if 'google.colab' in str(get_ipython()):
from google.colab import drive
drive.mount('/content/drive', force_remount=False)
!pip install mirzai
else:The autoreload extension is already loaded. To reload it, use:
%reload_ext autoreload3.2. Train & evaluate (PLSR)
Train & evaluate on multiple train/test splits with different random seeds
# Python utilities
from pathlib import Path
import pickle
# mirzai utilities
from mirzai.data.loading import load_kssl
from mirzai.data.selection import (select_y, select_tax_order, select_X)
from mirzai.data.transform import (log_transform_y, CO2_REGION)
#from mirzai.training.plsr import (compute_valid_curve, PLS_model, Evaluator)
from mirzai.training.plsr import (PLS_model, Learners)
from fastcore.transform import compose
# Data science stack
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
import warnings
warnings.filterwarnings('ignore')Load and transform
src_dir = 'data'
fnames = ['spectra-features.npy', 'spectra-wavenumbers.npy',
'depth-order.npy', 'target.npy',
'tax-order-lu.pkl', 'spectra-id.npy']
X, X_names, depth_order, y, tax_lookup, X_id = load_kssl(src_dir, fnames=fnames)
data = X, y, X_id, depth_order
transforms = [select_y, select_tax_order, select_X, log_transform_y]
X, y, X_id, depth_order = compose(*transforms)(data)print(f'X shape: {X.shape}')
print(f'y shape: {y.shape}')
print(f'Wavenumbers:\n {X_names}')
print(f'depth_order (first 3 rows):\n {depth_order[:3, :]}')
print(f'Taxonomic order lookup:\n {tax_lookup}')X shape: (40132, 1764)
y shape: (40132,)
Wavenumbers:
[3999 3997 3995 ... 603 601 599]
depth_order (first 3 rows):
[[43. 2.]
[ 0. 0.]
[ 0. 1.]]
Taxonomic order lookup:
{'alfisols': 0, 'mollisols': 1, 'inceptisols': 2, 'entisols': 3, 'spodosols': 4, 'undefined': 5, 'ultisols': 6, 'andisols': 7, 'histosols': 8, 'oxisols': 9, 'vertisols': 10, 'aridisols': 11, 'gelisols': 12}Experiment
Setup
split_ratio = 0.1
seeds = range(20)
dest_dir = Path('files/dumps/plsr/train_eval')Train on all Soil Taxonomic Orders
dest_dir_model = Path('files/dumps/plsr/train_eval/all/models')
seeds = range(20)
learners = Learners(tax_lookup, seeds=seeds)
learners.train((X, y, depth_order[:, -1]),
n_cpts_range=range(40, 70, 2),
delta=2e-3,
dest_dir_model=dest_dir_model)--------------------------------------------------------------------------------
Seed: 0
--------------------------------------------------------------------------------
# of components chosen: 50
--------------------------------------------------------------------------------
Seed: 1
--------------------------------------------------------------------------------
# of components chosen: 52
--------------------------------------------------------------------------------
Seed: 2
--------------------------------------------------------------------------------
# of components chosen: 54
--------------------------------------------------------------------------------
Seed: 3
--------------------------------------------------------------------------------
# of components chosen: 56
--------------------------------------------------------------------------------
Seed: 4
--------------------------------------------------------------------------------
# of components chosen: 52
--------------------------------------------------------------------------------
Seed: 5
--------------------------------------------------------------------------------
# of components chosen: 54
--------------------------------------------------------------------------------
Seed: 6
--------------------------------------------------------------------------------
# of components chosen: 58
--------------------------------------------------------------------------------
Seed: 7
--------------------------------------------------------------------------------
# of components chosen: 62
--------------------------------------------------------------------------------
Seed: 8
--------------------------------------------------------------------------------
# of components chosen: 48
--------------------------------------------------------------------------------
Seed: 9
--------------------------------------------------------------------------------
# of components chosen: 52
--------------------------------------------------------------------------------
Seed: 10
--------------------------------------------------------------------------------
# of components chosen: 58
--------------------------------------------------------------------------------
Seed: 11
--------------------------------------------------------------------------------
# of components chosen: 48
--------------------------------------------------------------------------------
Seed: 12
--------------------------------------------------------------------------------
# of components chosen: 48
--------------------------------------------------------------------------------
Seed: 13
--------------------------------------------------------------------------------
# of components chosen: 48
--------------------------------------------------------------------------------
Seed: 14
--------------------------------------------------------------------------------
# of components chosen: 44
--------------------------------------------------------------------------------
Seed: 15
--------------------------------------------------------------------------------
# of components chosen: 68
--------------------------------------------------------------------------------
Seed: 16
--------------------------------------------------------------------------------
# of components chosen: 44
--------------------------------------------------------------------------------
Seed: 17
--------------------------------------------------------------------------------
# of components chosen: 50
--------------------------------------------------------------------------------
Seed: 18
--------------------------------------------------------------------------------
# of components chosen: 64
--------------------------------------------------------------------------------
Seed: 19
--------------------------------------------------------------------------------
# of components chosen: 58Evaluate on all
src_dir_model = Path('files/dumps/plsr/train_eval/all/models')
seeds = range(20)
learners = Learners(tax_lookup, seeds=seeds)
perfs_global_all, y_hats_all, y_trues_all = learners.evaluate((X, y, depth_order[:, -1]),
src_dir_model=src_dir_model)# Save spectific seed y_hat, y_true to plot "Observed vs. predicted" scatterplots
dest_dir_predicted = Path('files/dumps/')
seed = 1
with open(dest_dir_predicted/f'predicted-true-plsr-seed-{seed}.pickle', 'wb') as f:
pickle.dump((y_hats_all[seed].to_numpy(), y_trues_all[seed].to_numpy()), f)perfs_global_all.describe()| rpd | rpiq | r2 | lccc | rmse | mse | mae | mape | bias | stb | |
|---|---|---|---|---|---|---|---|---|---|---|
| count | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 |
| mean | 1.658763 | 2.272073 | 0.636279 | 0.780492 | 1.057159 | 1.306802 | 0.335821 | 135.026036 | -0.000259 | -0.000547 |
| std | 0.022627 | 0.033917 | 0.009834 | 0.005444 | 0.446290 | 1.222881 | 0.016914 | 3.283663 | 0.003848 | 0.007405 |
| min | 1.623276 | 2.204548 | 0.620402 | 0.771287 | 0.671250 | 0.450576 | 0.308898 | 130.007174 | -0.006300 | -0.012485 |
| 25% | 1.640177 | 2.256584 | 0.628185 | 0.777881 | 0.756833 | 0.572822 | 0.325999 | 132.585609 | -0.003009 | -0.005756 |
| 50% | 1.660981 | 2.267233 | 0.637440 | 0.780165 | 0.861632 | 0.742537 | 0.331193 | 134.434034 | -0.001383 | -0.002745 |
| 75% | 1.666828 | 2.296689 | 0.639975 | 0.784414 | 1.093822 | 1.216215 | 0.344576 | 136.850054 | 0.002217 | 0.004202 |
| max | 1.707217 | 2.328779 | 0.656813 | 0.793939 | 2.242110 | 5.027059 | 0.372859 | 140.504871 | 0.007079 | 0.013396 |
src_dir_model = Path('files/dumps/plsr/train_eval/all/models')
seeds = range(20)
for k, v in {k: v for k, v in tax_lookup.items() if k != 'oxisols'}.items():
print(80*'-')
print(f'Test metrics on {k}')
print(80*'-')
learners = Learners(tax_lookup, seeds=seeds)
perfs_global, _, _ = learners.evaluate((X, y, depth_order[:, -1]),
order=v,
src_dir_model=src_dir_model)
print(perfs_global.describe())--------------------------------------------------------------------------------
Test metrics on alfisols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.488581 2.023110 0.544065 0.727878 0.411130 0.198526
std 0.077245 0.111888 0.048141 0.028838 0.176213 0.187989
min 1.335787 1.825327 0.438167 0.673686 0.207121 0.042899
25% 1.429706 1.971203 0.509535 0.713507 0.282230 0.079724
50% 1.493973 2.024866 0.550797 0.731100 0.390250 0.152373
75% 1.533459 2.092083 0.573742 0.749183 0.462969 0.214501
max 1.625880 2.267384 0.620742 0.769198 0.845685 0.715184
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.162636 82.627329 -0.001926 -0.004745
std 0.020796 4.371945 0.008761 0.023061
min 0.128730 73.976273 -0.018855 -0.047651
25% 0.146705 79.772143 -0.009633 -0.025287
50% 0.159680 83.069107 -0.000834 -0.002193
75% 0.179517 85.662533 0.003653 0.009967
max 0.195241 93.340300 0.011908 0.032238
--------------------------------------------------------------------------------
Test metrics on mollisols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.582966 2.090417 0.599952 0.742203 0.769248 0.629664
std 0.034982 0.078073 0.017652 0.012356 0.199794 0.322337
min 1.525933 1.972214 0.570089 0.722419 0.526864 0.277585
25% 1.558545 2.042070 0.587883 0.734876 0.600345 0.360454
50% 1.579016 2.077462 0.598502 0.738004 0.735055 0.541195
75% 1.610838 2.141203 0.614204 0.752421 0.917944 0.842758
max 1.642984 2.250635 0.629145 0.764131 1.123730 1.262768
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.307466 94.298259 0.001662 0.003734
std 0.018656 3.052491 0.008008 0.018876
min 0.279430 87.394573 -0.013899 -0.033922
25% 0.289349 92.537801 -0.004175 -0.009866
50% 0.308846 94.409650 0.001370 0.003153
75% 0.320209 96.313669 0.008152 0.018857
max 0.340558 98.688058 0.014374 0.032833
--------------------------------------------------------------------------------
Test metrics on inceptisols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.484931 2.062821 0.543190 0.715217 0.560004 0.327634
std 0.052383 0.113698 0.032576 0.021247 0.121523 0.141674
min 1.393604 1.834631 0.483181 0.680782 0.383973 0.147435
25% 1.443211 1.980935 0.518045 0.696649 0.462424 0.214457
50% 1.496851 2.072912 0.552023 0.716683 0.563604 0.317676
75% 1.515227 2.150964 0.562994 0.733597 0.621687 0.386500
max 1.588288 2.274688 0.602151 0.752080 0.802342 0.643753
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.258099 117.611669 -0.025303 -0.051641
std 0.030971 6.840199 0.011152 0.023835
min 0.206618 105.015942 -0.046778 -0.093141
25% 0.236181 114.370854 -0.033645 -0.068934
50% 0.253719 116.712206 -0.022784 -0.045203
75% 0.276829 119.444905 -0.017516 -0.032937
max 0.332711 134.933979 -0.007504 -0.014204
--------------------------------------------------------------------------------
Test metrics on entisols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.495344 2.102047 0.535041 0.733857 0.442698 0.204343
std 0.160012 0.348114 0.099822 0.051184 0.093817 0.089328
min 1.229760 1.616803 0.334548 0.616401 0.310090 0.096156
25% 1.435865 1.802874 0.511824 0.723841 0.396742 0.157491
50% 1.505428 2.070569 0.555707 0.748509 0.417861 0.174608
75% 1.587539 2.257373 0.600692 0.761758 0.506173 0.256319
max 1.916463 2.839117 0.725927 0.830845 0.663214 0.439853
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.227277 124.306496 -0.013363 -0.027085
std 0.033105 15.203068 0.022993 0.048705
min 0.172270 102.336085 -0.045416 -0.111028
25% 0.209802 113.533998 -0.030312 -0.054147
50% 0.226905 119.691934 -0.020139 -0.041128
75% 0.247482 131.740359 -0.003590 -0.006261
max 0.287580 161.354752 0.034848 0.076445
--------------------------------------------------------------------------------
Test metrics on spodosols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.884385 2.713797 0.699472 0.824446 0.482583 0.249341
std 0.238908 0.639696 0.079492 0.043133 0.131607 0.130359
min 1.390288 1.803467 0.474431 0.721559 0.276331 0.076359
25% 1.712985 2.360031 0.652982 0.792727 0.385248 0.148728
50% 1.870642 2.527430 0.709356 0.835642 0.477664 0.228171
75% 2.015491 3.008323 0.749113 0.844006 0.557888 0.312005
max 2.398842 4.506760 0.823003 0.895621 0.706393 0.498991
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.262912 157.167552 0.014543 0.020080
std 0.058003 19.733897 0.028305 0.043673
min 0.171519 126.395882 -0.019549 -0.047892
25% 0.219475 142.879484 -0.001157 -0.002089
50% 0.263831 156.169654 0.007113 0.011952
75% 0.296229 171.764491 0.019708 0.037011
max 0.359890 188.821685 0.101276 0.143619
--------------------------------------------------------------------------------
Test metrics on undefined
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.699452 2.327902 0.653167 0.788484 1.152385 1.582740
std 0.032766 0.070709 0.013227 0.007203 0.517839 1.675655
min 1.648992 2.217120 0.632006 0.777364 0.766317 0.587241
25% 1.677440 2.275579 0.644376 0.781214 0.880365 0.775066
50% 1.692550 2.328214 0.650700 0.788912 0.935093 0.874408
75% 1.724165 2.352930 0.663386 0.795110 1.060103 1.132949
max 1.757505 2.475099 0.676041 0.798610 2.462503 6.063923
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.424512 152.988067 0.004325 0.007633
std 0.029141 7.248857 0.006910 0.012443
min 0.362205 141.274022 -0.006812 -0.012853
25% 0.407466 148.161140 -0.001129 -0.002094
50% 0.424653 151.755239 0.003296 0.005683
75% 0.443348 157.515627 0.009594 0.017346
max 0.475289 168.416540 0.016393 0.029619
--------------------------------------------------------------------------------
Test metrics on ultisols
-------------------------------------------------------------------------------- rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.201589 1.665164 0.296985 0.581148 0.326529 0.109017
std 0.067696 0.152486 0.081464 0.056106 0.050224 0.034916
min 1.079312 1.352102 0.136519 0.489327 0.252949 0.063983
25% 1.151943 1.560671 0.242226 0.546748 0.295477 0.087307
50% 1.207474 1.676421 0.310236 0.584230 0.321999 0.103689
75% 1.237247 1.789812 0.343037 0.616695 0.350845 0.123099
max 1.329557 1.890884 0.430578 0.701262 0.450793 0.203214
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.181925 80.909836 0.039877 0.094603
std 0.025022 9.029538 0.014663 0.031388
min 0.150119 67.915820 0.013417 0.031652
25% 0.164187 74.465798 0.025832 0.068632
50% 0.174361 80.450314 0.037480 0.100163
75% 0.197729 88.393075 0.050413 0.112495
max 0.249080 101.466523 0.065988 0.147605
--------------------------------------------------------------------------------
Test metrics on andisols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.561501 2.013977 0.584894 0.728846 0.592610 0.428996
std 0.054676 0.214017 0.028844 0.027988 0.286190 0.451123
min 1.462280 1.607348 0.528559 0.677337 0.344830 0.118907
25% 1.522535 1.939558 0.564517 0.712462 0.397968 0.158418
50% 1.560284 2.031585 0.585664 0.727166 0.479231 0.229689
75% 1.594774 2.100454 0.603265 0.746108 0.630045 0.397882
max 1.659770 2.348595 0.633790 0.781835 1.299265 1.688089
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.281092 106.394236 0.015788 0.033443
std 0.049266 11.345530 0.017747 0.037467
min 0.205634 84.861332 -0.015052 -0.033935
25% 0.241596 98.732944 0.005526 0.013816
50% 0.288964 106.024153 0.013751 0.028963
75% 0.307314 112.704328 0.029538 0.064234
max 0.390000 130.842886 0.043541 0.094777
--------------------------------------------------------------------------------
Test metrics on histosols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.711375 2.500980 0.637454 0.778795 1.165003 1.492735
std 0.214934 0.635935 0.090172 0.055099 0.377669 1.060323
min 1.374536 1.609684 0.462316 0.677843 0.634321 0.402363
25% 1.566681 2.034666 0.586609 0.753988 0.927356 0.860062
50% 1.681201 2.410043 0.641241 0.776610 1.093495 1.195750
75% 1.820641 2.907901 0.693132 0.811247 1.278496 1.634970
max 2.161752 3.628460 0.782561 0.867816 2.175235 4.731648
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.610852 243.155805 0.001554 -0.003005
std 0.133948 46.102071 0.035374 0.045302
min 0.380170 172.238410 -0.064465 -0.110639
25% 0.546008 210.254481 -0.010255 -0.015323
50% 0.567238 231.883135 -0.000548 -0.000847
75% 0.674668 284.666143 0.009428 0.010923
max 0.865779 324.890279 0.077207 0.078856
--------------------------------------------------------------------------------
Test metrics on vertisols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.524148 2.101732 0.547234 0.732481 0.390509 0.161691
std 0.175537 0.331334 0.101618 0.061666 0.098375 0.090253
min 1.233878 1.556100 0.333783 0.607804 0.279383 0.078055
25% 1.428539 1.912700 0.503573 0.695811 0.327822 0.107521
50% 1.478822 2.143325 0.536402 0.728507 0.361795 0.130898
75% 1.632835 2.287894 0.619066 0.781692 0.423892 0.179688
max 1.895009 2.787665 0.717768 0.831755 0.673910 0.454155
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.238076 98.744299 -0.052767 -0.134787
std 0.035953 12.347859 0.015789 0.043975
min 0.183381 83.009898 -0.075401 -0.206095
25% 0.206193 89.681748 -0.061264 -0.162066
50% 0.231768 94.428492 -0.055714 -0.138102
75% 0.259825 106.099271 -0.046249 -0.110102
max 0.309477 127.327009 -0.015200 -0.042732
--------------------------------------------------------------------------------
Test metrics on aridisols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.378008 1.827588 0.444671 0.686068 2.131294 11.685892
std 0.164302 0.273668 0.154683 0.067033 2.742162 28.829063
min 1.015784 1.353244 0.024153 0.533021 0.484385 0.234629
25% 1.294828 1.616959 0.398816 0.656175 0.657982 0.433422
50% 1.420750 1.811119 0.501060 0.704963 1.069559 1.148925
75% 1.471227 2.008619 0.534730 0.727574 1.995471 4.232994
max 1.643348 2.293749 0.626948 0.771557 10.787938 116.379602
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.500131 145.908265 -0.010957 -0.025541
std 0.250221 56.170231 0.018981 0.046494
min 0.294676 96.568375 -0.047555 -0.126001
25% 0.352105 113.472506 -0.021256 -0.045881
50% 0.403889 124.989228 -0.007535 -0.013668
75% 0.562606 154.904255 0.004738 0.008801
max 1.257594 322.681936 0.015741 0.036257
--------------------------------------------------------------------------------
Test metrics on gelisols
--------------------------------------------------------------------------------
rpd rpiq r2 lccc rmse mse \
count 20.000000 20.000000 20.000000 20.000000 20.000000 20.000000
mean 1.692878 2.534223 0.626330 0.766703 0.646289 0.435929
std 0.200619 0.761655 0.096257 0.059995 0.138561 0.188936
min 1.274587 1.094882 0.368254 0.619306 0.418552 0.175186
25% 1.607986 2.005345 0.602193 0.747244 0.552801 0.305750
50% 1.651799 2.517481 0.624528 0.765263 0.631649 0.398982
75% 1.849051 2.810388 0.698211 0.819691 0.723203 0.523037
max 2.033034 3.926091 0.752298 0.864522 0.954032 0.910176
mae mape bias stb
count 20.000000 20.000000 20.000000 20.000000
mean 0.389866 193.838336 -0.088982 -0.145424
std 0.084392 31.399886 0.034638 0.082395
min 0.255306 134.782910 -0.161743 -0.406283
25% 0.324760 169.096695 -0.107498 -0.178604
50% 0.390097 191.552850 -0.092811 -0.129146
75% 0.459532 213.219983 -0.058120 -0.106705
max 0.542090 257.645578 -0.023931 -0.030365 Evaluate on Mollisols
src_dir_model = Path('files/dumps/plsr/train_eval/all/models')
seeds = range(20)
order = 1
learners = Learners(tax_lookup, seeds=seeds)
perfs_global_mollisols, _, _ = learners.evaluate((X, y, depth_order[:, -1]),
order=1,
src_dir_model=src_dir_model)perfs_global_mollisols.describe()| rpd | rpiq | r2 | lccc | rmse | mse | mae | mape | bias | stb | |
|---|---|---|---|---|---|---|---|---|---|---|
| count | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 |
| mean | 1.582966 | 2.090417 | 0.599952 | 0.742203 | 0.769248 | 0.629664 | 0.307466 | 94.298259 | 0.001662 | 0.003734 |
| std | 0.034982 | 0.078073 | 0.017652 | 0.012356 | 0.199794 | 0.322337 | 0.018656 | 3.052491 | 0.008008 | 0.018876 |
| min | 1.525933 | 1.972214 | 0.570089 | 0.722419 | 0.526864 | 0.277585 | 0.279430 | 87.394573 | -0.013899 | -0.033922 |
| 25% | 1.558545 | 2.042070 | 0.587883 | 0.734876 | 0.600345 | 0.360454 | 0.289349 | 92.537801 | -0.004175 | -0.009866 |
| 50% | 1.579016 | 2.077462 | 0.598502 | 0.738004 | 0.735055 | 0.541195 | 0.308846 | 94.409650 | 0.001370 | 0.003153 |
| 75% | 1.610838 | 2.141203 | 0.614204 | 0.752421 | 0.917944 | 0.842758 | 0.320209 | 96.313669 | 0.008152 | 0.018857 |
| max | 1.642984 | 2.250635 | 0.629145 | 0.764131 | 1.123730 | 1.262768 | 0.340558 | 98.688058 | 0.014374 | 0.032833 |
Evaluate on Gelisols
src_dir_model = Path('files/dumps/plsr/train_eval/all/models')
seeds = range(20)
order = 12
learners = Learners(tax_lookup, seeds=seeds)
perfs_global_gelisols, _, _ = learners.evaluate((X, y, depth_order[:, -1]),
order=order,
src_dir_model=src_dir_model)perfs_global_gelisols.describe()| rpd | rpiq | r2 | lccc | rmse | mse | mae | mape | bias | stb | |
|---|---|---|---|---|---|---|---|---|---|---|
| count | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 |
| mean | 1.692878 | 2.534223 | 0.626330 | 0.766703 | 0.646289 | 0.435929 | 0.389866 | 193.838336 | -0.088982 | -0.145424 |
| std | 0.200619 | 0.761655 | 0.096257 | 0.059995 | 0.138561 | 0.188936 | 0.084392 | 31.399886 | 0.034638 | 0.082395 |
| min | 1.274587 | 1.094882 | 0.368254 | 0.619306 | 0.418552 | 0.175186 | 0.255306 | 134.782910 | -0.161743 | -0.406283 |
| 25% | 1.607986 | 2.005345 | 0.602193 | 0.747244 | 0.552801 | 0.305750 | 0.324760 | 169.096695 | -0.107498 | -0.178604 |
| 50% | 1.651799 | 2.517481 | 0.624528 | 0.765263 | 0.631649 | 0.398982 | 0.390097 | 191.552850 | -0.092811 | -0.129146 |
| 75% | 1.849051 | 2.810388 | 0.698211 | 0.819691 | 0.723203 | 0.523037 | 0.459532 | 213.219983 | -0.058120 | -0.106705 |
| max | 2.033034 | 3.926091 | 0.752298 | 0.864522 | 0.954032 | 0.910176 | 0.542090 | 257.645578 | -0.023931 | -0.030365 |
Evaluate on Vertisols
src_dir_model = Path('files/dumps/plsr/train_eval/all/models')
seeds = range(20)
order = 10
learners = Learners(tax_lookup, seeds=seeds)
perfs_global_vertisols, _, _ = learners.evaluate((X, y, depth_order[:, -1]),
order=order,
src_dir_model=src_dir_model)perfs_global_vertisols.describe()| rpd | rpiq | r2 | lccc | rmse | mse | mae | mape | bias | stb | |
|---|---|---|---|---|---|---|---|---|---|---|
| count | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 |
| mean | 1.524148 | 2.101732 | 0.547234 | 0.732481 | 0.390509 | 0.161691 | 0.238076 | 98.744299 | -0.052767 | -0.134787 |
| std | 0.175537 | 0.331334 | 0.101618 | 0.061666 | 0.098375 | 0.090253 | 0.035953 | 12.347859 | 0.015789 | 0.043975 |
| min | 1.233878 | 1.556100 | 0.333783 | 0.607804 | 0.279383 | 0.078055 | 0.183381 | 83.009898 | -0.075401 | -0.206095 |
| 25% | 1.428539 | 1.912700 | 0.503573 | 0.695811 | 0.327822 | 0.107521 | 0.206193 | 89.681748 | -0.061264 | -0.162066 |
| 50% | 1.478822 | 2.143325 | 0.536402 | 0.728507 | 0.361795 | 0.130898 | 0.231768 | 94.428492 | -0.055714 | -0.138102 |
| 75% | 1.632835 | 2.287894 | 0.619066 | 0.781692 | 0.423892 | 0.179688 | 0.259825 | 106.099271 | -0.046249 | -0.110102 |
| max | 1.895009 | 2.787665 | 0.717768 | 0.831755 | 0.673910 | 0.454155 | 0.309477 | 127.327009 | -0.015200 | -0.042732 |
Train and test on Mollisols
dest_dir_model = Path('files/dumps/plsr/train_eval/mollisols/models')
seeds = range(20)
order = 1
learners = Learners(tax_lookup, seeds=seeds)
learners.train((X, y, depth_order[:, -1]),
order=order,
n_cpts_range=range(30, 60, 2),
delta=2e-3,
dest_dir_model=dest_dir_model)--------------------------------------------------------------------------------
Seed: 0
--------------------------------------------------------------------------------
# of components chosen: 46
--------------------------------------------------------------------------------
Seed: 1
--------------------------------------------------------------------------------
# of components chosen: 52
--------------------------------------------------------------------------------
Seed: 2
--------------------------------------------------------------------------------
# of components chosen: 34
--------------------------------------------------------------------------------
Seed: 3
--------------------------------------------------------------------------------
# of components chosen: 40
--------------------------------------------------------------------------------
Seed: 4
--------------------------------------------------------------------------------
# of components chosen: 46
--------------------------------------------------------------------------------
Seed: 5
--------------------------------------------------------------------------------
# of components chosen: 34
--------------------------------------------------------------------------------
Seed: 6
--------------------------------------------------------------------------------
# of components chosen: 40
--------------------------------------------------------------------------------
Seed: 7
--------------------------------------------------------------------------------
# of components chosen: 40
--------------------------------------------------------------------------------
Seed: 8
--------------------------------------------------------------------------------
# of components chosen: 40
--------------------------------------------------------------------------------
Seed: 9
--------------------------------------------------------------------------------
# of components chosen: 34
--------------------------------------------------------------------------------
Seed: 10
--------------------------------------------------------------------------------
# of components chosen: 54
--------------------------------------------------------------------------------
Seed: 11
--------------------------------------------------------------------------------
# of components chosen: 34
--------------------------------------------------------------------------------
Seed: 12
--------------------------------------------------------------------------------
# of components chosen: 34
--------------------------------------------------------------------------------
Seed: 13
--------------------------------------------------------------------------------
# of components chosen: 34
--------------------------------------------------------------------------------
Seed: 14
--------------------------------------------------------------------------------
# of components chosen: 56
--------------------------------------------------------------------------------
Seed: 15
--------------------------------------------------------------------------------
# of components chosen: 34
--------------------------------------------------------------------------------
Seed: 16
--------------------------------------------------------------------------------
# of components chosen: 34
--------------------------------------------------------------------------------
Seed: 17
--------------------------------------------------------------------------------
# of components chosen: 40
--------------------------------------------------------------------------------
Seed: 18
--------------------------------------------------------------------------------
# of components chosen: 50
--------------------------------------------------------------------------------
Seed: 19
--------------------------------------------------------------------------------
# of components chosen: 34src_dir_model = Path('files/dumps/plsr/train_eval/mollisols/models')
seeds = range(20)
order = 1
learners = Learners(tax_lookup, seeds=seeds)
perfs_local_mollisols, _, _ = learners.evaluate((X, y, depth_order[:, -1]),
order=order,
src_dir_model=src_dir_model)
perfs_local_mollisols.describe()| rpd | rpiq | r2 | lccc | rmse | mse | mae | mape | bias | stb | |
|---|---|---|---|---|---|---|---|---|---|---|
| count | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 |
| mean | 1.729961 | 2.284160 | 0.664719 | 0.803311 | 0.682758 | 0.501425 | 0.271352 | 102.920252 | 0.000623 | 0.001149 |
| std | 0.049480 | 0.085176 | 0.019570 | 0.011727 | 0.192671 | 0.281706 | 0.014817 | 3.289161 | 0.008043 | 0.018940 |
| min | 1.620689 | 2.157094 | 0.618909 | 0.776560 | 0.431012 | 0.185771 | 0.243688 | 95.089908 | -0.016145 | -0.039405 |
| 25% | 1.704100 | 2.221866 | 0.655256 | 0.801153 | 0.535298 | 0.286687 | 0.261261 | 100.989098 | -0.005376 | -0.012943 |
| 50% | 1.734100 | 2.282031 | 0.667110 | 0.803882 | 0.609356 | 0.371682 | 0.273628 | 103.076593 | -0.001222 | -0.002843 |
| 75% | 1.759684 | 2.322080 | 0.676704 | 0.809385 | 0.793591 | 0.630660 | 0.275906 | 105.139386 | 0.007436 | 0.017265 |
| max | 1.827361 | 2.468705 | 0.700207 | 0.826625 | 1.028421 | 1.057650 | 0.303412 | 108.383273 | 0.013749 | 0.033221 |
Train and test on Gelisols
dest_dir_model = Path('files/dumps/plsr/train_eval/gelisols/models')
seeds = range(20)
order = 12
learners = Learners(tax_lookup, seeds=seeds)
learners.train((X, y, depth_order[:, -1]),
order=order,
n_cpts_range=range(2, 10, 1),
delta=1e-2,
dest_dir_model=dest_dir_model)--------------------------------------------------------------------------------
Seed: 0
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 1
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 2
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 3
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 4
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 5
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 6
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 7
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 8
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 9
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 10
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 11
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 12
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 13
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 14
--------------------------------------------------------------------------------
# of components chosen: 8
--------------------------------------------------------------------------------
Seed: 15
--------------------------------------------------------------------------------
# of components chosen: 7
--------------------------------------------------------------------------------
Seed: 16
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 17
--------------------------------------------------------------------------------
# of components chosen: 7
--------------------------------------------------------------------------------
Seed: 18
--------------------------------------------------------------------------------
# of components chosen: 7
--------------------------------------------------------------------------------
Seed: 19
--------------------------------------------------------------------------------
# of components chosen: 9src_dir_model = Path('files/dumps/plsr/train_eval/gelisols/models')
seeds = range(20)
order = 12
learners = Learners(tax_lookup, seeds=seeds)
perfs_local_gelisols, _, _ = learners.evaluate((X, y, depth_order[:, -1]),
order=order,
src_dir_model=src_dir_model)
perfs_local_gelisols.describe()| rpd | rpiq | r2 | lccc | rmse | mse | mae | mape | bias | stb | |
|---|---|---|---|---|---|---|---|---|---|---|
| count | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 |
| mean | 2.077886 | 3.084523 | 0.746774 | 0.858587 | 0.646908 | 0.452289 | 0.319878 | 205.707385 | -0.002118 | -0.006263 |
| std | 0.308216 | 0.896429 | 0.076636 | 0.042218 | 0.188620 | 0.248623 | 0.069856 | 44.668977 | 0.040502 | 0.064437 |
| min | 1.547630 | 1.263899 | 0.570562 | 0.773000 | 0.348828 | 0.121681 | 0.208846 | 122.575289 | -0.067213 | -0.125266 |
| 25% | 1.823879 | 2.564071 | 0.692319 | 0.827514 | 0.530327 | 0.281372 | 0.269832 | 172.185662 | -0.025272 | -0.038202 |
| 50% | 2.053352 | 3.090948 | 0.756956 | 0.857239 | 0.614913 | 0.378319 | 0.308076 | 204.408200 | -0.007451 | -0.014036 |
| 75% | 2.299863 | 3.404348 | 0.805802 | 0.899488 | 0.817420 | 0.669407 | 0.361622 | 227.222367 | 0.016758 | 0.020807 |
| max | 2.652731 | 5.047128 | 0.851434 | 0.914563 | 0.913832 | 0.835089 | 0.452944 | 302.505106 | 0.103336 | 0.170306 |
Train and test on Vertisols
dest_dir_model = Path('files/dumps/plsr/train_eval/vertisols/models')
seeds = range(20)
order = 10
learners = Learners(tax_lookup, seeds=seeds)
learners.train((X, y, depth_order[:, -1]),
order=order,
n_cpts_range=range(2, 30, 1),
delta=1e-2,
dest_dir_model=dest_dir_model)--------------------------------------------------------------------------------
Seed: 0
--------------------------------------------------------------------------------
# of components chosen: 11
--------------------------------------------------------------------------------
Seed: 1
--------------------------------------------------------------------------------
# of components chosen: 11
--------------------------------------------------------------------------------
Seed: 2
--------------------------------------------------------------------------------
# of components chosen: 20
--------------------------------------------------------------------------------
Seed: 3
--------------------------------------------------------------------------------
# of components chosen: 18
--------------------------------------------------------------------------------
Seed: 4
--------------------------------------------------------------------------------
# of components chosen: 12
--------------------------------------------------------------------------------
Seed: 5
--------------------------------------------------------------------------------
# of components chosen: 10
--------------------------------------------------------------------------------
Seed: 6
--------------------------------------------------------------------------------
# of components chosen: 17
--------------------------------------------------------------------------------
Seed: 7
--------------------------------------------------------------------------------
# of components chosen: 8
--------------------------------------------------------------------------------
Seed: 8
--------------------------------------------------------------------------------
# of components chosen: 15
--------------------------------------------------------------------------------
Seed: 9
--------------------------------------------------------------------------------
# of components chosen: 11
--------------------------------------------------------------------------------
Seed: 10
--------------------------------------------------------------------------------
# of components chosen: 9
--------------------------------------------------------------------------------
Seed: 11
--------------------------------------------------------------------------------
# of components chosen: 12
--------------------------------------------------------------------------------
Seed: 12
--------------------------------------------------------------------------------
# of components chosen: 19
--------------------------------------------------------------------------------
Seed: 13
--------------------------------------------------------------------------------
# of components chosen: 11
--------------------------------------------------------------------------------
Seed: 14
--------------------------------------------------------------------------------
# of components chosen: 12
--------------------------------------------------------------------------------
Seed: 15
--------------------------------------------------------------------------------
# of components chosen: 15
--------------------------------------------------------------------------------
Seed: 16
--------------------------------------------------------------------------------
# of components chosen: 14
--------------------------------------------------------------------------------
Seed: 17
--------------------------------------------------------------------------------
# of components chosen: 12
--------------------------------------------------------------------------------
Seed: 18
--------------------------------------------------------------------------------
# of components chosen: 8
--------------------------------------------------------------------------------
Seed: 19
--------------------------------------------------------------------------------
# of components chosen: 11src_dir_model = Path('files/dumps/plsr/train_eval/vertisols/models')
seeds = range(20)
order = 10
learners = Learners(tax_lookup, seeds=seeds)
perfs_local_vertisols, _, _ = learners.evaluate((X, y, depth_order[:, -1]),
order=order,
src_dir_model=src_dir_model)
perfs_local_vertisols.describe()| rpd | rpiq | r2 | lccc | rmse | mse | mae | mape | bias | stb | |
|---|---|---|---|---|---|---|---|---|---|---|
| count | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 | 20.000000 |
| mean | 2.009008 | 2.772106 | 0.737918 | 0.855452 | 0.285306 | 0.084297 | 0.187213 | 93.891373 | -0.004065 | -0.011017 |
| std | 0.233000 | 0.461799 | 0.071342 | 0.035866 | 0.055230 | 0.033674 | 0.024124 | 11.393788 | 0.020619 | 0.053431 |
| min | 1.410521 | 1.902507 | 0.490198 | 0.741338 | 0.182039 | 0.033138 | 0.129494 | 76.538749 | -0.041485 | -0.121708 |
| 25% | 1.894380 | 2.447912 | 0.717390 | 0.849941 | 0.247126 | 0.061071 | 0.173790 | 85.490789 | -0.016195 | -0.042687 |
| 50% | 2.024185 | 2.805916 | 0.752705 | 0.860249 | 0.283139 | 0.080169 | 0.188396 | 91.373366 | -0.002301 | -0.006435 |
| 75% | 2.081399 | 3.036756 | 0.766082 | 0.870251 | 0.299627 | 0.089804 | 0.204240 | 102.965234 | 0.007755 | 0.018465 |
| max | 2.549723 | 3.779666 | 0.843884 | 0.916982 | 0.418275 | 0.174954 | 0.223513 | 113.539148 | 0.036139 | 0.086563 |
Compile metrics for “local vs. global” Fig. 6
def format_metrics(*dfs):
perfs = {'r2': {'mean': [], 'std': []},
'mape': {'mean': [], 'std': []}}
for df in dfs:
for metric in ['r2', 'mape']:
mean, std = df.describe().loc[['mean', 'std'], metric].items()
perfs[metric]['mean'].append(mean[1])
perfs[metric]['std'].append(std[1])
return perfsperfs = {}
perfs['global'] = format_metrics(perfs_global_mollisols, perfs_global_gelisols, perfs_global_vertisols)
perfs['local'] = format_metrics(perfs_local_mollisols, perfs_local_gelisols, perfs_local_vertisols)perfs{'global': {'r2': {'mean': [0.5999520975577833,
0.6263296407122689,
0.5472338139745212],
'std': [0.01765197454826005, 0.09625659382541236, 0.10161782452206956]},
'mape': {'mean': [94.29825920533892, 193.83833562261376, 98.74429900717739],
'std': [3.0524914231457974, 31.399886042493737, 12.347859402802634]}},
'local': {'r2': {'mean': [0.6647188332987304,
0.7467735410966299,
0.737918199125363],
'std': [0.01957008650036342, 0.07663555355310155, 0.07134199041747523]},
'mape': {'mean': [102.92025183047231, 205.70738511998024, 93.89137312512031],
'std': [3.289160840803382, 44.668977271871476, 11.393787965933955]}}}dest_dir = Path('files/dumps/plsr')
with open(dest_dir/'global_vs_local.pickle', 'wb') as f:
pickle.dump(perfs, f)