void enkf_config_node_set_internalize(enkf_config_node_type * node, int report_step) {
ert_impl_type impl_type = enkf_config_node_get_impl_type( node );
if (impl_type == CONTAINER) {
int inode;
int container_size = enkf_config_node_container_size( node );
for (inode = 0; inode < container_size; inode++) {
enkf_config_node_type * child_node = enkf_config_node_container_iget( node , inode );
enkf_config_node_set_internalize( child_node , report_step );
}
} else {
if (node->internalize == NULL)
node->internalize = bool_vector_alloc( 0 , false );
bool_vector_iset( node->internalize , report_step , true);
}
}
static void state_map_select_matching__( state_map_type * map , bool_vector_type * select_target , int select_mask , bool select) {
state_map_assert_writable(map);
pthread_rwlock_rdlock( &map->rw_lock );
{
{
const int * map_ptr = int_vector_get_ptr( map->state );
int size = util_int_min(int_vector_size( map->state ), bool_vector_size(select_target));
for (int i=0; i < size; i++) {
int state_value = map_ptr[i];
if (state_value & select_mask)
bool_vector_iset( select_target , i , select);
}
}
pthread_rwlock_unlock( &map->rw_lock );
}
}
/**
This function will load an active map from the enkf_fs filesystem.
*/
void gen_data_config_load_active( gen_data_config_type * config , enkf_fs_type * fs, int report_step , bool force_load) {
bool fs_changed = false;
if (fs != config->read_fs) {
config->read_fs = fs;
fs_changed = true;
}
if (!config->dynamic)
return; /* This is used as a GEN_PARAM instance - and the loading of mask is not an option. */
pthread_mutex_lock( &config->update_lock );
{
if ( force_load || (int_vector_iget( config->data_size_vector , report_step ) > 0)) {
if (config->active_report_step != report_step || fs_changed) {
char * filename = util_alloc_sprintf("%s_active" , config->key );
FILE * stream = enkf_fs_open_excase_tstep_file( fs , filename , report_step);
if (stream != NULL) {
bool_vector_fread( config->active_mask , stream );
fclose( stream );
} else {
int gen_data_size = int_vector_safe_iget( config->data_size_vector, report_step );
if (gen_data_size < 0) {
fprintf(stderr,"** Fatal internal error in function:%s \n",__func__);
fprintf(stderr,"\n");
fprintf(stderr," 1: The active mask file:%s was not found \n",filename);
fprintf(stderr," 2: The size of the gen_data vectors has not been set\n");
fprintf(stderr,"\n");
fprintf(stderr,"We can not create a suitable active_mask. Code should call gen_data_config_has_active_mask()\n\n");
util_abort("%s: fatal internal error - could not create a suitable active_mask \n",__func__);
} else {
fprintf(stdout,"** Info: could not locate active data elements file %s, filling active vector with true all elements active \n",filename);
bool_vector_reset( config->active_mask );
bool_vector_iset( config->active_mask, gen_data_size - 1, true);
}
}
free( filename );
}
}
config->active_report_step = report_step;
}
pthread_mutex_unlock( &config->update_lock );
}
void forward_initialize_node(enkf_main_type * enkf_main, const char * init_file, enkf_node_type * field_node) {
{
const int ens_size = enkf_main_get_ensemble_size( enkf_main );
bool_vector_type * iactive = bool_vector_alloc(0, false);
bool_vector_iset( iactive , ens_size - 1 , true );
enkf_main_create_run_path(enkf_main , iactive , 0);
bool_vector_free(iactive);
}
{
int iens = 0;
enkf_state_type * state = enkf_main_iget_state( enkf_main , iens );
enkf_fs_type * fs = enkf_main_get_fs(enkf_main);
run_arg_type * run_arg = run_arg_alloc_ENSEMBLE_EXPERIMENT( fs , 0 ,0 , "simulations/run0");
enkf_state_forward_init( state , run_arg);
}
}
void test_state() {
rng_type * rng = rng_alloc( MZRAN , INIT_DEFAULT );
int ens_size = 10;
int active_size = 8;
int rows = 100;
matrix_type * state = matrix_alloc(1,1);
bool_vector_type * ens_mask = bool_vector_alloc(ens_size , false);
matrix_type * A = matrix_alloc( rows , active_size);
matrix_type * A2 = matrix_alloc( rows, active_size );
matrix_type * A3 = matrix_alloc( 1,1 );
for (int i=0; i < active_size; i++)
bool_vector_iset( ens_mask , i + 1 , true );
matrix_random_init(A , rng);
rml_enkf_common_store_state( state , A , ens_mask );
test_assert_int_equal( matrix_get_rows( state ) , rows );
test_assert_int_equal( matrix_get_columns( state ) , ens_size );
{
int g;
int a = 0;
for (g=0; g < ens_size; g++) {
if (bool_vector_iget( ens_mask , g )) {
test_assert_true( matrix_columns_equal( state , g , A , a ));
a++;
}
}
}
rml_enkf_common_recover_state( state , A2 , ens_mask);
rml_enkf_common_recover_state( state , A3 , ens_mask);
test_assert_true( matrix_equal( A , A2 ));
test_assert_true( matrix_equal( A , A3 ));
bool_vector_free( ens_mask );
matrix_free( state );
matrix_free( A );
}
void obs_vector_ensemble_chi2(const obs_vector_type * obs_vector ,
enkf_fs_type * fs,
bool_vector_type * valid ,
int step1 ,
int step2 ,
int iens1 ,
int iens2 ,
state_enum load_state ,
double ** chi2) {
int step;
enkf_node_type * enkf_node = enkf_node_alloc( obs_vector->config_node );
node_id_type node_id;
node_id.state = load_state;
for (step = step1; step <= step2; step++) {
int iens;
node_id.report_step = step;
{
void * obs_node = vector_iget( obs_vector->nodes , step);
if (obs_node == NULL) {
for (iens = iens1; iens < iens2; iens++)
chi2[step][iens] = 0;
} else {
for (iens = iens1; iens < iens2; iens++) {
node_id.iens = iens;
if (enkf_node_try_load( enkf_node , fs , node_id))
chi2[step][iens] = obs_vector_chi2__(obs_vector , step , enkf_node , node_id);
else {
chi2[step][iens] = 0;
// Missing data - this member will be marked as invalid in the misfit calculations.
bool_vector_iset( valid , iens , false );
}
}
}
}
}
enkf_node_free( enkf_node );
}
void model_config_set_load_state( model_config_type * config , int report_step) {
bool_vector_iset(config->__load_state , report_step , true);
}
void model_config_init(model_config_type * model_config ,
const config_type * config ,
int ens_size ,
const ext_joblist_type * joblist ,
int last_history_restart ,
const sched_file_type * sched_file ,
const ecl_sum_type * refcase) {
model_config->forward_model = forward_model_alloc( joblist );
model_config_set_refcase( model_config , refcase );
if (config_item_set( config , FORWARD_MODEL_KEY )) {
char * config_string = config_alloc_joined_string( config , FORWARD_MODEL_KEY , " ");
forward_model_parse_init( model_config->forward_model , config_string );
free(config_string);
}
if (config_item_set( config , ENKF_SCHED_FILE_KEY))
model_config_set_enkf_sched_file(model_config , config_get_value(config , ENKF_SCHED_FILE_KEY ));
if (config_item_set( config, RUNPATH_KEY)) {
model_config_add_runpath( model_config , DEFAULT_RUNPATH_KEY , config_get_value(config , RUNPATH_KEY) );
model_config_select_runpath( model_config , DEFAULT_RUNPATH_KEY );
}
{
history_source_type source_type = DEFAULT_HISTORY_SOURCE;
if (config_item_set( config , HISTORY_SOURCE_KEY)) {
const char * history_source = config_iget(config , HISTORY_SOURCE_KEY, 0,0);
source_type = history_get_source_type( history_source );
}
if (!model_config_select_history( model_config , source_type , sched_file , refcase ))
if (!model_config_select_history( model_config , DEFAULT_HISTORY_SOURCE , sched_file , refcase ))
if (!model_config_select_any_history( model_config , sched_file , refcase))
fprintf(stderr,"** Warning:: Do not have enough information to select a history source \n");
}
if (model_config->history != NULL) {
int num_restart = history_get_last_restart( model_config->history );
bool_vector_iset( model_config->internalize_state , num_restart - 1 , false );
bool_vector_iset( model_config->__load_state , num_restart - 1 , false );
}
/*
The full treatment of the SCHEDULE_PREDICTION_FILE keyword is in
the ensemble_config file, because the functionality is implemented
as (quite) plain GEN_KW instance. Here we just check if it is
present or not.
*/
if (config_item_set(config , SCHEDULE_PREDICTION_FILE_KEY))
model_config->has_prediction = true;
else
model_config->has_prediction = false;
if (config_item_set(config , CASE_TABLE_KEY))
model_config_set_case_table( model_config , ens_size , config_iget( config , CASE_TABLE_KEY , 0,0));
if (config_item_set( config , ENSPATH_KEY))
model_config_set_enspath( model_config , config_get_value(config , ENSPATH_KEY));
if (config_item_set( config , JOBNAME_KEY))
model_config_set_jobname_fmt( model_config , config_get_value(config , JOBNAME_KEY));
if (config_item_set( config , RFTPATH_KEY))
model_config_set_rftpath( model_config , config_get_value(config , RFTPATH_KEY));
if (config_item_set( config , DBASE_TYPE_KEY))
model_config_set_dbase_type( model_config , config_get_value(config , DBASE_TYPE_KEY));
if (config_item_set( config , MAX_RESAMPLE_KEY))
model_config_set_max_internal_submit( model_config , config_get_value_as_int( config , MAX_RESAMPLE_KEY ));
{
const char * export_file_name;
if (config_item_set( config , GEN_KW_EXPORT_FILE_KEY))
export_file_name = config_get_value(config, GEN_KW_EXPORT_FILE_KEY);
else
export_file_name = DEFAULT_GEN_KW_EXPORT_FILE;
model_config_set_gen_kw_export_file(model_config, export_file_name);
}
}
void stepwise_estimate( stepwise_type * stepwise , double deltaR2_limit , int CV_blocks) {
int nvar = matrix_get_columns( stepwise->X0 );
int nsample = matrix_get_rows( stepwise->X0 );
double currentR2 = -1;
bool_vector_type * active_rows = bool_vector_alloc( nsample , true );
/*Reset beta*/
for (int i = 0; i < nvar; i++) {
matrix_iset(stepwise->beta, i , 0 , 0.0);
}
bool_vector_set_all( stepwise->active_set , false );
double MSE_min = 10000000;
double Prev_MSE_min = MSE_min;
double minR2 = -1;
while (true) {
int best_var = 0;
Prev_MSE_min = MSE_min;
/*
Go through all the inactive variables, and calculate the
resulting prediction error IF this particular variable is added;
keep track of the variable which gives the lowest prediction error.
*/
for (int ivar = 0; ivar < nvar; ivar++) {
if (!bool_vector_iget( stepwise->active_set , ivar)) {
double newR2 = stepwise_test_var(stepwise , ivar , CV_blocks);
if ((minR2 < 0) || (newR2 < minR2)) {
minR2 = newR2;
best_var = ivar;
}
}
}
/*
If the best relative improvement in prediction error is better
than @deltaR2_limit, the corresponding variable is added to the
active set, and we return to repeat the loop one more
time. Otherwise we just exit.
*/
{
MSE_min = minR2;
double deltaR2 = MSE_min / Prev_MSE_min;
if (( currentR2 < 0) || deltaR2 < deltaR2_limit) {
bool_vector_iset( stepwise->active_set , best_var , true );
currentR2 = minR2;
bool_vector_set_all(active_rows, true);
stepwise_estimate__( stepwise , active_rows );
} else {
/* The gain in prediction error is so small that we just leave the building. */
/* NB! Need one final compuation of beta (since the test_var function does not reset the last tested beta value !) */
bool_vector_set_all(active_rows, true);
stepwise_estimate__( stepwise , active_rows );
break;
}
if (bool_vector_count_equal( stepwise->active_set , true) == matrix_get_columns( stepwise->X0 )) {
stepwise_estimate__( stepwise , active_rows );
break; /* All variables are active. */
}
}
}
stepwise_set_R2(stepwise, currentR2);
bool_vector_free( active_rows );
}
static double stepwise_test_var( stepwise_type * stepwise , int test_var , int blocks) {
double prediction_error = 0;
bool_vector_iset( stepwise->active_set , test_var , true ); // Temporarily activate this variable
{
int nvar = matrix_get_columns( stepwise->X0 );
int nsample = matrix_get_rows( stepwise->X0 );
int block_size = nsample / blocks;
bool_vector_type * active_rows = bool_vector_alloc( nsample, true );
/*True Cross-Validation: */
int * randperms = util_calloc( nsample , sizeof * randperms );
for (int i=0; i < nsample; i++)
randperms[i] = i;
/* Randomly perturb ensemble indices */
rng_shuffle_int( stepwise->rng , randperms , nsample );
for (int iblock = 0; iblock < blocks; iblock++) {
int validation_start = iblock * block_size;
int validation_end = validation_start + block_size - 1;
if (iblock == (blocks - 1))
validation_end = nsample - 1;
/*
Ensure that the active_rows vector has a block consisting of
the interval [validation_start : validation_end] which is set to
false, and the remaining part of the vector is set to true.
*/
{
bool_vector_set_all(active_rows, true);
/*
If blocks == 1 that means all datapoint are used in the
regression, and then subsequently reused in the R2
calculation.
*/
if (blocks > 1) {
for (int i = validation_start; i <= validation_end; i++) {
bool_vector_iset( active_rows , randperms[i] , false );
}
}
}
/*
Evaluate the prediction error on the validation part of the
dataset.
*/
{
stepwise_estimate__( stepwise , active_rows );
{
int irow;
matrix_type * x_vector = matrix_alloc( 1 , nvar );
//matrix_type * e_vector = matrix_alloc( 1 , nvar );
for (irow=validation_start; irow <= validation_end; irow++) {
matrix_copy_row( x_vector , stepwise->X0 , 0 , randperms[irow]);
//matrix_copy_row( e_vector , stepwise->E0 , 0 , randperms[irow]);
{
double true_value = matrix_iget( stepwise->Y0 , randperms[irow] , 0 );
double estimated_value = stepwise_eval__( stepwise , x_vector );
prediction_error += (true_value - estimated_value) * (true_value - estimated_value);
//double e_estimated_value = stepwise_eval__( stepwise , e_vector );
//prediction_error += e_estimated_value*e_estimated_value;
}
}
matrix_free( x_vector );
}
}
}
free( randperms );
bool_vector_free( active_rows );
}
/*inactivate the test_var-variable after completion*/
bool_vector_iset( stepwise->active_set , test_var , false );
return prediction_error;
}
