def gabelopsitu():
runevaler("opsitu",
2000,
GabelTrainer,
GabelTorchEvaler,
eval_gabel_ann,
networklayers=[70, 50, 20],
lr=.03,
dropoutrate=0.3,
validate_on_k=10,
filenamepostfix="gabel")
def esnniris():
runevaler("iris",
200,
ESNNSystem,
TorchEvaler,
eval_dual_ann,
networklayers=[40, 10, 2],
lr=.02,
dropoutrate=0.05,
validate_on_k=10,
filenamepostfix="esnn")
def gabeliris():
runevaler("iris",
2000,
GabelTrainer,
GabelTorchEvaler,
eval_dual_ann,
networklayers=[40, 40, 40],
lr=.01,
dropoutrate=0.00,
validate_on_k=10,
filenamepostfix="gabel")
def choprairis():
runevaler("iris",
200,
ChopraTrainer,
ChopraTorchEvaler,
eval_dual_ann,
networklayers=[40, 40],
lr=.05,
dropoutrate=0.05,
validate_on_k=10,
filenamepostfix="chopra")
def testTorchESNNOPSITU2(datasetname, epochs, eval_on_k, filename):
runevaler(datasetname,
epochs,
ESNNSystem,
TorchEvaler,
eval_dual_ann,
networklayers=[40, 40, 40],
lr=.09,
dropoutrate=0.5,
eval_on_k=eval_on_k,
filenamepostfix=filename)
def plotGableOpsitu(lr=0.03,
networklayers=[50, 30, 3],
dropout=0.05,
epochs=200,
validate_on_k=10,
n=5):
model, \
dsl, \
train, \
test, \
evaler, \
best_model, \
losses, \
lossdf = runevaler("opsitu", epochs, GabelTrainer, GabelTorchEvaler,
eval_gabel_ann, networklayers=networklayers,
lr=lr, dropoutrate=dropout, validate_on_k=validate_on_k, n=n,
filenamepostfix="gabel")
df1 = makematrixdata(best_model,
dsl.getFeatures()[train],
dsl.getTargets()[train],
10,
type=1)
plot2heatmap(df1, 10, annot=True, outputfile="gabel-matrix.pdf")
df2 = makematrixdata(model,
dsl.getFeatures()[train],
dsl.getTargets()[train],
10,
type=1)
plot2heatmap(df2, 10, annot=True, outputfile="training-gabel-matrix.pdf")
return losses, lossdf
def plotCHOPRAOpsitu(lr=0.2,
networklayers=[50, 30, 3],
dropout=0.05,
epochs=200,
validate_on_k=10,
n=5):
model, \
dsl, \
train, \
test, \
evaler, \
best_model, \
losses,\
lossdf = runevaler("opsitu", epochs, ChopraTrainer, ChopraTorchEvaler,
eval_dual_ann, networklayers=networklayers, lr=lr,
dropoutrate=dropout, validate_on_k=validate_on_k, n=n, filenamepostfix="chopra")
df = makematrixdata(best_model,
dsl.getFeatures()[train],
dsl.getTargets()[train],
10,
type=2)
plot2heatmap(df, 10, annot=True, outputfile="chopra-matrix.pdf")
df = makematrixdata(model,
dsl.getFeatures()[train],
dsl.getTargets()[train],
10,
type=2)
plot2heatmap(df, 10, annot=True, outputfile="training-chopra-matrix.pdf")
return losses, lossdf
def esnnopsitu():
model, \
test_data, \
test_target, \
evaler,\
best_model = runevaler("opsitu", 200, ESNNSystem, TorchEvaler,
eval_dual_ann, networklayers=[70, 20, 3], lr=.02,
dropoutrate=0.05, validate_on_k=10, filenamepostfix="esnn")
return model, best_model, test_data, test_target
def chopraopsitu():
model, \
test_data, \
test_target, \
evaler, \
best_model = runevaler("opsitu", 2000, ChopraTrainer, ChopraTorchEvaler,
eval_dual_ann, networklayers=[30, 10, 2],
lr=.01, dropoutrate=0.02, validate_on_k=10,
filenamepostfix="chopra")
print(model(evaler.x1s[0:1, ], evaler.x1s[0:1, ]))
print(model(evaler.x1s[1:2, ], evaler.x1s[1:2, ]))
print(model(evaler.x1s[1:2, ], evaler.x1s[0:1, ]))