Here is what I have done so far
import itertools
final_param_list = []
param_list_gen = [[8, 16, 32], ["Sigmoid", "ReLU", "Leaky ReLU"], [10, 20, 50], [1,2]]
for element in itertools.product(*param_list_gen):
final_param_list.append(element)
the output looks like
[(8, 'Sigmoid', 10, 1), (8, 'Sigmoid', 10, 2), ....]
For each list the values at each index are:
index0 = batch size
index1 = activation funtion
index2 = number of nodes
index3 = number of layers
So in the first list
batch_size = 8
activation='Sigmoid'
units=10
layers=1
What I want to be able to do is loop through the lists in final_param_list = [] and not only set each param but I want to add a hidden layer only when layers=2. I could go the easy route and just create two separate models, one with one hidden layer and the other with 2 hidden layers, and loop through them individually but I want to do something a little more elegant then that.
NOTE: I'm aware that some of this could probably be done with gridsearch AND I am aware that hidden layer 1 and 2 will have the same parameters. Ultimately, I will work my way toward being able to tune them separately but the solution as I have described it will suffice for now.
Ended up exploring a solution that @mkrieger1 made in the comments. Seems like it worked. Here is my code.
for param in final_param_list:
# ------ model 1 - 1 hidden layer ------ #
# Check to see if we are calling for one or two layers . If one layer then proceed
if param[3] == 1:
# hidden layer 1
q2model1.add(Dense(param[0]))
if param[1] != 'LeakyReLU':
q2model1.add(Activation(param[1]))
else:
q2model1.add(LeakyReLU(alpha=0.1))
# output layer
q2model1.add(Dense(class_num, activation='softmax'))
# ------ model 1 - 2 hidden layers ------ #
else:
# hidden layer 1
q2model1.add(Dense(param[0]))
if param[1] != 'LeakyReLU':
q2model1.add(Activation(param[1]))
else:
q2model1.add(LeakyReLU(alpha=0.1))
# hidden layer 2
q2model1.add(Dense(param[0]))
if param[1] != 'LeakyReLU':
q2model1.add(Activation(param[1]))
else:
q2model1.add(LeakyReLU(alpha=0.1))
# output layer
q2model1.add(Dense(class_num, activation='softmax'))
q2model1.compile(loss='sparse_categorical_crossentropy', optimizer='RMSProp',
metrics=['accuracy'])
history = q2model1.fit(X1, y1, epochs=20)
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.