test_integration_3inversions.py 11.3 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
7
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", [
        {"mode": "4dvar", "minimizer": "M1QN3"},
Antoine Berchet's avatar
Antoine Berchet committed
21
22
23
        {"mode": "4dvar", "minimizer": "congrad"},
        {"mode": "analytical"},
        {"mode": "ensrf"}
24
    ]
25
)
Antoine Berchet's avatar
Antoine Berchet committed
26
def test_integration_inversion(dummy_config_inversion, settings, pytestconfig):
27
28
29
30
31
32
    """
    Integration test that runs the dummy_forward model.
    """

    tmpdir, config, tag = dummy_config_inversion
    
Antoine Berchet's avatar
Antoine Berchet committed
33
34
35
36
    # Force reloading observation operator
    # from forward to make computation faster
    config["model"]["reload_H"] = "{}/../H_matrix.pickle".format(tmpdir)
    
37
    # Changing mode
Antoine Berchet's avatar
Antoine Berchet committed
38
    nsimmax = 10
39
40
    if config["datavect"]["components"]["fluxes"]\
            ["parameters"]["CH4"]["hresol"] == "hpixels":
Antoine Berchet's avatar
Antoine Berchet committed
41
        nsimmax = 25
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
    
    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
60
                "df1": 0.00001
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
            },
            "save_out_netcdf": True
        }
        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
81
    config["platform"] = {"plugin": {"name": "docker", "version": "cif"}}
Antoine Berchet's avatar
Antoine Berchet committed
82
83
84
85
86
87
88
89
90
91
92
93

    # 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)
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
    
    # 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
120
    
121
122
123
    # 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
124
    
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
    # 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"])

    current_dir = os.path.abspath(os.path.dirname(os.path.realpath(__file__)))
    example_dir = \
        os.path.abspath(os.path.join(current_dir, "../../examples/dummy/"))

    dummy_config_file = \
Antoine Berchet's avatar
Antoine Berchet committed
157
        os.path.join(example_dir, "config_{}.yml".format(tag))
158
159
160
    with open(dummy_config_file, "w") as outfile:
        ordered_dump(outfile, config)
    
Antoine Berchet's avatar
Antoine Berchet committed
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
    # 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))
                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))
                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:
210
211
        title = "Analytical"
        
Antoine Berchet's avatar
Antoine Berchet committed
212
213
214
        # Load results
        file_config = "{}/dummy_config.yml".format(tmpdir)
        inv_setup = Setup.from_yaml(file_config)
215
216
        inv_setup = Setup.from_dict(inv_setup, convert_none=True)
        inv_setup = inv_setup.load_config(inv_setup)
217
218
219
220
221
222
223
224
225
226
227
        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
228
        j_b = dx[np.newaxis, :].dot(bfull.dot(dx[:, np.newaxis])).sum() / 2
229
230
231
        
        # 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
232
233
            f.write("{} {} {}\n".format(0, j_o, j_b))
            f.write("{} {} {}\n".format(nsimmax, j_o, j_b))
234
    
Antoine Berchet's avatar
Antoine Berchet committed
235
236
237
238
239
240
241
    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
242
    
Antoine Berchet's avatar
Antoine Berchet committed
243
244
245
246
247
248
249
250
251
252
253
254
255
        # 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 = ""
256
        if config["datavect"]["components"][
Antoine Berchet's avatar
Antoine Berchet committed
257
258
259
260
261
262
                "fluxes"]["parameters"]["CH4"]["hresol"] == "hpixels":
            if config["datavect"]["components"][
                    "fluxes"]["parameters"]["CH4"]["hcorrelations"]["sigma"] == 500:
                resol = "lowcorr"
            else:
                resol = "highcorr"
263
    
Antoine Berchet's avatar
Antoine Berchet committed
264
265
266
        elif config["datavect"]["components"]["fluxes"] \
                ["parameters"]["CH4"]["hresol"] == "ibands":
            resol = "bands"
267
    
Antoine Berchet's avatar
Antoine Berchet committed
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
        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)
        plt.savefig("{}/../map_dx_{}_{}.pdf".format(tmpdir, title, resol))
        plt.close()