test_integration_3inversions.py 18.7 KB
Newer Older
1
2
3
import os
import time
import shutil
4
5
6
import numpy as np
import matplotlib.pyplot as plt
import xarray as xr
Antoine Berchet's avatar
Antoine Berchet committed
7
from pathlib import Path
8
9
10
11
12
13
14
15
16
17
18
19
20

import pytest

from pycif.utils.classes.setup import Setup
from pycif.utils.datastores.dump import read_datastore
from pycif.utils.yml import ordered_dump
from pycif.utils.path import init_dir


@pytest.mark.dummy
@pytest.mark.article
@pytest.mark.parametrize(
    "settings", [
Antoine Berchet's avatar
Antoine Berchet committed
21
        {"mode": "4dvar", "minimizer": "M1QN3"},
22
23
24
25
26
27
28
        pytest.param({"mode": "4dvar", "minimizer": "M1QN3", "montecarlo": 10},
                     marks=pytest.mark.uncertainties),
        {"mode": "4dvar", "minimizer": "congrad"},
        {"mode": "ensrf"},
        pytest.param({"mode": "ensrf", "nsample": 5},
                     marks=pytest.mark.uncertainties),
        {"mode": "analytical"}
29
    ]
30
)
Antoine Berchet's avatar
Antoine Berchet committed
31
def test_integration_inversion(dummy_config_inversion, settings, pytestconfig):
32
33
34
35
36
37
    """
    Integration test that runs the dummy_forward model.
    """

    tmpdir, config, tag = dummy_config_inversion
    
Antoine Berchet's avatar
Antoine Berchet committed
38
39
40
41
    # Force reloading observation operator
    # from forward to make computation faster
    config["model"]["reload_H"] = "{}/../H_matrix.pickle".format(tmpdir)
    
Antoine Berchet's avatar
Antoine Berchet committed
42
43
    # Update number of simulations depending on pytests options
    marker = pytestconfig.getoption('-m')
44
    nsimmax = 10
Antoine Berchet's avatar
Antoine Berchet committed
45
46
47
    montecarlo = settings.get("montecarlo", 10)
    nsample = settings.get("nsample", 5)
    if "allsimulations" not in marker:
Antoine Berchet's avatar
Antoine Berchet committed
48
49
        nsimmax = 3
        montecarlo = 3
Antoine Berchet's avatar
Antoine Berchet committed
50
51
52
        nsample = 1
        
    # Changing mode
53
    if config["datavect"]["components"]["flux"]\
54
            ["parameters"]["CH4"]["hresol"] == "hpixels":
Antoine Berchet's avatar
Antoine Berchet committed
55
        nsimmax = int(2.5 * nsimmax)
56
    
57
    elif config["datavect"]["components"]["flux"]\
58
            ["parameters"]["CH4"]["hresol"] == "global" \
59
            and settings.get("minimizer", "") == "congrad":
60
61
62
63
64
65
66
67
68
69
70
71
72
        nsimmax = 1
        
    if settings["mode"] == "4dvar":
        mode = {
            "plugin": {"name": "4dvar", "version": "std"},
            "minimizer": {
                "plugin": {"name": settings["minimizer"], "version": "std"},
                "simulator": {
                    "plugin": {"name": "gausscost", "version": "std"},
                    "reload_from_previous": True
                },
                "maxiter": nsimmax,
                "nsim": nsimmax,
73
74
                "epsg": 0.0002,
                "df1": 0.5
75
76
77
            },
            "save_out_netcdf": True
        }
78
79
80
81
        if settings["minimizer"] == "congrad":
            mode["minimizer"]["save_uncertainties"] = True
        
        if "montecarlo" in settings:
82
83
84
85
86
            mode["montecarlo"] = {
                "nsample": settings["montecarlo"],
                "perturb_x": True,
                "perturb_y": False
            }
87
        
88
89
90
91
92
93
94
95
96
97
        prior_dir = os.path.join(tmpdir, "obsoperator/fwd_-001/obsvect")
        posterior_dir = os.path.join(tmpdir, "obsoperator/fwd_-002/obsvect")
    
    elif settings["mode"] == "analytical":
        mode = {"plugin": {"name": "analytic", "version": "std",
                           "dump_nc_base_control": True}}
        prior_dir = os.path.join(tmpdir, "obsvect_prior")
        posterior_dir = os.path.join(tmpdir, "obsvect_posterior")
    
    elif settings["mode"] == "ensrf":
Antoine Berchet's avatar
Antoine Berchet committed
98
        nsimmax = 2 * settings.get("nsample", 1) * nsimmax
99
100
101
102
103
104
        mode = {"plugin": {"name": "EnSRF", "version": "std"},
                "nsample": nsimmax}
        prior_dir = os.path.join(tmpdir, "obsvect_prior")
        posterior_dir = os.path.join(tmpdir, "obsvect_posterior")
      
    config["mode"] = mode
105
106
107
108
109
110
    config["platform"] = {
        "plugin":
            {"name": "LSCE", "version": "obelix"}
            if os.getenv("PYCIF_PLATFORM") == "LSCE"
            else {"name": "docker", "version": "cif"}
    }
Antoine Berchet's avatar
Antoine Berchet committed
111
112
113

    # Save ID for later plot of cost function
    resolution = \
114
        config["datavect"]["components"]["flux"]["parameters"]["CH4"][
Antoine Berchet's avatar
Antoine Berchet committed
115
116
            "hresol"]
    sigma = config["datavect"]["components"]\
117
                ["flux"]["parameters"]["CH4"]["hcorrelations"]["sigma"]
Antoine Berchet's avatar
Antoine Berchet committed
118
119
120
121
122
    correlations = "lowcorr" if sigma == 500 else "highcorr"
    test_id = (settings["mode"], settings.get("minimizer", ""),
               resolution, correlations)

    pytest.test_ids[test_id] = (tmpdir, nsimmax)
123
124
125
    
    # Removing MCF to make computation quicker
    del config["model"]["chemistry"]["acspecies"]["MCF"]
126
    del config["datavect"]["components"]["flux"]["parameters"]["MCF"]
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
    del config["datavect"]["components"]["concs"]["parameters"]["MCF"]
    
    # Dump yml config file
    dummy_config_file = os.path.join(tmpdir, "dummy_config.yml")
    with open(dummy_config_file, "w") as outfile:
        ordered_dump(outfile, config)
    
    # Run as job
    setup = Setup.yaml_to_setup(dummy_config_file)
    setup.todo_init = ["platform"]
    setup = Setup.load_config(setup)
    platform = setup.platform

    job_file = os.path.join(tmpdir,
                            "job_pycif_{}".format(tmpdir[-5:]))

    exe = "{} -m pycif {}".format(platform.python, dummy_config_file)

    job_id = platform.submit_job(
        exe,
        job_file
    )
Antoine Berchet's avatar
Antoine Berchet committed
149
    
150
151
152
    # Wait for the end of the execution
    while not platform.check_jobs([job_id]):
        time.sleep(platform.sleep_time)
Antoine Berchet's avatar
Antoine Berchet committed
153
    
154
155
156
157
158
159
160
161
162
    # Get results and check that posterior closer to prior
    components = os.listdir(prior_dir)
    for comp in components:
        comp_dir_prior = os.path.join(prior_dir, comp)
        comp_dir_post = os.path.join(posterior_dir, comp)
        parameters = os.listdir(comp_dir_prior)
        for param in parameters:
            param_dir = os.path.join(comp_dir_prior, param)
            monitor_prior = \
Antoine Berchet's avatar
Antoine Berchet committed
163
                read_datastore(os.path.join(param_dir, "monitor.nc"))["maindata"]
164
165
166

            param_dir = os.path.join(comp_dir_post, param)
            monitor_post = \
Antoine Berchet's avatar
Antoine Berchet committed
167
                read_datastore(os.path.join(param_dir, "monitor.nc"))["maindata"]
168
169
170
171
172
173
174
175
176
177
178
179
180
            assert (
                    (monitor_prior.loc[:, "obs"]
                     - monitor_prior.loc[:, "sim"]).pow(2).sum()
                    - (monitor_post.loc[:, "obs"]
                       - monitor_post.loc[:, "sim"]).pow(2).sum()
                    > 0
            )

    # Dump configuration into CIF examples
    tag += "_{}_{}".format(settings["mode"],
                           "" if settings["mode"] != "4dvar"
                           else settings["minimizer"])

Antoine Berchet's avatar
Antoine Berchet committed
181
    current_dir = os.path.abspath(os.path.dirname(os.path.realpath(__file__)))
182
183
    root_dir = os.path.abspath(os.path.join(current_dir, "../../"))
    pytest_dir = os.path.abspath(tmpdir + "/../")
Antoine Berchet's avatar
Antoine Berchet committed
184
    example_dir = \
185
        os.path.abspath(os.path.join(root_dir, "examples_artifact/dummy/"))
Antoine Berchet's avatar
Antoine Berchet committed
186
    Path(example_dir).mkdir(parents=True, exist_ok=True)
187
188
    
    config["workdir"] = "{}/inversion_{}/".format(pytest_dir, tag)
189
190

    dummy_config_file = \
Antoine Berchet's avatar
Antoine Berchet committed
191
        os.path.join(example_dir, "config_inversion_{}.yml".format(tag))
192
    with open(dummy_config_file, "w") as outfile:
193
        ordered_dump(outfile, config,
194
195
                     ref_directories={"outdir": pytest_dir,
                                      "rootdir": root_dir},
196
                     replace_values={"rootdir": "/tmp/CIF/"})
197
    
Antoine Berchet's avatar
Antoine Berchet committed
198
199
200
201
202
203

    # Stop here if not full article computation
    marker = pytestconfig.getoption('-m')
    if "allsimulations" not in marker:
        return
    
Antoine Berchet's avatar
Antoine Berchet committed
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
    # Loop with different number of simulations for non analytical inversions
    title = ""
    obs_root = "obsvect_posterior"
    control_root = "controlvect"
    if settings["mode"] != "analytical":
        for nsim in range(1, nsimmax + 1, 2):
            file_config = "{}/dummy_config.yml".format(tmpdir)
            inv_setup = Setup.from_yaml(file_config)
            
            if settings.get("minimizer", None) == "congrad":
                title = "CONGRAD"
                obs_root = "obsoperator/fwd_-002/obsvect"
                control_root = "obsoperator/fwd_-002/controlvect"
                inv_setup["mode"]["minimizer"]["maxiter"] = \
                    max(1, int(nsim))
Antoine Berchet's avatar
Antoine Berchet committed
219
                nsim *= 2
Antoine Berchet's avatar
Antoine Berchet committed
220
221
222
223
224
225
226
227
228
229
                if os.path.isdir("{}/obsoperator/fwd_-002".format(tmpdir)):
                    shutil.rmtree("{}/obsoperator/fwd_-002".format(tmpdir))
            elif settings.get("minimizer", None) == "M1QN3":
                title = "M1QN3"
                obs_root = "obsoperator/fwd_-002/obsvect"
                control_root = "obsoperator/fwd_-002/controlvect"
                inv_setup["mode"]["minimizer"]["maxiter"] = \
                    max(1, int(nsim))
                inv_setup["mode"]["minimizer"]["nsim"] = \
                    max(1, int(nsim))
Antoine Berchet's avatar
Antoine Berchet committed
230
                nsim *= 2
Antoine Berchet's avatar
Antoine Berchet committed
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
                if os.path.isdir("{}/obsoperator/fwd_-002".format(tmpdir)):
                    shutil.rmtree("{}/obsoperator/fwd_-002".format(tmpdir))
            elif settings["mode"] == "ensrf":
                title = "EnSRF"
                inv_setup["mode"]["nsample"] = nsim
            
            # Execute a partial inversion with less simulations
            inv_setup = Setup.from_dict(inv_setup, convert_none=True)
            inv_setup = inv_setup.load_config(inv_setup)
            controlvect, obsvect = inv_setup.mode.execute()
            
            # Compute the cost function
            departures = obsvect.ysim - obsvect.yobs
            j_o = 0.5 * (departures * obsvect.rinvprod(departures)).sum()
            
            dx = controlvect.x - controlvect.xb
            bfull = np.linalg.inv(controlvect.build_b(inv_setup.controlvect))
            j_b = dx[np.newaxis, :].dot(bfull.dot(dx[:, np.newaxis])).sum() / 2
            
            # Dump to text file for later plot
            with open("{}/varying_cost_function.txt".format(tmpdir), "a") as f:
                f.write("{} {} {}\n".format(nsim, j_o, j_b))
    
    else:
255
256
        title = "Analytical"
        
Antoine Berchet's avatar
Antoine Berchet committed
257
258
259
        # Load results
        file_config = "{}/dummy_config.yml".format(tmpdir)
        inv_setup = Setup.from_yaml(file_config)
260
261
        inv_setup = Setup.from_dict(inv_setup, convert_none=True)
        inv_setup = inv_setup.load_config(inv_setup)
262
263
264
265
266
267
268
269
270
271
272
        controlvect, obsvect = inv_setup.mode.execute()
        
        # Compute the cost function
        departures = obsvect.ysim - obsvect.yobs
        j_o = 0.5 * (departures * obsvect.rinvprod(departures)).sum()
        
        dx = controlvect.x - controlvect.xb
        bfull = np.linalg.inv(
            inv_setup.controlvect
                .build_b(inv_setup.controlvect)
        )
Antoine Berchet's avatar
Antoine Berchet committed
273
        j_b = dx[np.newaxis, :].dot(bfull.dot(dx[:, np.newaxis])).sum() / 2
274
275
276
        
        # Dump to text file for later plot
        with open("{}/varying_cost_function.txt".format(tmpdir), "a") as f:
Antoine Berchet's avatar
Antoine Berchet committed
277
278
            f.write("{} {} {}\n".format(0, j_o, j_b))
            f.write("{} {} {}\n".format(nsimmax, j_o, j_b))
279
    
Antoine Berchet's avatar
Antoine Berchet committed
280
281
282
283
284
285
286
    marker = pytestconfig.getoption('-m')
    if "article" in marker and "not article" not in marker:
        # Domain limits
        xmin = inv_setup.domain.xmin
        xmax = inv_setup.domain.xmax
        ymin = inv_setup.domain.ymin
        ymax = inv_setup.domain.ymax
287
    
Antoine Berchet's avatar
Antoine Berchet committed
288
289
        # Read observations
        file_obs = "{}/obsvect/concs/CH4/monitor.nc".format(tmpdir)
290
        monitor_ref = read_datastore(file_obs)["metadata"]
Antoine Berchet's avatar
Antoine Berchet committed
291
292
293
        coords = monitor_ref.loc[:, ["lon", "lat", "alt"]].drop_duplicates()
        
        # Compute fluxes from control vector
294
295
        file_flx = "{}/{}/flux/" \
                   "controlvect_flux_CH4.nc".format(tmpdir, control_root)
Antoine Berchet's avatar
Antoine Berchet committed
296
297
        ds = xr.open_dataset(file_flx)
        dflx = ds["x_phys"].mean(axis=(0, 1)) - ds["xb_phys"].mean(axis=(0, 1))
298
        dx = ds["x"].mean(axis=(0, 1)) - ds["xb"].mean(axis=(0, 1))
Antoine Berchet's avatar
Antoine Berchet committed
299
300
301
        
        # Fetch resolution for figure name
        resol = ""
302
        if config["datavect"]["components"][
303
                "flux"]["parameters"]["CH4"]["hresol"] == "hpixels":
Antoine Berchet's avatar
Antoine Berchet committed
304
            if config["datavect"]["components"][
305
                    "flux"]["parameters"]["CH4"][
Antoine Berchet's avatar
Antoine Berchet committed
306
                    "hcorrelations"]["sigma"] == 500:
Antoine Berchet's avatar
Antoine Berchet committed
307
308
309
                resol = "lowcorr"
            else:
                resol = "highcorr"
310
    
311
        elif config["datavect"]["components"]["flux"] \
Antoine Berchet's avatar
Antoine Berchet committed
312
313
                ["parameters"]["CH4"]["hresol"] == "ibands":
            resol = "bands"
314
    
315
        elif config["datavect"]["components"]["flux"] \
Antoine Berchet's avatar
Antoine Berchet committed
316
317
318
                ["parameters"]["CH4"]["hresol"] == "global":
            resol = "global"
        
319
        # Plot the figure of physical fluxes increments
Antoine Berchet's avatar
Antoine Berchet committed
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
        plt.figure(figsize=(21, 11))
        
        ax0 = plt.axes([0.05, 0.05, 0.73, 0.87])
        im = plt.imshow(dflx, extent=(xmin, xmax, ymin, ymax), cmap="YlOrRd",
                        vmin=-0.2, vmax=0.5)
        sc = plt.scatter(coords["lon"], coords["lat"], c=coords["alt"],
                         cmap="Blues", linewidths=1, edgecolors="k", s=600)
        plt.yticks(fontsize=25)
        plt.xticks(fontsize=25)
        
        ax1 = plt.axes([0.74, 0.05, 0.05, 0.87])
        cb1 = plt.colorbar(im, cax=ax1)
        plt.yticks(fontsize=25)
        plt.ylabel("Fluxes (a.u.)", fontsize=30)
        
        ax2 = plt.axes([0.86, 0.05, 0.05, 0.87])
        cb2 = plt.colorbar(sc, cax=ax2)
        plt.yticks(fontsize=25)
        plt.ylabel("Station altitude (m a.g.l)", fontsize=30)
        
        ax0.set_title(title, fontsize=45)
Antoine Berchet's avatar
Antoine Berchet committed
341
342
343
344
345
346
347
        
        current_dir = os.path.abspath(
            os.path.dirname(os.path.realpath(__file__)))
        figure_dir = \
            os.path.abspath(os.path.join(current_dir,
                                         "../../figures_artifact/"))
        Path(figure_dir).mkdir(parents=True, exist_ok=True)
348
349
        plt.savefig("{}/map_dx_{}_{}_{}.pdf".format(
            figure_dir, title, resol, nsimmax))
Antoine Berchet's avatar
Antoine Berchet committed
350
        plt.close()
351
        
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
        # Plot the figure of scalar fluxes increments
        plt.figure(figsize=(21, 11))
        
        ax0 = plt.axes([0.05, 0.05, 0.73, 0.87])
        im = plt.imshow(dx, extent=(xmin, xmax, ymin, ymax), cmap="YlOrRd")
        sc = plt.scatter(coords["lon"], coords["lat"], c=coords["alt"],
                         cmap="Blues", linewidths=1, edgecolors="k", s=600)
        plt.yticks(fontsize=25)
        plt.xticks(fontsize=25)
        
        ax1 = plt.axes([0.74, 0.05, 0.05, 0.87])
        cb1 = plt.colorbar(im, cax=ax1)
        plt.yticks(fontsize=25)
        plt.ylabel("Fluxes (a.u.)", fontsize=30)
        
        ax2 = plt.axes([0.86, 0.05, 0.05, 0.87])
        cb2 = plt.colorbar(sc, cax=ax2)
        plt.yticks(fontsize=25)
        plt.ylabel("Station altitude (m a.g.l)", fontsize=30)
        
        ax0.set_title(title, fontsize=45)
        
        current_dir = os.path.abspath(
            os.path.dirname(os.path.realpath(__file__)))
        figure_dir = \
            os.path.abspath(os.path.join(current_dir,
                                         "../../figures_artifact/"))
        Path(figure_dir).mkdir(parents=True, exist_ok=True)
        plt.savefig("{}/map_dx_scale_{}_{}_{}.pdf".format(
            figure_dir, title, resol, nsimmax))
        plt.close()
        
384
385
386
387
388
389
390
391
392
393
        # Plot uncertainty reduction
        if "pa_std" in ds:
            dstd = ds["b_std"].mean(axis=(0, 1)) \
                   - ds["pa_std"].mean(axis=(0, 1))
        
            # Plot the figure
            plt.figure(figsize=(21, 11))
            
            ax0 = plt.axes([0.05, 0.05, 0.73, 0.87])
            im = plt.imshow(dstd, extent=(xmin, xmax, ymin, ymax),
Antoine Berchet's avatar
Antoine Berchet committed
394
                            cmap="YlOrRd", vmin=0, vmax=1)
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
            sc = plt.scatter(coords["lon"], coords["lat"], c=coords["alt"],
                             cmap="Blues", linewidths=1, edgecolors="k", s=600)
            plt.yticks(fontsize=25)
            plt.xticks(fontsize=25)
            
            ax1 = plt.axes([0.74, 0.05, 0.05, 0.87])
            cb1 = plt.colorbar(im, cax=ax1)
            plt.yticks(fontsize=25)
            plt.ylabel("Uncertainty reduction (a.u.)", fontsize=30)
            
            ax2 = plt.axes([0.86, 0.05, 0.05, 0.87])
            cb2 = plt.colorbar(sc, cax=ax2)
            plt.yticks(fontsize=25)
            plt.ylabel("Station altitude (m a.g.l)", fontsize=30)
            
            ax0.set_title(title, fontsize=45)
            
            current_dir = os.path.abspath(
                os.path.dirname(os.path.realpath(__file__)))
            figure_dir = \
                os.path.abspath(os.path.join(current_dir,
                                             "../../figures_artifact/"))
            Path(figure_dir).mkdir(parents=True, exist_ok=True)
418
419
            plt.savefig("{}/map_dstd_{}_{}_{}.pdf"
                        .format(figure_dir, title, resol, nsimmax))
420
421
422
423
424
425
            plt.close()
        
        # Plot matrix of uncertainty reduction
        if hasattr(controlvect, "pa"):
            pa = controlvect.pa
            
Antoine Berchet's avatar
Antoine Berchet committed
426
427
428
429
            plt.figure(figsize=(21, 11))
            ax0 = plt.axes([0.05, 0.05, 0.73, 0.87])
            im = plt.imshow(pa[:int(controlvect.dim / 2),
                               :int(controlvect.dim / 2)],
430
                            vmin=-0.5, vmax=0.5, cmap="RdBu")
Antoine Berchet's avatar
Antoine Berchet committed
431
432
433
434
435
436
            plt.xticks(fontsize=20)
            plt.yticks(fontsize=20)
            
            ax1 = plt.axes([0.74, 0.05, 0.05, 0.87])
            cb1 = plt.colorbar(im, cax=ax1)
            plt.yticks(fontsize=25)
437
438
439
            plt.ylabel("Posterior uncertainties", fontsize=30)

            ax0.set_title(title, fontsize=45)
440
441
442
443
444
445
446
            
            current_dir = os.path.abspath(
                os.path.dirname(os.path.realpath(__file__)))
            figure_dir = \
                os.path.abspath(os.path.join(current_dir,
                                             "../../figures_artifact/"))
            Path(figure_dir).mkdir(parents=True, exist_ok=True)
447
            plt.savefig("{}/posterior_matrix_{}_{}_{}.pdf"
Antoine Berchet's avatar
Antoine Berchet committed
448
449
                        .format(figure_dir, title, resol, nsimmax),
                        bbox_inches='tight')
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
            plt.close()
            
            # Prior matrix
            bfull = controlvect.build_b(controlvect)
            plt.figure(figsize=(21, 11))
            ax0 = plt.axes([0.05, 0.05, 0.73, 0.87])
            im = plt.imshow(bfull[:int(controlvect.dim / 2),
                                  :int(controlvect.dim / 2)],
                            vmin=-0.5, vmax=0.5, cmap="RdBu")
            plt.xticks(fontsize=20)
            plt.yticks(fontsize=20)

            ax1 = plt.axes([0.74, 0.05, 0.05, 0.87])
            cb1 = plt.colorbar(im, cax=ax1)
            plt.yticks(fontsize=25)
465
            plt.ylabel("Prior uncertainties", fontsize=30)
466
467
468
469
470
471
472
473

            current_dir = os.path.abspath(
                os.path.dirname(os.path.realpath(__file__)))
            figure_dir = \
                os.path.abspath(os.path.join(current_dir,
                                             "../../figures_artifact/"))
            Path(figure_dir).mkdir(parents=True, exist_ok=True)
            plt.savefig("{}/prior_matrix_{}_{}_{}.pdf"
Antoine Berchet's avatar
Antoine Berchet committed
474
475
                        .format(figure_dir, title, resol, nsimmax),
                        bbox_inches='tight')
476
            plt.close()