fig = plt.figure('Worm crawling', figsize=(8, 4.5), dpi=300)
gs1 = gridspec.GridSpec(1, 2, width_ratios=[1, 3])
gs1.update(left=0.075,
right=0.95,
wspace=0.1,
bottom=0.15,
top=0.9,
hspace=0.15)
# recreate worms
#=============================================================================#
# # load eigenworms
#=============================================================================#
eigenwormsFull = dh.loadEigenBasis(filename='utility/Eigenworms.dat',
nComp=5,
new=True)
eigenworms = dh.loadEigenBasis(filename='utility/Eigenworms.dat',
nComp=3,
new=True)
pc1, pc2, pc3, avTrue, thetaTrue = moving['Behavior']['Eigenworm1'],moving['Behavior']['Eigenworm2'],\
moving['Behavior']['Eigenworm3'], moving['Behavior']['AngleVelocity'], moving['Behavior']['Theta']
pcs = np.vstack([pc3, pc2, pc1])
# actual centerline
cl = moving['CL']
#=============================================================================#
# load predicted worm
#=============================================================================#
def main():
show = 1
#=============================================================================#
# # data parameters
#=============================================================================#
dataPars = {'medianWindow':5, # smooth eigenworms with gauss filter of that size, must be odd
'gaussWindow':5, # sgauss window for angle velocity derivative. must be odd
'rotate':True, # rotate Eigenworms using previously calculated rotation matrix
'windowGCamp': 5 # gauss window for red and green channel
}
folder = "AML32_moving/"
dataLog = "AML32_moving/AML32_moving_datasets.txt"
outLoc = "Analysis/AML32_moving_results.hdf5"
#=============================================================================#
# # load eigenworms
#=============================================================================#
eigenworms = dh.loadEigenBasis(filename='utility/Eigenworms.dat', nComp=3, new = True)
#=============================================================================#
# load datasets from hdf5/matlab in a dictionary
#=============================================================================#
dataSets = dh.loadMultipleDatasets(dataLog, pathTemplate=folder, dataPars = dataPars, nDatasets = 2)
keyList = np.sort(dataSets.keys())
# pick one dataset
key = keyList[1]
data = dataSets[key]
pc1, pc2, pc3, avTrue, thetaTrue = data['Behavior']['Eigenworm1'],data['Behavior']['Eigenworm2'],\
data['Behavior']['Eigenworm3'], data['Behavior']['AngleVelocity'], data['Behavior']['Theta']
pcs = np.vstack([pc3,pc2, pc1])
# actual centerline
cl= data['CL']
#=============================================================================#
# for debugging recreate an existing, approximated shape from 3 eigenworms
#=============================================================================#
pcsNew, meanAngle, lengths, refPoint = dh.calculateEigenwormsFromCL(cl, eigenworms)
pc3New, pc2New, pc1New = pcsNew
cl = dh.calculateCLfromEW(pcsNew, eigenworms, meanAngle, lengths, refPoint)
# transform eigenworms exactly the same way. Otherwise we get some artefacts from nans
#_,_,_, av, theta = dh.transformEigenworms(pcsNew, dataPars)
r = (pcsNew[2]**2+pcsNew[1]**2)
#=============================================================================#
# here we reconstruct from the true angular velocity to check the math. This is smoothed, so we need to compare with this version
#=============================================================================#
xt, yt, zt = recrWorm(avTrue, pc3, thetaTrue, r=r)
pcsR = np.vstack([zt,yt, xt])
clApprox = dh.calculateCLfromEW(pcsR, eigenworms, meanAngle, lengths, refPoint)
#=============================================================================#
# load predicted worm
#=============================================================================#
resultDict = dh.loadDictFromHDF(outLoc)
results = resultDict[key]
avP = results['ElasticNet']['AngleVelocity']['output'][:len(pcs[0])]
tP = results['ElasticNet']['Eigenworm3']['output'][:len(pcs[0])]
print 'R2'
print results['ElasticNet']['AngleVelocity']['score'],results['ElasticNet']['AngleVelocity']['scorepredicted']
print results['ElasticNet']['Eigenworm3']['score'],results['ElasticNet']['Eigenworm3']['scorepredicted']
#=============================================================================#
# reconstruct worm from predicted angular velocity.
#=============================================================================#
xP, yP, zP = recrWorm(avP, tP, thetaTrue,r=r)
pcsP = np.vstack([zP,yP, xP])
clPred = dh.calculateCLfromEW(pcsP, eigenworms, meanAngle, lengths, refPoint)
# center around midpoint
originalCMS = np.tile(np.mean(cl, axis=1)[:,np.newaxis,:], (1,99,1))
clApprox -= originalCMS
cl -= originalCMS
clPred -= originalCMS# +(0,100)
if show:
# animate centerlines
fig = plt.figure('prediction')
frames = np.arange(1750,2550)
#frames = np.arange(1750,1760)
f, (ax0, ax1, ax) = plt.subplots(2,1)
ax0.plot(avTrue[frames], label = 'True', color=mp.colorBeh['AngleVelocity'])
ax0.plot(avP[frames], label = 'Predicted', color=mp.colorPred['AngleVelocity'])
plt.xlabel('Frames')
plt.ylabel('Wave velocity')
plt.legend()
ax1.plot(pc3[frames], color=mp.colorBeh['Eigenworm3'], label = 'True')
ax1.plot(tP[frames], color=mp.colorPred['Eigenworm3'], label = 'Predicted')
plt.xlabel('Frames')
plt.ylabel('Turns')
plt.show()
fig = plt.figure('wiggly centerlines')
ax = fig.add_subplot(111, adjustable='box', aspect=0.66)
ax.set_ylim(-400, 400)
ax.set_xlim(-400, 800)
plt.tight_layout()
#frames = results['Training']['AngleVelocity']['Test']
mp.make_animation3(fig, ax, data1= clPred, data2=clApprox +(500,0) , frames = frames, save=True)
#mp.make_animation2(fig, ax, data1= clApprox +(500,0), data2=clApprox +(500,0) , frames = np.arange(120, 1000), color=mp.UCred[0])
plt.show()
# show overlay of eigenworms, correlations etc.
plt.figure('True reconstruction relative to original centerlines')
for i in range(3):
plt.subplot(3,1,i+1)
#plt.plot(pcsNew[i], alpha=0.5, label='centerline projected')
plt.plot(pcs[i], alpha=0.5, label='dataHandler')
plt.plot(pcsR[i], alpha=0.5, label='true reconstructed')
plt.legend()
plt.ylabel('Projection on Eigenworm {}'.format(3-i))
plt.xlabel('Time (Volumes)')
plt.tight_layout()
plt.show()
#example centerlines
i=99
plt.figure()
plt.subplot(111, aspect = 'equal')
plt.plot(clApprox[i,:,0]+30*i,clApprox[i,:,1], '0.5', label = 'Approximated centerline by 4 Eigenworms')
plt.plot(cl[i,:,0]+30*i,cl[i,:,1], 'r', label = 'Real Centerline')
plt.plot(clPred[i,:,0]+30*i,clPred[i,:,1], 'b', label = 'Predicted Centerline')
for i in range(100, 200, 10):
plt.plot(clApprox[i,:,0]+30*i,clApprox[i,:,1], '0.5')
plt.plot(cl[i,:,0]+30*i,cl[i,:,1], 'r')
plt.plot(clPred[i,:,0]+30*i,clPred[i,:,1], 'b')
plt.legend()
plt.tight_layout()
plt.show()
#
cut = 1700
sz = 3
a = 0.5
plt.figure(figsize=(12,7))
plt.subplot(231)
plt.scatter(yt[:cut], xt[:cut], label = 'True reconstructed', alpha=a, s = sz)
plt.scatter(pcs[1][:cut], pcs[2][:cut], label = 'True projected', alpha=a, s = sz)
plt.ylabel('Projection on Eigenworm 1')
plt.xlabel('Projection on Eigenworm 2')
plt.legend()
plt.subplot(234)
plt.scatter(yt[cut:], xt[cut:], label = 'True reconstructed', alpha=a, s = sz)
plt.scatter(pcs[1][cut:], pcs[2][cut:], label = 'True projected', alpha=a, s = sz)
plt.legend()
plt.ylabel('Projection on Eigenworm 1')
plt.xlabel('Projection on Eigenworm 2')
plt.subplot(232)
plt.scatter(xt[:cut], pcs[2][:cut], label = 'First part R^2 = {:.2f}'.format(explained_variance_score(xt[:cut], pcs[2][:cut])), alpha=a, s = sz, color ='C2' )
plt.ylabel('Projection on Eigenworm 1')
plt.xlabel('Reconstruction')
plt.legend()
plt.subplot(235)
plt.scatter(xt[cut:], pcs[2][cut:], label = 'Second part R^2 = {:.2f}'.format(explained_variance_score(xt[cut:], pcs[2][cut:])), alpha=a, s = sz, color ='C2' )
plt.legend()
plt.ylabel('Projection on Eigenworm 1')
plt.xlabel('Reconstruction')
plt.subplot(233)
plt.scatter(yt[:cut], pcs[1][:cut], label = 'First part R^2 = {:.2f}'.format(explained_variance_score(yt[:cut], pcs[1][:cut])), alpha=a, s = sz, color ='C3' )
plt.ylabel('Projection on Eigenworm 2')
plt.xlabel('Reconstruction')
plt.legend()
plt.subplot(236)
plt.scatter(yt[cut:], pcs[1][cut:], label = 'Second part R^2 = {:.2f}'.format(explained_variance_score(yt[cut:], pcs[1][cut:])), alpha=a, s = sz, color ='C3' )
plt.legend()
plt.ylabel('Projection on \n Eigenworm 2')
plt.xlabel('Reconstruction')
plt.tight_layout()
plt.show()
plt.figure('Predicted velocity and turns')
plt.subplot(211)
plt.plot(avTrue, label = 'True')
plt.plot(avP, label = 'Predicted')
plt.legend()
plt.subplot(212)
plt.plot(pc3)
plt.plot(tP)
plt.tight_layout()
plt.show()
# show predicted reconstruction
plt.figure('Predicted Eigenworm reconstruction')
plt.subplot(311)
plt.plot(xP,label = 'Prediction reconstructed', alpha=0.5)
plt.plot(xt, label = 'True projected', alpha=0.5)
plt.ylabel('Projection on Eigenworm 1')
plt.xlabel('Time (Volumes)')
plt.subplot(312)
plt.plot(yP,label = 'Prediction reconstructed', alpha=0.5)
plt.plot(yt, label = 'True projected', alpha=0.5)
plt.ylabel('Projection on Eigenworm 2')
plt.xlabel('Time (Volumes)')
plt.subplot(313)
plt.plot(zP, label = 'Prediction reconstructed', alpha=0.5)
plt.plot(zt, label = 'True projected', alpha=0.5)
plt.ylabel('Projection on Eigenworm 3')
plt.xlabel('Time (Volumes)')
plt.legend()
plt.tight_layout()
#
cut = 1700
sz = 3
a = 0.5
plt.figure(figsize=(12,7))
plt.subplot(231)
plt.scatter(yt[:cut], xt[:cut], label = 'True reconstructed', alpha=a, s = sz)
plt.scatter(yP[:cut], xP[:cut], label = 'Predicted', alpha=a, s = sz)
plt.ylabel('Projection on Eigenworm 1')
plt.xlabel('Projection on Eigenworm 2')
plt.legend()
plt.subplot(234)
plt.scatter(yt[cut:], xt[cut:], label = 'True reconstructed', alpha=a, s = sz)
plt.scatter(yP[cut:],xP[cut:], label = 'Predicted', alpha=a, s = sz)
plt.legend()
plt.ylabel('Projection on Eigenworm 1')
plt.xlabel('Projection on Eigenworm 2')
plt.subplot(232)
plt.scatter(xt[:cut], xP[:cut], label = 'First part R^2 = {:.2f}'.format(explained_variance_score(xt[:cut], xP[:cut])), alpha=a, s = sz, color ='C2' )
plt.legend()
plt.ylabel('Predicted')
plt.xlabel('Reconstruction')
plt.subplot(235)
plt.scatter(xt[cut:], xP[cut:], label = 'Second part R^2 = {:.2f}'.format(explained_variance_score(xt[cut:], xP[cut:])), alpha=a, s = sz, color ='C2' )
plt.legend()
plt.ylabel('Predicted')
plt.xlabel('Reconstruction')
plt.subplot(233)
plt.scatter(yt[:cut], yP[:cut], label = 'First part R^2 = {:.2f}'.format(explained_variance_score(yt[:cut], yP[:cut])), alpha=a, s = sz, color ='C3' )
plt.legend()
plt.ylabel('Predicted')
plt.xlabel('Reconstruction')
plt.legend()
plt.subplot(236)
plt.scatter(yt[cut:], yP[cut:], label = 'Second part R^2 = {:.2f}'.format(explained_variance_score(yt[cut:],yP[cut:])), alpha=a, s = sz, color ='C3' )
plt.legend()
plt.ylabel('Predicted')
plt.xlabel('Reconstruction')
plt.tight_layout()
plt.show()
pca = PCA(n_components=nComp)
pcs = pca.fit_transform(allCenterlines)
newEigenworms = pca.components_
print 'Explained variance with 2 components: ', np.cumsum(
pca.explained_variance_ratio_)[2]
print 'Explained variance with 4 components: ', np.cumsum(
pca.explained_variance_ratio_)[3]
# save data as file
#np.savetxt('Eigenworms.dat', newEigenworms)
# plot old Eigenworms for comparison
oldEigenworms = dh.loadEigenBasis(filename='utility/eigenWorms.mat',
nComp=4,
new=False)
plt.figure()
ax4 = plt.subplot(111)
ax4.fill_between(np.arange(0.5, nComp + 0.5),
pca.explained_variance_ratio_ * 100,
step='post',
color='k',
alpha=0.75)
#ax4.step(np.arange(1,nComp+1),np.cumsum(results['expVariance'])*100, where = 'pre')
ax4.plot(np.arange(1, nComp + 1),
np.cumsum(pca.explained_variance_ratio_) * 100,
'ko-',
lw=1)
ax4.set_ylabel('Explained variance (%)')
dataPars = {'medianWindow':1, # smooth eigenworms with median filter of that size, must be odd
'savGolayWindow':13, # savitzky-golay window for angle velocity derivative. must be odd
'rotate':False, # rotate Eigenworms using previously calculated rotation matrix
'savGolayWindowGCamp': 5 # savitzky-golay window for red and green channel
}
folder = "../BrainScanner20170613_134800_MS/"
ewfile = "eigenWorms.mat"
#RMatrix = np.loadtxt(folder+'../'+'Rotationmatrix.dat')
points = dh.loadPoints(folder)
#print points
cl, wc, ep = dh.loadCenterlines(folder)
#cl = dh.loadCenterlines(folder, wormcentered=0)
data = dh.loadData(folder, dataPars)
eigenworms = dh.loadEigenBasis(ewfile)[:3]
print eigenworms.shape
eigenworms = resize(eigenworms, (3,99))
pc1,pc2,pc3 = [data['Behavior']['Eigenworm1'],data['Behavior']['Eigenworm2'],data['Behavior']['Eigenworm3']]
# reverse rotate the Eigenworms
#pcs = np.array(np.dot(-RMatrix, np.vstack([pc1, pc2, pc3]))).T
# switch pc1 and two--they otherwise are switched everywhere :-(
#IMPORTANT TODO: Change to correct this
pcs = np.vstack([pc3, pc1, pc2]).T
print 'prefactors', pcs.shape
print 'Eigenworms', eigenworms.shape
index = 1000