def main():
max_episodes = 50000
# replay_buffer = deque()
# with tf.Session(config=tf.ConfigProto(device_count={'GPU':0})) as sess:
with tf.Session() as sess:
mainDQN = dqn.DQN(sess, pa.network_input_height, pa.network_input_width, pa.network_output_height, name="main")
# mainDQN = dqn.DQN(sess, pa.horizon * (pa.renewable) + pa.num_queue * (pa.renewable + 1), pa.network_output_height, name="main")
tf.compat.v1.global_variables_initializer().run()
trajs = deque()
trajs_rew = deque()
for episode in range(max_episodes):
e = 1. / ((episode / 1000) + 1)
done = False
step_count = 0
state = env.observe()
env.reset()
info = []
traj = []
show_result = 0;
while not done:
if np.random.rand(1) < e:
action = env.random_action()
# if episode % 10 == 0:
# print(info)
# print(action)
else:
action = np.argmax(mainDQN.predict(state))
# if episode % 10 == 0:
# print(info)
# print(mainDQN.predict(state))
# print("action:"+str(action))
# env.step(action)
next_state, reward, done, info = env.step(action)
traj.append((state, action, reward, next_state, done))
# replay_buffer.append((state, action, reward, next_state, done))
# if len(replay_buffer) > REPLAY_MEMORY:
# replay_buffer.popleft()
if done:
show_result = reward
state = next_state
step_count += 1
if step_count > 100:
break
trajs.append(traj)
if episode % 10 == 0:
print("Episode: {} steps: {} reward: {}".format(episode, step_count, show_result))
# print(info)
if episode % 10 == 1:
if episode != 1:
# minibatch = random.sample(trajs, 10)
loss, _ = replay_train(mainDQN, trajs)
print("Loss: ", loss)
trajs = deque()
def get_decoder(self, layer_of_activations="conv2d_15"):
# Looping through the old model and popping the encoder part + encoded layer
for i, l in enumerate(self.model.layers[0:19]):
self.model.layers.pop(0)
print(self.model.summary())
# Building a clean model that is the exact same architecture as the decoder part of the autoencoder
new_model = nb.build_decoder()
# Looping through both models and setting the weights on the new decoder
for i, l in enumerate(self.model.layers):
print(i, l.name, l.output_shape)
print(new_model.layers[i + 1].name,
new_model.layers[i + 1].output_shape)
new_model.layers[i + 1].set_weights(l.get_weights())
print(self.model.summary)
print(new_model.summary(200))
return new_model
import NetworkBuilder import os as os net = NetworkBuilder.NetworkBuilder() X, y = net.loadData() y0 = y[:, 0] y1 = y[:, 1] mlp_model = net.generateMLP() mlp_fit_model = net.fitModel(X, [y0, y1], mlp_model) mlp_model = net.loadWeights(model, "0001.hdf5") # examine model: mlp_model.summary() mlp_yhat = net.makePredictions(model, X) #--------------------------------- lstm_model = net.generateLSTM() # reshape for LSTM format X = X.reshape(X.shape[0], 1, X.shape[1]) lstm_fit_model = net.fitModel(X, [y0, y1], lstm_model) lstm_model = net.loadWeights(lstm_model, "0001.hdf5")
def binary_activation(x):
return nb.binary_activation(x)
from bokeh.plotting import output_file, show
output_file("RS_System.html")
template = """<span href="#" data-toggle="tooltip" title="<%= value %>"><%= value %></span>"""
byButton = True
curr = 0
title = PreText(text = "Search Rich Context Archive")
rs_title = PreText(text = "Recommendation Results: ")
df = pd.read_csv('data/ICPSR_ARCHIVE.csv')
df = df.replace(np.nan, '', regex=True)
with open('data/dataset_focals.json') as fl:
focals = json.load(fl)
G = nb.readNX('data/network_v2.5_contract.json')
all_keywords = [str(i) for i in G.nodes if (~str(i).startswith('data_')) & (~str(i).startswith('pub_')) & (~str(i).startswith('auth_'))]
output_file('ICPSR_Archive.html',mode='inline',root_dir=None)
source = ColumnDataSource(data=dict())
rs_source = ColumnDataSource(data=dict())
def update():
t = menu[nmenu.active][1]
logging.info(t)
current = df.sample(10)
if search.value != "":
if t != "":
current = df.loc[(df.ID.str.startswith(t))&(df.title.str.contains(search.value, case=False)==True)]
else:
def getRecommendations(Search_By='Keyword',
search='',
metric='Jaccard',
G=None):
all_res = pd.DataFrame()
nodes = []
if search == '':
print('MISSING ARGUMENT: Specify Search Value!')
return all_res
elif G is None:
G = nb.buildG()
nb.addCommunity(G)
allDatasets = [i for i in G.nodes if (str(i).startswith('data_'))]
allPubs = [i for i in G.nodes if (str(i).startswith('pub_'))]
all_keywords = [
str(i) for i in G.nodes if (~str(i).startswith('data_'))
& (~str(i).startswith('pub_')) & (~str(i).startswith('auth_'))
]
all_dataTitles = dict(
zip(datasets.title_lower,
['data_' + str(i) for i in datasets.data_set_id]))
all_pubTitles = dict(
zip(publications.title_lower,
['pub_' + str(i) for i in publications.publication_id]))
if Search_By == 'Keyword':
match = difflib.get_close_matches(search, all_keywords)
if len(match) > 0:
nodes = [
i for i in G.neighbors(match[0]) if str(i).startswith('data_')
] if len(match) > 0 else []
all_res = pd.DataFrame(
[int(i.replace('data_', '')) for i in nodes],
columns=['data_set_id'])
all_res['score'] = 1.0
nodes.extend([
i for i in G.neighbors(match[0]) if str(i).startswith('pub_')
])
elif Search_By == 'Dataset':
match = difflib.get_close_matches(
search,
list(all_dataTitles.keys()) + list(all_dataTitles.values()))
if len(match) > 0:
nodes = [match[0]
] if '_' in match[0] else [all_dataTitles[match[0]]]
elif Search_By == 'Publication Paper':
match = difflib.get_close_matches(
search,
list(all_pubTitles.keys()) + list(all_pubTitles.values()))
if len(match) > 0:
nodes = [match[0]
] if '_' in match[0] else [all_pubTitles[match[0]]]
else:
print('MISSING ARGUMENT: Specify Search Criteria!')
return all_res
# if len(match)>0:
# print(match[0])
if all_res.shape[0] < 10:
for n in nodes:
res = getNodeSim(n, G, metric)
all_res = res if res.shape[0] <= 0 else pd.concat([all_res, res],
sort=False)
all_res = all_res.reset_index(drop=True)
if all_res.shape[0] > 0:
all_res = pd.merge(all_res,
datasets[['data_set_id', 'title', 'description']],
how='left',
on='data_set_id')
all_res = all_res.sort_values(by='score',
ascending=False).reset_index(drop=True)
return all_res.iloc[:10]
else:
return all_res
name='x')
## labels
y_true = tf.placeholder(tf.float32, shape=[None, num_classes], name='y_true')
y_true_cls = tf.argmax(y_true, dimension=1)
def create_weights(shape):
return tf.Variable(tf.truncated_normal(shape, stddev=0.05))
def create_biases(size):
return tf.Variable(tf.constant(0.05, shape=[size]))
nb = NetworkBuilder.NetworkBuilder()
with tf.name_scope("ModelV2") as scope:
model = x
model = nb.flatten(model)
model = nb.attach_relu_layer(model)
model = nb.attach_sigmoid_layer(model)
model = nb.attach_dense_layer(model, 3)
y_pred = tf.nn.softmax(model, name='y_pred')
y_pred_cls = tf.argmax(y_pred, dimension=1)
session.run(tf.global_variables_initializer())
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=model,
labels=y_true)
cost = tf.reduce_mean(cross_entropy)