def postTreatDev(cccs, preds, gs, nDim): #First we calculate the bias gsMean = np.nanmean(gs['dev'][nDim]) predMean = np.nanmean(preds['dev']) bias = gsMean - predMean #We add the bias to the prediction and save if there is an improvement predCenter = preds['dev'] + bias cccBias = cccCalc(predCenter,gs['dev'][nDim]) if (cccBias > cccs['dev']): cccs['dev'] = cccBias preds['dev'] = predCenter else : bias = 0.0 #We now scale the prediction and do the same thing #First we calculate the scale stdGs = np.nanstd(gs['dev'][nDim]) stdPred = np.nanstd(preds['dev']) scale = stdGs/stdPred #We apply the scale and save if improvement predScale = np.multiply(preds['dev'],scale) cccScale = cccCalc(predScale,gs['dev'][nDim]) if (cccScale > cccs['dev']) : cccs['dev'] = cccScale preds['dev'] = predScale else : scale = 0.0 return cccs['dev'], preds['dev'], bias, scale
def postTreatTest(gs, pred, ccc, bias, scale, nDim):
for s in v.aPart:
gspt = np.array(gs[s])[nDim]
if (bias != 0.0):
#We add the bias to the prediction and save if there is an improvement
pred[s] = np.array(pred[s]) + bias
ccc[s] = cccCalc(pred[s], gspt)
if (scale != 0.0):
#We apply the scale and save if improvement
pred[s] = np.multiply(pred[s], scale)
ccc[s] = cccCalc(pred[s], gspt)
return ccc, pred
def unimodalPredDev(gs, feats, nDim):
parts = ['dev']
[cccs, preds] = [{} for i in range(2)]
for s in parts:
cccs[s] = -1.0
warnings.filterwarnings('ignore', category=ConvergenceWarning)
#Liblinear
for comp in v.C:
#Options for liblinear
options = "-s "+str(v.sVal)+" -c "+str(comp)+" -B 1 -q"
#We learn the model on train
model = train(gs['train'][nDim],feats['train'],options)
#We predict on data
for s in parts:
pred = np.array(predict(gs[s][nDim],feats[s],model,"-q"))[0]
#We calculate the correlation and store it
ccc = cccCalc(np.array(pred),gs[s][nDim])
if (ccc > cccs[s]):
preds[s] = pred
cccs[s] = ccc
function = "SVR"
alpha = comp
if (v.fullMode == True):
#We see if we can do better with sklearn
for nbFunc in range(len(v.lFunc)):
for c in v.parFunc[nbFunc]:
func = v.lFunc[nbFunc]
reg = func[0](alpha=c)
#One task prediction
if (func[1] == 0):
reg.fit(feats['train'],gs['train'][nDim])
for s in parts:
p = reg.predict(feats['dev'])
ccc = cccCalc(p,gs[s][nDim])
if (ccc > cccs[s]) :
preds[s] = p
cccs[s] = ccc
function = func[2]
alpha = c
#Multi task prediction
else :
reg.fit(feats['train'],np.transpose(gs['train']))
for s in parts:
p = reg.predict(feats['dev'])[:,nDim]
ccc = cccCalc(p,gs[s][nDim])
if (ccc > cccs[s]) :
preds[s] = p
cccs[s] = ccc
function = func[2]
alpha = c
return cccs, preds, function, alpha
def linRegMult(datas, func, c, part, cMode, cSize):
res = [func, c, [], [], {}, cMode, cSize]
warnings.filterwarnings('ignore', category=ConvergenceWarning)
#Getting the coefficient for each modality on Dev
if (c != 0):
reg = func[0](alpha=c)
else:
reg = func[0]()
for nDim in range(len(v.eName)):
for nMod in range(len(datas['dev'][nDim])):
if (nMod == 0):
preds = datas['dev'][nDim][nMod]
else:
preds = np.concatenate((preds, datas['dev'][nDim][nMod]),
axis=1)
reg.fit(preds, np.transpose(datas['gsdev']))
res[2] = reg.coef_
#Doing the new prediction
cccs = []
for nDim in range(len(v.eName)):
cccs = []
for s in part:
if (res[4].get(s, None) == None):
res[4][s] = []
res[4][s].append(predMulti(reg.coef_, datas[s], nDim, 1, cSize))
cccs.append(
round(cccCalc(res[4][s][nDim], datas['gs' + s][nDim]), 3))
res[3].append(cccs)
return res
def unimodalPredTest(gs, feats, nDim, func, c):
[cccs, preds] = [{} for i in range(2)]
for s in v.aPart:
cccs[s] = -1.0
warnings.filterwarnings('ignore', category=ConvergenceWarning)
if (func == "SVR"):
#Options for liblinear
options = "-s " + str(v.sVal) + " -c " + str(c) + " -B 1 -q"
#We learn the model on train
model = train(gs['train'][nDim], feats['train'], options)
#We predict on data
for s in v.aPart:
pred = np.array(predict(gs[s][nDim], feats[s], model, "-q"))[0]
#We calculate the correlation and store it
ccc = cccCalc(np.array(pred), gs[s][nDim])
if (ccc > cccs[s]):
preds[s] = pred
cccs[s] = ccc
else:
for f in v.lFunc:
if (f[2] == func):
fun = f
reg = fun[0](alpha=c)
if (fun[1] == 0):
reg.fit(feats['train'], gs['train'][nDim])
for s in v.aPart:
p = reg.predict(feats[s])
ccc = cccCalc(p, gs[s][nDim])
if (ccc > cccs[s]):
preds[s] = p
cccs[s] = ccc
else:
reg.fit(feats['train'], np.transpose(gs['train']))
for s in v.aPart:
p = reg.predict(feats[s])[:, nDim]
ccc = cccCalc(p, gs[s][nDim])
if (ccc > cccs[s]):
preds[s] = p
cccs[s] = ccc
return cccs, preds, func, c