def regtree(par, *data):
X_train, X_test, Y_train, Y_test = data
regTreeModel=tree.DecisionTreeRegressor\
(max_features=par[0],min_samples_split=par[1],min_samples_leaf=par[2],
min_weight_fraction_leaf=par[3],max_leaf_nodes=int(par[4]))
fitModel = linear_model.LinearRegression()
Yp,Yptrain,regTreeModel,fitModelList,predind=\
SSRS.RegressionTree(X_train,X_test,Y_train,Y_test,regTreeModel,fitModel,Field,
doFitSelection=0,doMultiBand=1)
rmse, rmse_band = SSRS.RMSEcal(Yp, Y_test)
print(rmse)
return rmse
Y = UCData
attrind = np.array(range(1, 51) + range(62, 78, 3))
Field = [Field[i] for i in range(1, 51) + range(62, 78, 3)]
X = AttrData[:, attrind]
X[np.isnan(X)] = 0
scaler = preprocessing.StandardScaler().fit(X)
Xn = scaler.fit_transform(X)
### cluster
model = KMeans(init='k-means++', n_clusters=6, n_init=10, max_iter=1000)
model = AffinityPropagation(preference=-150, verbose=True)
#model = Birch(branching_factor=10, n_clusters=4, threshold=0.3, compute_labels=True)
model = MeanShift(bandwidth=estimate_bandwidth(X, quantile=0.1, n_samples=100),
bin_seeding=True)
label = SSRS.Cluster(X, model)
### classification
model = tree.DecisionTreeClassifier()
model = GaussianNB()
model = svm.SVC()
model = SGDClassifier()
Tp = SSRS.Classification_cross(XXn, T=label, nfold=10, model=model)
SSRS.plotErrorMap(label, Tp)
### regression
regModel = linear_model.LinearRegression()
#regModel=svm.SVC()
regModel = KNeighborsRegressor(n_neighbors=10)
regModel = tree.DecisionTreeRegressor()
Xn=scaler.fit_transform(X) ### cluster model = KMeans(init='k-means++', n_clusters=6, n_init=10, max_iter=1000) model = AffinityPropagation(preference=-150,verbose=True) #model = Birch(branching_factor=10, n_clusters=4, threshold=0.3, compute_labels=True) model = MeanShift(bandwidth=estimate_bandwidth(X, quantile=0.1, n_samples=100), bin_seeding=True) label=SSRS.Cluster(X, model) ### classification model = tree.DecisionTreeClassifier() model = GaussianNB() model = svm.SVC() model = SGDClassifier() Tp = SSRS.Classification_cross(XXn, T=label, nfold=10, model=model) SSRS.plotErrorMap(label, Tp) ### regression regModel=linear_model.LinearRegression() #regModel=svm.SVC() regModel=KNeighborsRegressor(n_neighbors=10) regModel = tree.DecisionTreeRegressor() regModel = GaussianNB() rmse_band,Yp,Ytest=SSRS.RegressionLearn(X,XXn,0.2,regModel)
## plot tree
regModel.fit(X2_train, Y2_train)
savedir=r"/Volumes/wrgroup/Kuai/USGSCorr/figure_tree/"
savedir=r"Y:\Kuai\USGSCorr\figure_tree\\"
with open(savedir+"tree.dot", 'w') as f:
f = tree.export_graphviz(regModel, out_file=f,feature_names=[Field[i] for i in attrsel],
label='none',node_ids=True)
os.system("dot -Tpng tree.dot -o tree.png")
regTree=regModel.tree_
feature_names=[Field[i] for i in attrind]
Xin=X2_train
Yin=Y2_train
string,nodeind,leaf,label=SSRS.traverseTree(regTree,feature_names,Xin)
for i in range(0,regTree.node_count):
plt.figure()
plt.boxplot(Yin[nodeind[i],:])
plt.title(string[i],fontsize=8)
#plt.tight_layout()
plt.savefig(savedir+"Train_node%i"%i)
plt.close()
Xin=X2_test
Yin=Y2_test
string,nodeind,leaf,label=SSRS.traverseTree(regTree,feature_names,Xin)
for i in range(0,regTree.node_count):
plt.figure()
plt.boxplot(Yin[nodeind[i],:])
plt.title(string[i],fontsize=8)
def testModel(predListTest):
nmodel = predListTest.__len__()
nit = 50
# test 1: different size of training and test
testErr = np.ones([nmodel, 5, 50])
testsize = [0.2, 0.4, 0.5, 0.6, 0.7]
for k in range(0, nit):
print(k)
for j in range(0, nmodel):
for i in range(0, 5):
ind = range(0, nind)
X_train,X_test,Y_train,Y_test,ind_train,ind_test = \
cross_validation.train_test_split(Xn,dist,ind,test_size=testsize[i],random_state=k)
regTreeModel = tree.DecisionTreeRegressor(max_leaf_nodes=20,
min_samples_leaf=20)
fitModel = linear_model.LinearRegression()
predSel = predListTest[j]
predName = [Field[jj] for jj in predSel]
Yp,Yptrain,regTreeModel,fitModelList,predind=\
SSRS.RegressionTree(X_train[:,predSel],X_test[:,predSel],Y_train,Y_test,regTreeModel,fitModel,predName,
doFitSelection=0,doMultiBand=1)
rmse, rmse_band = SSRS.RMSECal(Yp, Y_test)
testErr[j, i, k] = rmse
# test 2: use 1 HUC2 as test
testErr1_huc2_rt = np.ones([nmodel, 18])
trainErr1_huc2_rt = np.ones([nmodel, 18])
IDhucfile = r"E:\work\SSRS\data\IDhuc_mb_4949.mat"
mat = sio.loadmat(IDhucfile)
IDhuc = mat["IDhuc"]
huc2 = IDhuc[indvalid, 1]
for k in range(0, nit):
print(k)
for i in range(0, 18):
ind = range(0, nind)
X_train,X_test,Y_train,Y_test,ind_train,ind_test = \
cross_validation.train_test_split(Xn,dist,ind,test_size=0.2,random_state=k)
ind_test = np.where(huc2 == i + 1)[0]
X_test = Xn[ind_test, :]
Y_test = dist[ind_test, :]
for j in range(0, nmodel):
regTreeModel = tree.DecisionTreeRegressor(max_leaf_nodes=20)
fitModel = linear_model.LinearRegression()
predSel = predListTest[j]
predName = [Field[jj] for jj in predSel]
Yp,Yptrain,regTreeModel,fitModelList,predind=\
SSRS.RegressionTree(X_train[:,predSel],X_test[:,predSel],Y_train,Y_test,regTreeModel,fitModel,predName,
doFitSelection=0,doMultiBand=1)
rmse, rmse_band = SSRS.RMSECal(Yptrain, Y_train)
trainErr1_huc2_rt[j, i] = rmse
rmse, rmse_band = SSRS.RMSECal(Yp, Y_test)
testErr1_huc2_rt[j, i] = rmse
# test 3: leave out 1 HUC2 one time
testErr1_huc2 = np.ones([nmodel, 18])
trainErr1_huc2 = np.ones([nmodel, 18])
IDhucfile = r"E:\work\SSRS\data\IDhuc_mb_4949.mat"
mat = sio.loadmat(IDhucfile)
IDhuc = mat["IDhuc"]
huc2 = IDhuc[indvalid, 1]
for i in range(0, 18):
ind_test = np.where(huc2 == i + 1)[0]
ind_train = np.where(huc2 != i + 1)[0]
X_train = Xn[ind_train, :]
X_test = Xn[ind_test, :]
Y_train = dist[ind_train, :]
Y_test = dist[ind_test, :]
for j in range(0, nmodel):
regTreeModel = tree.DecisionTreeRegressor(max_leaf_nodes=20)
fitModel = linear_model.LinearRegression()
predSel = predListTest[j]
predName = [Field[jj] for jj in predSel]
Yp,Yptrain,regTreeModel,fitModelList,predind=\
SSRS.RegressionTree(X_train[:,predSel],X_test[:,predSel],Y_train,Y_test,regTreeModel,fitModel,predName,
doFitSelection=0,doMultiBand=1)
rmse, rmse_band = SSRS.RMSECal(Yptrain, Y_train)
trainErr1_huc2[j, i] = rmse
rmse, rmse_band = SSRS.RMSECal(Yp, Y_test)
testErr1_huc2[j, i] = rmse
# test 4: leave out 2 HUC2 one time
testErr2_huc2 = np.ones([nmodel, 18 * 17])
trainErr2_huc2 = np.ones([nmodel, 18 * 17])
hucTab = np.ones([18 * 17, 2])
IDhucfile = r"E:\work\SSRS\data\IDhuc_mb_4949.mat"
mat = sio.loadmat(IDhucfile)
IDhuc = mat["IDhuc"]
huc2 = IDhuc[indvalid, 1]
n = -1
for i in range(0, 18):
print(i)
for j in range(0, 18):
if i == j:
continue
n = n + 1
hucTab[n, 0] = i
hucTab[n, 1] = j
ind_test = np.where((huc2 == i + 1) | (huc2 == j + 1))[0]
ind_train = np.where((huc2 != i + 1) & (huc2 != j + 1))[0]
X_train = Xn[ind_train, :]
X_test = Xn[ind_test, :]
Y_train = dist[ind_train, :]
Y_test = dist[ind_test, :]
for k in range(0, nmodel):
regTreeModel = tree.DecisionTreeRegressor(max_leaf_nodes=20)
fitModel = linear_model.LinearRegression()
predSel = predListTest[k]
predName = [Field[jj] for jj in predSel]
Yp,Yptrain,regTreeModel,fitModelList,predind=\
SSRS.RegressionTree(X_train[:,predSel],X_test[:,predSel],Y_train,Y_test,regTreeModel,fitModel,predName,
doFitSelection=0,doMultiBand=1)
rmse, rmse_band = SSRS.RMSECal(Yptrain, Y_train)
trainErr2_huc2[k, n] = rmse
rmse, rmse_band = SSRS.RMSECal(Yp, Y_test)
testErr2_huc2[k, n] = rmse
return testErr,trainErr1_huc2,testErr1_huc2,\
trainErr1_huc2_rt,testErr1_huc2_rt,\
trainErr2_huc2,testErr2_huc2,hucTab
X = np.delete(X, indnan, 0)
Y = np.delete(Y, indnan, 0)
indvalid = np.delete(indvalid, indnan, 0)
scaler = preprocessing.StandardScaler().fit(X)
Xn = scaler.fit_transform(X)
[nind, nband] = Y.shape
[nind, nattr] = X.shape
# test for k in kmean
score_cluster = np.zeros(8)
for i in range(2, 10):
print(i)
nc = i
model = KMeans(init='k-means++', n_clusters=nc, n_init=10, max_iter=1000)
label, center = SSRS.Cluster(Y, model, doplot=0)
score_cluster[i - 2] = metrics.silhouette_score(Y, label)
plt.plot(range(2, 10), score_cluster, '-*')
## cluster
nc = 6
model = KMeans(init='k-means++',
n_clusters=nc,
n_init=15,
max_iter=1000,
tol=1e-15,
verbose=True)
label, center = SSRS.Cluster(Y, model, doplot=0)
## PCA
pca = PCA(n_components=nband)
scaler = preprocessing.StandardScaler().fit(X)
Xn = scaler.fit_transform(X)
[nind, nband] = Y.shape
[nind, nattr] = X.shape
################################################################
# CLUSTER
################################################################
nc = 6
model = KMeans(init='k-means++',
n_clusters=nc,
n_init=15,
max_iter=1000,
tol=1e-15,
verbose=True)
label, center = SSRS.Cluster(Y, model, doplot=0)
## PCA
pca = PCA(n_components=nband)
pca.fit(Y)
Ypca = pca.transform(Y)
Cpca = pca.transform(center)
Ypca[:, 0] = -Ypca[:, 0]
Cpca[:, 0] = -Cpca[:, 0]
## rename clusters
ythe = np.array([0])
label, Cpca, center = SSRS.Cluster_rename(label, ythe, Cpca, center)
## plot PCA and cluster after resign name
SSRS.Cluster_plot(Y, label, center)
# nn = Regressor(
# layers=[
# Layer("Sigmoid", units=200),
# Layer("Sigmoid", units=200),
# Layer("Linear")],
# learning_rate=0.1,
# n_iter=200,verbose=1)
X_train,X_test,Y_train,Y_test = cross_validation.train_test_split(\
Xn,Y,test_size=0.2,random_state=0)
# predict correlation
Yp,Yptrain,regModelList=SSRS.Regression\
(X_train,X_test,Y_train,Y_test,multiband=1,regModel=regModel,doplot=0)
rmse, rmse_band = SSRS.RMSEcal(Yp, Y_test)
rmse_train, rmse_band_train = SSRS.RMSEcal(Yptrain, Y_train)
print(rmse)
print(rmse_train)
print(np.corrcoef(Yp[:, 0], Y_test[:, 0]))
par = [1.0, 16, 6, 20, 0.15]
regTreeModel=tree.DecisionTreeRegressor\
(max_features=par[0],max_depth=par[1],min_samples_split=par[2],min_samples_leaf=par[3],
min_weight_fraction_leaf=par[4],max_leaf_nodes=18)
fitModel = linear_model.LinearRegression()
Yp,Yptrain,regTreeModel,fitModelList,predind=SSRS.RegressionTree\
(X_train,X_test,Y_train,Y_test,regTreeModel,fitModel,Field,doFitSelection=0)
rmse, rmse_band = SSRS.RMSEcal(Yp, Y_test)
rmse_train, rmse_band_train = SSRS.RMSEcal(Yptrain, Y_train)
print(rmse)
Y2=np.argmax(Y[:,15:30],axis=1)
## Regression
regModel=linear_model.LinearRegression()
#regModel=svm.SVC()
regModel=KNeighborsRegressor(n_neighbors=20)
regModel=tree.DecisionTreeRegressor()
regModel=GaussianNB()
regModel=sklearn.linear_model.SGDRegressor()
regModel=RandomForestRegressor()
X_train,X_test,Y_train,Y_test = cross_validation.train_test_split(\
Xn,np.column_stack((Y1,Y2)),test_size=0.2,random_state=0)
Yp,rmse,rmse_train,rmse_band,rmse_band_train=SSRS.Regression\
(X_train,X_test,Y_train,Y_test,multiband=1,regModel=regModel,doplot=0)
print(rmse)
print(rmse_train)
## Classification
model = tree.DecisionTreeClassifier()
model = GaussianNB()
model = svm.SVC()
model = SGDClassifier()
model=sklearn.ensemble.RandomForestClassifier()
Yin=Y1
Tp = SSRS.Classification_cross(Xn, T=Yin, nfold=10, model=model)
SSRS.plotErrorMap(Yin, Tp)
np.sqrt(((Yin - Tp) ** 2).mean())
np.count_nonzero(np.abs(Yin-Tp)<2)/4627.