def test_train_loopClass(Xtrain,Ytrain,Xtest,Ytest,iterations,ridge_penalty,hiddens,plot_in_train=True):
"""or loops to iterate through training iterations, ridge penalty,
and hidden layer list
"""
results = {}
global nnet,random_network_seed
for niter in iterations:
for penalty in ridge_penalty:
for hidden in hiddens:
### Check / use random seed
if random_network_seed == None:
np.random.seed(None)
random_network_seed = int(np.random.randint(1, 100000))
np.random.seed(random_network_seed)
random.seed(random_network_seed)
tf.set_random_seed(0)
### Standardize the data
Xtrain,Xtest,stdVals = dSS.standardize_data(Xtrain,Xtest)
Xmean,Xstd = stdVals
### Define the model
model = defineNN(hidden,
input_shape=np.shape(Xtrain)[1],
output_shape=np.shape(Ytrain)[1],
ridgePenalty=penalty)
### Train the net
model, history = trainNN(model,Xtrain,
Ytrain,niter=niter,verbose=0)
### After training, use the network with training data to
### check that we don't have any errors and output RMSE
rmse_train = dSS.rmse(Ytrain,model.predict(Xtrain))
if type(Ytest) != bool:
rmse_test = 0.
rmse_test = dSS.rmse(Ytest,model.predict(Xtest))
else:
rmse_test = False
this_result = {'iters': niter,
'hiddens' : hidden,
'RMSE Train' : rmse_train,
'RMSE Test' : rmse_test,
'ridge penalty': penalty,
'zero mean' : rm_annual_mean,
'zero merid mean' : rm_merid_mean,
'land only?' : land_only,
'ocean only?' : ocean_only,
'Segment Seed' : random_segment_seed,
'Network Seed' : random_network_seed }
results.update(this_result)
global experiment_result
experiment_result = experiment_result.append(results,
ignore_index=True)
#'unlock' the random seed
np.random.seed(None)
random.seed(None)
tf.set_random_seed(None)
return experiment_result, model
# ### Loop over folds
for loop in np.arange(0,foldsN):
K.clear_session()
#---------------------------
#random_segment_seed = 34515
random_segment_seed = int(np.genfromtxt('/Users/zlabe/Documents/Research/ExtremeEvents/Data/SelectedSegmentSeed.txt',unpack=True))
#---------------------------
Xtrain,Ytrain,Xtest,Ytest,Xtest_shape,Xtrain_shape,data_train_shape,data_test_shape,testIndices = segment_data(data,classesl,segment_data_factor)
YtrainClassMulti = Ytrain
YtestClassMulti = Ytest
# For use later
XtrainS,XtestS,stdVals = dSS.standardize_data(Xtrain,Xtest)
Xmean, Xstd = stdVals
#---------------------------
random_network_seed = 87750
#---------------------------
# Create and train network
exp_result,model = test_train_loopClass(Xtrain,
YtrainClassMulti,
Xtest,
YtestClassMulti,
iterations=iterations,
ridge_penalty=ridge_penalty,
hiddens=hiddensList,
plot_in_train = True)