In Enlightengeo/GeoscienceBC_2019-008: Enlighten Geoscience BC 2019-008 Induced Seismicity Statistical Study Code

Geoscience BC Study Code - Machine Learning Wrapper

library(geosciencebc2019008)
run_date = Sys.Date() %>% str_replace_all(., "-", "_")
knitr::opts_chunk$set(eval = FALSE)

Load prepared data

This chunk loads our prepared data (see prepare_data.Rmd)

saf_wells_sf <- read_rds('../wcfd_data/wcfd_saf_well_data.rds')
wcfd_data = TRUE

ML Dataframe Prep

This chunk prepares our dataframe for the data science workflows (exploratory data analysis, feature selection, model generation, and model interpretation). It specifies auxillary columns (those useful for information), predictor columns, and our target column(s), which for classificatio is the seimogenic (T/F) association. Some of the columns, filtering, and imputation are specific for the classification problem, although the bits of code are very similar.

aux_cols <- c(
  "unique_surv_id", "wa_num", "ground_elevtn",
  "mean_easting", "mean_northing", "on_prod_date"
  )

class_comp_feats = c(
  "calc_total_fluid_m3", "mean_rate_m3_min", "mean_proppant_per_stage_t",
  "proppant_cat", "horiz_wells_in_5km", "min_midpoint_dist",
  "calc_completed_length_m"
  )

class_geo_feats = c(
  "paleozoic_structure_mss", "geothermal_gradient_degc_km", "shmin_grasby",
  "distance_listric_faults_berger_m", "distance_normal_faults_berger_m"
  )

targets <- c("seismogenic", "max_mag")

ml_df <- saf_wells_sf %>%
  st_drop_geometry() %>% 
  dplyr::select(
    all_of(aux_cols), all_of(class_comp_feats), all_of(class_geo_feats), all_of(targets)) %>%
  dplyr::mutate_if(is.logical, as.numeric) %>%
  dplyr::filter(
    shmin_grasby > 0,
    !is.na(mean_proppant_per_stage_t),
    is.finite(calc_completed_length_m),
    min_midpoint_dist > 0) %>%
  dplyr::mutate(seismogenic = as.factor(seismogenic))

# impute non-causal features
ml_df = impute(ml_df, cols = list(
  mean_rate_m3_min = imputeMean()
  ))$data

# one hot encode categorical variables
ml_df <- createDummyFeatures(
  ml_df, target = "seismogenic", cols = c("proppant_cat")) %>% 
  removeConstantFeatures(.) %>%
  dplyr::select(
    -proppant_cat.ceramic, -proppant_cat.resin.coated, 
    -proppant_cat.sand, -proppant_cat.unknown) %>%
  tidyr::drop_na()

write_csv(ml_df, paste0("../wcfd_data/","final_wcfd_mldf_class.csv"))
write_rds(ml_df, paste0("../wcfd_data/","final_wcfd_mldf_class.rds"))

Write csv and rds for future use

regr_comp_feats  = c(
  "calc_total_fluid_m3", "calc_total_proppant_t", "calc_completed_length_m",
  "n_stages", "min_midpoint_dist", "mean_rate_m3_min"
  )

regr_geo_feats = c(
  "pressure_depth_ratio_kpa_m", "top_montney_structure_mss", "third_order_residual_m",
  "geothermal_gradient_degc_km", "distance_listric_faults_berger_m",
  "distance_normal_faults_berger_m"
  )

targets <- c("seismogenic", "max_mag")

ml_df <- saf_wells_sf %>%
  st_drop_geometry() %>% 
  dplyr::select(
    all_of(aux_cols), all_of(regr_comp_feats), all_of(regr_geo_feats), all_of(targets)) %>%
  dplyr::mutate_if(is.logical, as.numeric) %>%
  dplyr::filter(
    min_midpoint_dist > 0,
    is.finite(calc_completed_length_m)) %>%
  dplyr::mutate(seismogenic = as.factor(seismogenic))

# impute non-causal features
ml_df = impute(ml_df, cols = list(
  mean_rate_m3_min = imputeMean()
  ))$data %>%
  tidyr::drop_na()

write_csv(ml_df, paste0("../wcfd_data/","final_wcfd_mldf_regr.csv"))
write_rds(ml_df, paste0("../wcfd_data/","final_wcfd_mldf_regr.rds"))

Enlightengeo/GeoscienceBC_2019-008 documentation built on Feb. 4, 2021, 12:43 p.m.