test_integration_3inversions.py 15 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", [
21
22
23
        {"mode": "4dvar", "minimizer": "M1QN3"},
        pytest.param({"mode": "4dvar", "minimizer": "M1QN3", "montecarlo": 10},
                     marks=pytest.mark.uncertainties),
Antoine Berchet's avatar
Antoine Berchet committed
24
        {"mode": "4dvar", "minimizer": "congrad"},
25
26
27
28
        {"mode": "ensrf"},
        pytest.param({"mode": "ensrf", "nsimmax": 250},
                     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)
    
42
    # Changing mode
43
    nsimmax = settings.get("nsimmax", 10)
44
45
    if config["datavect"]["components"]["fluxes"]\
            ["parameters"]["CH4"]["hresol"] == "hpixels":
Antoine Berchet's avatar
Antoine Berchet committed
46
        nsimmax = 25
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
    
    elif config["datavect"]["components"]["fluxes"]\
            ["parameters"]["CH4"]["hresol"] == "global" \
            and settings.get("minimizer", "congrad"):
        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,
                "epsg": 0.02,
Antoine Berchet's avatar
Antoine Berchet committed
65
                "df1": 0.00001
66
67
68
            },
            "save_out_netcdf": True
        }
69
70
71
72
73
74
        if settings["minimizer"] == "congrad":
            mode["minimizer"]["save_uncertainties"] = True
        
        if "montecarlo" in settings:
            mode["montecarlo"] = settings["montecarlo"]
        
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
        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":
        nsimmax *= 2
        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
Antoine Berchet's avatar
Antoine Berchet committed
92
    config["platform"] = {"plugin": {"name": "docker", "version": "cif"}}
Antoine Berchet's avatar
Antoine Berchet committed
93
94
95
96
97
98
99
100
101
102
103
104

    # Save ID for later plot of cost function
    resolution = \
        config["datavect"]["components"]["fluxes"]["parameters"]["CH4"][
            "hresol"]
    sigma = config["datavect"]["components"]\
                ["fluxes"]["parameters"]["CH4"]["hcorrelations"]["sigma"]
    correlations = "lowcorr" if sigma == 500 else "highcorr"
    test_id = (settings["mode"], settings.get("minimizer", ""),
               resolution, correlations)

    pytest.test_ids[test_id] = (tmpdir, nsimmax)
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
    
    # Removing MCF to make computation quicker
    del config["model"]["chemistry"]["acspecies"]["MCF"]
    del config["datavect"]["components"]["fluxes"]["parameters"]["MCF"]
    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
131
    
132
133
134
    # 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
135
    
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
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 = \
                read_datastore(os.path.join(param_dir, "monitor.nc"))

            param_dir = os.path.join(comp_dir_post, param)
            monitor_post = \
                read_datastore(os.path.join(param_dir, "monitor.nc"))
            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
163
    current_dir = os.path.abspath(os.path.dirname(os.path.realpath(__file__)))
164
165
    root_dir = os.path.abspath(os.path.join(current_dir, "../../"))
    pytest_dir = os.path.abspath(tmpdir + "/../")
Antoine Berchet's avatar
Antoine Berchet committed
166
    example_dir = \
167
        os.path.abspath(os.path.join(root_dir, "examples_artifact/dummy/"))
Antoine Berchet's avatar
Antoine Berchet committed
168
    Path(example_dir).mkdir(parents=True, exist_ok=True)
169
170
    
    config["workdir"] = "{}/inversion_{}/".format(pytest_dir, tag)
171
172

    dummy_config_file = \
Antoine Berchet's avatar
Antoine Berchet committed
173
        os.path.join(example_dir, "config_inversion_{}.yml".format(tag))
174
    with open(dummy_config_file, "w") as outfile:
175
176
177
178
        ordered_dump(outfile, config,
                     ref_directories={"rootdir": root_dir,
                                      "outdir": pytest_dir},
                     replace_values={"rootdir": "/tmp/CIF/"})
179
    
Antoine Berchet's avatar
Antoine Berchet committed
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
    # 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
195
                nsim *= 2
Antoine Berchet's avatar
Antoine Berchet committed
196
197
198
199
200
201
202
203
204
205
                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
206
                nsim *= 2
Antoine Berchet's avatar
Antoine Berchet committed
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
                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:
231
232
        title = "Analytical"
        
Antoine Berchet's avatar
Antoine Berchet committed
233
234
235
        # Load results
        file_config = "{}/dummy_config.yml".format(tmpdir)
        inv_setup = Setup.from_yaml(file_config)
236
237
        inv_setup = Setup.from_dict(inv_setup, convert_none=True)
        inv_setup = inv_setup.load_config(inv_setup)
238
239
240
241
242
243
244
245
246
247
248
        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
249
        j_b = dx[np.newaxis, :].dot(bfull.dot(dx[:, np.newaxis])).sum() / 2
250
251
252
        
        # 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
253
254
            f.write("{} {} {}\n".format(0, j_o, j_b))
            f.write("{} {} {}\n".format(nsimmax, j_o, j_b))
255
    
Antoine Berchet's avatar
Antoine Berchet committed
256
257
258
259
260
261
262
    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
263
    
Antoine Berchet's avatar
Antoine Berchet committed
264
265
266
267
268
269
270
271
272
273
274
275
276
        # Read observations
        file_obs = "{}/obsvect/concs/CH4/monitor.nc".format(tmpdir)
        monitor_ref = read_datastore(file_obs)
        coords = monitor_ref.loc[:, ["lon", "lat", "alt"]].drop_duplicates()
        
        # Compute fluxes from control vector
        file_flx = "{}/{}/fluxes/" \
                   "controlvect_fluxes_CH4.nc".format(tmpdir, control_root)
        ds = xr.open_dataset(file_flx)
        dflx = ds["x_phys"].mean(axis=(0, 1)) - ds["xb_phys"].mean(axis=(0, 1))
        
        # Fetch resolution for figure name
        resol = ""
277
        if config["datavect"]["components"][
Antoine Berchet's avatar
Antoine Berchet committed
278
279
280
281
282
283
                "fluxes"]["parameters"]["CH4"]["hresol"] == "hpixels":
            if config["datavect"]["components"][
                    "fluxes"]["parameters"]["CH4"]["hcorrelations"]["sigma"] == 500:
                resol = "lowcorr"
            else:
                resol = "highcorr"
284
    
Antoine Berchet's avatar
Antoine Berchet committed
285
286
287
        elif config["datavect"]["components"]["fluxes"] \
                ["parameters"]["CH4"]["hresol"] == "ibands":
            resol = "bands"
288
    
Antoine Berchet's avatar
Antoine Berchet committed
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
        elif config["datavect"]["components"]["fluxes"] \
                ["parameters"]["CH4"]["hresol"] == "global":
            resol = "global"
        
        # Plot the figure
        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
315
316
317
318
319
320
321
        
        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)
322
323
        plt.savefig("{}/map_dx_{}_{}_{}.pdf".format(
            figure_dir, title, resol, nsimmax))
Antoine Berchet's avatar
Antoine Berchet committed
324
        plt.close()
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
        
        # 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),
                            cmap="YlOrRd", vmin=0, 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("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)
360
361
            plt.savefig("{}/map_dstd_{}_{}_{}.pdf"
                        .format(figure_dir, title, resol, nsimmax))
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
            plt.close()
        
        # Plot matrix of uncertainty reduction
        if hasattr(controlvect, "pa"):
            bfull = controlvect.build_b(controlvect)
            pa = controlvect.pa
            
            plt.imshow((bfull - pa)[:int(controlvect.dim / 2),
                                    :int(controlvect.dim / 2)])
            
            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)
378
379
            plt.savefig("{}/uncertaintyreduc_matrix_{}_{}_{}.pdf"
                        .format(figure_dir, title, resol, nsimmax))
380
381
            plt.close()