while (a < len(list)):
if list_Evo[a][i] > b:
b = list_Evo[a][i]
if list_Evo[a][i] < c:
c = list_Evo[a][i]
a = a + 1
list_max_env[i] = b
list_min_env[i] = c
i = i + 1
#Creating a new file where average will be stored:
averagepath = testpath + "Model_Ver_" + paul[:
2] + "_average/plots_Model_Ver_" + paul[:
2] + "_average/"
assure_path_exists(testpath + "Model_Ver_" + paul[:2] + "_average/")
assure_path_exists(testpath + "Model_Ver_" + paul[:2] +
"_average/plots_Model_Ver_" + paul[:2] + "_average/")
f = open(averagepath + "FOM_evo_data.txt", "w+")
f.write("\n".join(map(str, list_ave_Evo)))
f.close()
copyfile(testpath + "Model_Ver_10/plots_Model_Ver_10/FOM_cut_data.txt",
averagepath + "FOM_cut_data.txt")
print("Model_Ver_" + paul[:2] + "-average saved! :-)")
if args.env:
f = open(averagepath + "FOM_max_data.txt", "w+")
f.write("\n".join(map(str, list_max_env)))
}
learning_rate = args.learningRate
my_decay = args.decay
myAdam = Adam(lr=learning_rate, decay=my_decay)
compileArgs['optimizer'] = myAdam
if args.verbose:
print "Opening file"
from commonFunctions import StopDataLoader, FullFOM, getYields, getDefinedClassifier, assure_path_exists
filepath = args.outputDir
baseName = filepath.replace(cfg.lgbk + "Searches/", "")
baseName = baseName.replace("/", "")
fileToPlot = "ROC_" + baseName
assure_path_exists(filepath + "/accuracy/" + "dummy.txt")
assure_path_exists(filepath + "/loss/" + "dummy.txt")
os.chdir(filepath)
#fileToPlot = "mGS:outputs_run_"+test_point+"_"+str(learning_rate)+"_"+str(my_decay)
f = open(fileToPlot + '.txt', 'w')
for y in [1, 2, 3]: # LAYERS
for x in range(2, 101): # NEURONS
if args.verbose:
print " ==> #LAYERS:", y, " #NEURONS:", x, " <=="
print("Starting the training")
model = getDefinedClassifier(len(trainFeatures), 1, compileArgs, x,
y, args.dropoutRate)
args = parser.parse_args()
n_layers = args.layers
n_neurons = args.neurons
n_epochs = args.epochs
batch_size = args.batchSize #len(XDev)/100
learning_rate = args.learningRate
my_decay = args.decay
dropout_rate = args.dropoutRate
regularizer = args.regularizer
iteration = args.iteration
dateSubmission = datetime.datetime.now().strftime("%Y-%m-%d_%H:%M")
baseName = cfg.lgbk + "SingleNN/" + dateSubmission + "/"
assure_path_exists(baseName + "dummy.txt")
for i in range(1, iteration + 1):
shPath = baseName + 'trainNN_Ver' + str(i) + '.sh'
with open(shPath, 'w') as f:
f.write("#!/bin/bash\n")
f.write("#$ -cwd\n")
f.write("#$ -pe mcore 3\n")
f.write("#$ -l container=True\n")
f.write("#$ -v CONTAINER=CENTOS7\n")
f.write("#...$ -v CONTAINER=UBUNTU16\n")
f.write("#$ -l gpu,release=el7\n")
f.write("cd /exper-sw/cmst3/cmssw/users/dbastos/StopNN/\n")
f.write("module load root-6.10.02\n")
f.write("python trainNN.py -z -l " + str(n_layers) + " -n " +
str(n_neurons) + " -e " + str(n_epochs) + " -a " +
#model_name = model_name.replace("Lr5_","Lr5.0_")
model_name = model_name + "_TP" + test_point + "_DT" + suffix
elif args.runNum != None:
filepath = cfg.lgbk + "Searches/run" + str(args.runNum)
model_name = "L" + str(args.layers) + "_N" + str(
args.neurons) + "_" + test_point + "_run" + str(args.runNum)
elif args.local:
filepath = "/home/diogo/PhD/SingleNN/" + model_name
os.chdir(filepath + "/")
if args.bk:
plots_path = filepath + "/plots_" + model_name + "_2bk/"
else:
plots_path = filepath + "/plots_" + model_name + "/"
assure_path_exists(plots_path)
if args.verbose:
print "Loading Model ..."
## Load your trainned model
with open(model_name + '.json', 'r') as json_file:
loaded_model_json = json_file.read()
model = model_from_json(loaded_model_json)
model.load_weights(model_name + ".h5")
model.compile(loss='binary_crossentropy', optimizer='adam')
if args.verbose:
print("Getting predictions")
devPredict = model.predict(XDev)
**trainParams)
# Time of the training
training_time = time.time() - start
if args.verbose:
print "Training took ", training_time, " seconds"
lr_list = lrm.lrates
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']
# assure_path_exists() is defined in commonFunctions.py
assure_path_exists(filepath + "accuracy/")
assure_path_exists(filepath + "loss/")
# Saving accuracy and loss values in a pickle file for later plotting
pickle.dump(acc, open(filepath + "accuracy/acc_" + name + ".pickle", "wb"))
pickle.dump(loss, open(filepath + "loss/loss_" + name + ".pickle", "wb"))
pickle.dump(val_acc,
open(filepath + "accuracy/val_acc_" + name + ".pickle", "wb"))
pickle.dump(val_loss,
open(filepath + "loss/val_loss_" + name + ".pickle", "wb"))
# Saving the trainned model
model_json = model.to_json() # model structure
with open(filepath + name + ".json", "w") as json_file:
json_file.write(model_json)