param prop(list) The property to perform SIS over as a python list
350
         */
351
352
353
354
355
356
        inline void sis(py::list prop)
        {
            std::vector<double> prop_vec = python_conv_utils::from_list<double>(prop);
            sis(prop_vec);
        }

357
        // DocString: feat_space_phi_selected_py
358
        /**
359
         * @brief The selected feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
360
361
         * @return _phi_selected as a python list
         */
362
        py::list phi_selected_py();
363

364
        // DocString: feat_space_phi0_py
365
        /**
366
         * @brief The initial feature space (cpp definition in <python/feature_creation/FeatureSpace.cpp>)
367
368
         * @return _phi0 as a python list
         */
369
        py::list phi0_py();
370

371
        // DocString: feat_space_scores_py
372
        /**
373
         * @brief The vector of projection scores for SIS
374
375
         * @return _scores as a numpy array
         */
376
        inline np::ndarray scores_py(){return python_conv_utils::to_ndarray<double>(_scores);};
377

378
        // DocString: feat_space_task_sizes_py
379
        /**
380
         * @brief The vector storing the number of samples in each task
381
382
         * @return _task_sizes as a python list
         */
383
        inline py::list task_sizes_py(){return python_conv_utils::to_list<int>(_task_sizes);};
384

385
        // DocString: feat_space_allowed_ops_py
386
        /**
387
         * @brief The list of allowed operator nodes
388
389
         * @return _allowed_ops as a python list
         */
390
        inline py::list allowed_ops_py(){return python_conv_utils::to_list<std::string>(_allowed_ops);}
391

392
        // DocString: feat_space_start_gen_py
393
        /**
394
         * @brief The index in _phi where each generation starts
395
396
         * @return _start_gen as a python list
         */
397
        inline py::list start_gen_py(){return python_conv_utils::to_list<int>(_start_gen);}
398

399
    #endif
Thomas Purcell's avatar
Thomas Purcell committed
400
401
402
};

#endif