我们从Python开源项目中，提取了以下15个代码示例，用于说明如何使用tensorflow.summary()。
def close(self): """Close the simulation, freeing resources. Notes ----- The simulation cannot be restarted after it is closed. This is not a technical limitation, just a design decision made for all Nengo simulators. """ if not self.closed: # note: we use getattr in case it crashes before the object is # created if getattr(self, "sess", None) is not None: self.sess.close() self.sess = None if getattr(self, "summary", None) is not None: self.summary.close() self.closed = True
def learn_actor(self, s, x_ma, epoch): # batch update _, police_grads = self.sess.run(self.train_ops, feed_dict={S: s, X_MA: x_ma}) # the following method for soft replace target params is computational expansive # target_params = (1-tau) * target_params + tau * eval_params # self.sess.run([tf.assign(t, (1 - self.tau) * t + self.tau * e) for t, e in zip(self.t_params, self.e_params)]) summary = tf.Summary() # summary.value.add(tag='info/c_gradient{}'.format(self.agent_id), simple_value=float(_c_grad)) summary.value.add(tag='info/police_grads{}'.format(self.agent_id), simple_value=np.mean([np.mean(_) for _ in police_grads])) writer.add_summary(summary, epoch) writer.flush() # instead of above method, I use a hard replacement here if self.t_replace_counter % self.t_replace_iter == 0: self.sess.run([tf.assign(t, e) for t, e in zip(self.t_params, self.e_params)]) self.t_replace_counter += 1
def __init__(self, config): self.config = config self.data = DataSet(self.config) self.add_placeholders() self.summarizer = tf.summary self.net = Network(config) self.saver = tf.train.Saver() self.epoch_count, self.second_epoch_count = 0, 0 self.outputs, self.prob = self.net.neural_search() self.hyperparams = self.net.gen_hyperparams(self.outputs) self.hype_list = [1 for i in range(self.config.hyperparams)] #[7, 7, 24, 5, 5, 36, 3, 3, 48, 64] self.reinforce_loss = self.net.REINFORCE(self.prob) self.tr_cont_step = self.net.train_controller(self.reinforce_loss, self.val_accuracy) self.cNet, self.y_pred = self.init_child(self.hype_list) self.cross_loss, self.accuracy, self.tr_model_step = self.grow_child() self.init = tf.global_variables_initializer() self.local_init = tf.local_variables_initializer()
def tf_log(summary, collection=SCALAR_SUMMARIES): """Add tf.summary object to collection named collection""" tf.add_to_collection(collection, summary)
def on_epoch_end(self, epoch, logs=None): logs = logs or {} if self.model.validation_data and self.histogram_freq: if epoch % self.histogram_freq == 0: # TODO: implement batched calls to sess.run # (current call will likely go OOM on GPU) if self.model.uses_learning_phase: cut_v_data = len(self.model.inputs) val_data = self.model.validation_data[:cut_v_data] + [0] tensors = self.model.inputs + [K.learning_phase()] else: val_data = self.model.validation_data tensors = self.model.inputs feed_dict = dict(zip(tensors, val_data)) result = self.sess.run([self.merged], feed_dict=feed_dict) summary_str = result[0] self.writer.add_summary(summary_str, epoch) for name, value in logs.items(): if name in ['batch', 'size']: continue summary = tf.Summary() summary_value = summary.value.add() summary_value.simple_value = value.item() summary_value.tag = name self.writer.add_summary(summary, epoch) self.writer.flush()
def visualize(self, net_inputs, net_outputs): # input a batch of training examples of form [Tensor1, Tensor2, ... Tensor_n] # net_inputs: usually images; net_outputs: usually the labels # this function visualize the data that is read in, do not return anything but use tf.summary # visualize the video using multiple images # their is no way to visualize time sequence now, so isvalid and isstop couldn't be visualized if not FLAGS.no_image_input: decoded = net_inputs[0] visualize = tf.cast(decoded[0,:,:,:,:], tf.uint8) tf.image_summary("video_seq", visualize, max_images=FLAGS.n_sub_frame)
def visualize(self, net_inputs, net_outputs): # input a batch of training examples of form [Tensor1, Tensor2, ... Tensor_n] # net_inputs: usually images; net_outputs: usually the labels # this function visualize the data that is read in, do not return anything but use tf.summary raise NotImplemented()
def learn_critic(self, x_ma, a_ma, r, x2_ma, s, a, s2, epoch=0): # ATTENTION!!!! # the key point is that we use constant a_ma to replace critic's tensor: self.a_ma # here we must replace this tensor, otherwise whole network crash # because critic must use constant a_ma to do gradient, # while actor must use its network tensor a_ma to do gradient # this is the trick!! _c_grad, _c_loss, _a_grads = self.sess.run( self.train_ops, feed_dict={X_MA: x_ma, self.a_ma: a_ma, R: r, X2_MA: x2_ma, S: s, S2: s2}) summary = tf.Summary() # summary.value.add(tag='info/c_gradient{}'.format(self.agent_id), # simple_value=float(_c_grad)) summary.value.add(tag='info/c_loss{}'.format(self.agent_id), simple_value=float(_c_loss)) writer.add_summary(summary, epoch) writer.flush() # the following method for soft replace target params is computational expansive # target_params = (1-tau) * target_params + tau * eval_params # self.sess.run([tf.assign(t, (1 - self.tau) * t + self.tau * e) for t, e in zip(self.t_params, self.e_params)]) # instead of above method, we use a hard replacement here if self.t_replace_counter % self.t_replace_iter == 0: self.sess.run([tf.assign(t, e) for t, e in zip(self.t_params, self.e_params)]) self.t_replace_counter += 1 # ------------------- Memory -------------------
def add_summaries(self, sess): if self.config.load: path_ = "../results/tensorboard" else : path_ = "../bin/results/tensorboard" summary_writer_train = tf.summary.FileWriter(path_ + "/train", sess.graph) summary_writer_val = tf.summary.FileWriter(path_ + "/val", sess.graph) summary_writer_test = tf.summary.FileWriter(path_+ "/test", sess.graph) summary_writers = {'train': summary_writer_train, 'val': summary_writer_val, 'test': summary_writer_test} return summary_writers
def set_model(self, model): self.model = model self.sess = K.get_session() if self.histogram_freq and self.merged is None: for layer in self.model.layers: for weight in layer.weights: if hasattr(tf, 'histogram_summary'): tf.histogram_summary(weight.name, weight) else: tf.summary.histogram(weight.name, weight) if self.write_images: w_img = tf.squeeze(weight) shape = w_img.get_shape() if len(shape) > 1 and shape[0] > shape[1]: w_img = tf.transpose(w_img) if len(shape) == 1: w_img = tf.expand_dims(w_img, 0) w_img = tf.expand_dims(tf.expand_dims(w_img, 0), -1) if hasattr(tf, 'image_summary'): tf.image_summary(weight.name, w_img) else: tf.summary.image(weight.name, w_img) if hasattr(layer, 'output'): if hasattr(tf, 'histogram_summary'): tf.histogram_summary('{}_out'.format(layer.name), layer.output) else: tf.summary.histogram('{}_out'.format(layer.name), layer.output) if hasattr(tf, 'merge_all_summaries'): self.merged = tf.merge_all_summaries() else: self.merged = tf.summary.merge_all() if self.write_graph: if hasattr(tf, 'summary') and hasattr(tf.summary, 'FileWriter'): self.writer = tf.summary.FileWriter(self.log_dir, self.sess.graph) elif parse_version(tf.__version__) >= parse_version('0.8.0'): self.writer = tf.train.SummaryWriter(self.log_dir, self.sess.graph) else: self.writer = tf.train.SummaryWriter(self.log_dir, self.sess.graph_def) else: if hasattr(tf, 'summary') and hasattr(tf.summary, 'FileWriter'): self.writer = tf.summary.FileWriter(self.log_dir) else: self.writer = tf.train.SummaryWriter(self.log_dir)
def __init__(self, sess, state_dim, action_dim, learning_rate, gamma, t_replace_iter, a_ma, a2_ma, agent_id, agent_num): self.agent_id = agent_id self.agent_num = agent_num self.sess = sess self.s_dim = state_dim self.a_dim = action_dim self.lr = learning_rate self.gamma = gamma self.t_replace_iter = t_replace_iter self.t_replace_counter = 0 with tf.variable_scope('Critic{}'.format(self.agent_id)): # Input (s, a), output q local_a = a_ma[agent_id] self.a_ma = tf.concat(a_ma, axis=1) self.q = self._build_critic_net(X_MA, self.a_ma, 'eval_net', trainable=True) # Input (s_, a_), output q_ for q_target a2_ma = tf.concat(a2_ma, axis=1) self.q_ = self._build_critic_net(X2_MA, a2_ma, 'target_net', trainable=False) # target_q is based on a_ from Actor's target_net self.e_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='Critic{}/eval_net'.format(agent_id)) self.t_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='Critic{}/target_net'.format(agent_id)) with tf.variable_scope('target_q{}'.format(self.agent_id)): self.target_q = R + self.gamma * self.q_ with tf.variable_scope('TD_error{}'.format(self.agent_id)): self.loss = tf.reduce_mean(tf.squared_difference(self.target_q, self.q)) # MSE with tf.variable_scope('C_train{}'.format(self.agent_id)): self.train_ops = [] self.train_ops.append(tf.train.AdamOptimizer(self.lr).minimize( self.loss, var_list=self.e_params)) # C train only update c network, don't update a self.train_ops.append(self.loss) # for tf.summary with tf.variable_scope('a_grad{}'.format(self.agent_id)): # tensor of gradients of each sample (None, a_dim) self.a_grads = tf.gradients(self.q, local_a)[0] # only get dq/da, throw dq/dw self.train_ops.append(self.a_grads)
