Browse Source

1.4.4, includes simple DOI estimate and logging of inversion results into CSV file

tags/1.6.1
Trevor Irons 5 years ago
parent
commit
aa4b3f4ef3
4 changed files with 83 additions and 5 deletions
  1. 75
    2
      akvo/tressel/invertTA.py
  2. 6
    1
      akvo/tressel/logbarrier.py
  3. 1
    1
      akvo/tressel/mrsurvey.py
  4. 1
    1
      setup.py

+ 75
- 2
akvo/tressel/invertTA.py View File

14
 from matplotlib.ticker import AutoMinorLocator
14
 from matplotlib.ticker import AutoMinorLocator
15
 from matplotlib.ticker import LogLocator
15
 from matplotlib.ticker import LogLocator
16
 from matplotlib.ticker import FormatStrFormatter
16
 from matplotlib.ticker import FormatStrFormatter
17
+from matplotlib.colors import Normalize
18
+
17
 import cmocean
19
 import cmocean
18
 from akvo.tressel.lemma_yaml import * 
20
 from akvo.tressel.lemma_yaml import * 
19
 
21
 
22
+import pandas as pd
23
+
20
 def buildKQT(K0,tg,T2Bins):
24
 def buildKQT(K0,tg,T2Bins):
21
     """ 
25
     """ 
22
         Constructs a QT inversion kernel from an initial amplitude one.  
26
         Constructs a QT inversion kernel from an initial amplitude one.  
129
     K0 = K0[0]
133
     K0 = K0[0]
130
     #plt.matshow(K0)
134
     #plt.matshow(K0)
131
     
135
     
136
+    # VERY Simple DOI
137
+    SNR = np.sum(.01*K0, axis=1) / VS[0][0]
138
+    SNR[SNR>1] = 1
139
+    SNRidx = 0 
140
+    while SNR[SNRidx] >= 1:
141
+        SNRidx += 1
142
+    #print(SNR)
143
+    #plt.plot(ifaces[0:-1], SNR)
144
+    #plt.gca().axhline(y=VS[0][0], xmin=0, xmax=ifaces[-1], color='r')
145
+    #plt.gca().axhline(y=1, xmin=0, xmax=ifaces[-1], color='r')
146
+    #plt.show()
147
+
132
     ###############################################
148
     ###############################################
133
     # Build full kernel
149
     # Build full kernel
134
     ############################################### 
150
     ############################################### 
135
     T2Bins = np.logspace( np.log10(cont["T2Bins"]["low"]), np.log10(cont["T2Bins"]["high"]), cont["T2Bins"]["number"], endpoint=True, base=10)  
151
     T2Bins = np.logspace( np.log10(cont["T2Bins"]["low"]), np.log10(cont["T2Bins"]["high"]), cont["T2Bins"]["number"], endpoint=True, base=10)  
136
     KQT = buildKQT(K0,tg,T2Bins)
152
     KQT = buildKQT(K0,tg,T2Bins)
137
-
138
  
153
  
139
     ###############################################
154
     ###############################################
140
     # Invert
155
     # Invert
145
 
160
 
146
     ###############################################
161
     ###############################################
147
     # Appraise
162
     # Appraise
148
-    ############################################### 
163
+    ###############################################
164
+ 
149
     pre = np.dot(KQT,inv) 
165
     pre = np.dot(KQT,inv) 
150
     PRE = np.reshape( pre, np.shape(V)  )
166
     PRE = np.reshape( pre, np.shape(V)  )
151
     plt.matshow(PRE, cmap='Blues')
167
     plt.matshow(PRE, cmap='Blues')
165
     T2Bins = np.append( T2Bins, T2Bins[-1] + (T2Bins[-1]-T2Bins[-2]) )
181
     T2Bins = np.append( T2Bins, T2Bins[-1] + (T2Bins[-1]-T2Bins[-2]) )
166
     
182
     
167
     INV = np.reshape(inv, (len(ifaces)-1,cont["T2Bins"]["number"]) )
183
     INV = np.reshape(inv, (len(ifaces)-1,cont["T2Bins"]["number"]) )
184
+
185
+    #alphas = np.tile(SNR, (len(T2Bins)-1,1))
186
+    #colors = Normalize(1e-6, np.max(INV.T), clip=True)(INV.T)
187
+    #colors = cmocean.cm.tempo(colors)
188
+    ##colors[..., -1] = alphas
189
+    #print(np.shape(colors)) 
190
+    #print(np.shape(INV.T)) 
191
+
192
+    #greys = np.full((*(INV.T).shape, 3), 70, dtype=np.uint8)
193
+
168
     Y,X = meshgrid( ifaces, T2Bins )
194
     Y,X = meshgrid( ifaces, T2Bins )
169
     fig = plt.figure( figsize=(pc2in(20.0),pc2in(22.)) )
195
     fig = plt.figure( figsize=(pc2in(20.0),pc2in(22.)) )
170
     ax1 = fig.add_axes( [.2,.15,.6,.7] )
196
     ax1 = fig.add_axes( [.2,.15,.6,.7] )
171
     im = ax1.pcolor(X, Y, INV.T, cmap=cmocean.cm.tempo) #cmap='viridis')
197
     im = ax1.pcolor(X, Y, INV.T, cmap=cmocean.cm.tempo) #cmap='viridis')
198
+    #im = ax1.pcolor(X[0:SNRidx,:], Y[0:SNRidx,:], INV.T[0:SNRidx,:], cmap=cmocean.cm.tempo) #cmap='viridis')
199
+    #im = ax1.pcolor(X[SNRidx::,:], Y[SNRidx::,:], INV.T[SNRidx::,:], cmap=cmocean.cm.tempo, alpha=.5) #cmap='viridis')
200
+    #im = ax1.pcolormesh(X, Y, INV.T, alpha=alphas) #, cmap=cmocean.cm.tempo) #cmap='viridis')
201
+    #im = ax1.pcolormesh(X, Y, INV.T, alpha=alphas) #, cmap=cmocean.cm.tempo) #cmap='viridis')
202
+    #ax1.axhline( y=ifaces[SNRidx], xmin=T2Bins[0], xmax=T2Bins[-1], color='black'  )
172
     im.set_edgecolor('face')
203
     im.set_edgecolor('face')
173
     ax1.set_xlim( T2Bins[0], T2Bins[-1] )
204
     ax1.set_xlim( T2Bins[0], T2Bins[-1] )
174
     ax1.set_ylim( ifaces[-1], ifaces[0] )
205
     ax1.set_ylim( ifaces[-1], ifaces[0] )
192
     ax2.set_ylim( ifaces[-1], ifaces[0] )
223
     ax2.set_ylim( ifaces[-1], ifaces[0] )
193
     ax2.xaxis.set_major_locator( MaxNLocator(nbins = 3) )   
224
     ax2.xaxis.set_major_locator( MaxNLocator(nbins = 3) )   
194
     ax2.get_xaxis().set_major_formatter(FormatStrFormatter('%0.2f'))
225
     ax2.get_xaxis().set_major_formatter(FormatStrFormatter('%0.2f'))
226
+    ax2.axhline( y=ifaces[SNRidx], xmin=0, xmax=1, color='black', linestyle='dashed'  )
195
     #ax2.xaxis.set_label_position('bottom') 
227
     #ax2.xaxis.set_label_position('bottom') 
196
 
228
 
197
     plt.savefig("akvoInversion.pdf")
229
     plt.savefig("akvoInversion.pdf")
230
+
231
+    #############
232
+    # water plot#
233
+
234
+    fig2 = plt.figure( figsize=(pc2in(20.0),pc2in(22.)) )
235
+    ax = fig2.add_axes( [.2,.15,.6,.7] )
198
     
236
     
237
+    # Bound water cutoff 
238
+    Bidx = T2Bins[0:-1]<33.0
239
+    twater = np.sum(INV, axis=1)
240
+    bwater = np.sum(INV[:,Bidx], axis=1)
241
+    
242
+    ax.plot( twater, (ifaces[0:-1]+ifaces[1::])/2, label="NMR total water", color='blue' )
243
+    ax.plot( bwater, (ifaces[0:-1]+ifaces[1::])/2, label="NMR bound water", color='green' )
244
+    
245
+    ax.fill_betweenx((ifaces[0:-1]+ifaces[1::])/2 , twater, bwater, where=twater >= bwater, facecolor='blue', alpha=.5)
246
+    ax.fill_betweenx((ifaces[0:-1]+ifaces[1::])/2 , bwater, 0, where=bwater >= 0, facecolor='green', alpha=.5)
247
+    
248
+    ax.set_xlabel(r"$\theta_N$ (m$^3$/m$^3$)")
249
+    ax.set_ylabel(r"depth (m)")
250
+    
251
+    ax.set_ylim( ifaces[-1], ifaces[0] )
252
+    ax.set_xlim( 0, ax.get_xlim()[1] )
253
+    
254
+    ax.axhline( y=ifaces[SNRidx], xmin=0, xmax=1, color='black', linestyle='dashed'  )
255
+    
256
+    plt.savefig("akvoInversionWC.pdf")
257
+    plt.legend()  
258
+ 
259
+
260
+    # Report results into a text file 
261
+    fr = pd.DataFrame( INV, columns=T2Bins[0:-1] )
262
+    fr.insert(0, "layer top", ifaces[0:-1] )
263
+    fr.insert(1, "layer bottom", ifaces[1::] )
264
+    fr.insert(2, "NMR total water", np.sum(INV, axis=1) )
265
+    fr.insert(3, "NMR bound water", bwater )
266
+    fr.insert(4, "Layer SNR", SNR )
267
+
268
+    fr.to_csv("akvoInversion.csv")    
269
+    #fr.to_excel("akvoInversion.xlsx")    
270
+
271
+ 
199
     plt.show()
272
     plt.show()
200
 
273
 
201
 if __name__ == "__main__":
274
 if __name__ == "__main__":

+ 6
- 1
akvo/tressel/logbarrier.py View File

146
 
146
 
147
         xp = np.copy(x) # prior step x 
147
         xp = np.copy(x) # prior step x 
148
 
148
 
149
+        # quick and dirty solution
150
+        #b2a = np.dot(A.conj().transpose(), np.dot(WdTWd, b-b_pre) ) - alpha*np.dot(WmTWm,(x-xr))
151
+        #xg = nnls(ATWdTWdA + Phi_m, b2a)
152
+        #x = xg[0]
153
+
149
         while ( (phib / (phid+alpha*phim)) > EPSILON  or First==True ):
154
         while ( (phib / (phid+alpha*phim)) > EPSILON  or First==True ):
150
-        #while ( First==True ):
155
+        #while ( False ): # skip the hard stuff
151
 
156
 
152
             First = False
157
             First = False
153
             # Log barrier, keep each element above minVal
158
             # Log barrier, keep each element above minVal

+ 1
- 1
akvo/tressel/mrsurvey.py View File

1046
                     ht = signal.hilbert(xn)*np.exp(-1j*wL*self.DATADICT[pulse]["TIMES"])
1046
                     ht = signal.hilbert(xn)*np.exp(-1j*wL*self.DATADICT[pulse]["TIMES"])
1047
                     #############################################################
1047
                     #############################################################
1048
                     # Quadrature signal 
1048
                     # Quadrature signal 
1049
-                    RE[pulse][chan][ipm,:] = -np.real(ht[clip::])  # negative for consistency with VC  
1049
+                    RE[pulse][chan][ipm,:] =  np.real(ht[clip::])  # *-1 for negative for consistency with VC ??
1050
                     IM[pulse][chan][ipm,:] =  np.imag(ht[clip::])
1050
                     IM[pulse][chan][ipm,:] =  np.imag(ht[clip::])
1051
                     REmax[pulse] = max(REmax[pulse], np.max(np.real(ht[clip::])))
1051
                     REmax[pulse] = max(REmax[pulse], np.max(np.real(ht[clip::])))
1052
                     IMmax[pulse] = max(IMmax[pulse], np.max(np.imag(ht[clip::])))
1052
                     IMmax[pulse] = max(IMmax[pulse], np.max(np.imag(ht[clip::])))

+ 1
- 1
setup.py View File

21
     long_description = fh.read()
21
     long_description = fh.read()
22
 
22
 
23
 setup(name='Akvo',
23
 setup(name='Akvo',
24
-      version='1.4.3',
24
+      version='1.4.4',
25
       python_requires='>3.7.0', # due to pyLemma
25
       python_requires='>3.7.0', # due to pyLemma
26
       description='Surface nuclear magnetic resonance workbench',
26
       description='Surface nuclear magnetic resonance workbench',
27
       long_description=long_description,
27
       long_description=long_description,

Loading…
Cancel
Save