forked from vlimant/summer16-NikolausHowe
-
Notifications
You must be signed in to change notification settings - Fork 1
/
long_train.py
35 lines (25 loc) · 1.17 KB
/
long_train.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
# coding: utf-8
# ## Here we train single-hidden-layer fully-connected nets with different numbers of nodes
# Import functions
import setGPU0
from io_functions import *
from draw_functions import *
# 1 is signal; 0 is background
train_data, test_data, train_labels, test_labels = train_test(shape=(10000,), split=0.33)
# Choose our list of number of nodes
num_nodes = [100]
# Make a one-hidden-layer network with that number of nodes for each entry in num_nodes.
# Train it for 100 epochs, save the model, the weights, and loss history.
for number in num_nodes:
# Construct and compile a network
model = Sequential()
model.add(Dense(number, input_dim=10000, activation='sigmoid'))
model.add(Dropout(0.5))
model.add(Dense(number, activation='sigmoid'))
model.add(Dropout(0.5))
model.add(Dense(1, init='uniform', activation='sigmoid'))
model.compile(loss='binary_crossentropy', optimizer='sgd')
# Train the network
my_fit = model.fit(train_data, train_labels, nb_epoch=1000, batch_size=1000, verbose=1)
# Store the model, the weights, and the loss history
store_model(model, my_fit.history['loss'], 'dense'+str(number)+'long')