This is a variational autoencoder network, I have to define a sampling method to generate latent z, I thinks it might be something wrong with this. This py file is doing training, the other py file is doing predicting online, so I need to save the keras model, there is nothing wrong with saving model, but when I load model from 'h5' file, it shows an error:
NameError: name 'latent_dim' is not defined
The following is code:
df_test = df[df['label']==cluster_num].iloc[:,:data_num.shape[1]]
data_scale_ = preprocessing.StandardScaler().fit(df_test.values)
data_num_ = data_scale.transform(df_test.values)
models_deep_learning_scaler.append(data_scale_)
batch_size = data_num_.shape[0]//10
original_dim = data_num_.shape[1]
latent_dim = data_num_.shape[1]*2
intermediate_dim = data_num_.shape[1]*10
nb_epoch = 1
epsilon_std = 0.001
x = Input(shape=(original_dim,))
init_drop = Dropout(0.2, input_shape=(original_dim,))(x)
h = Dense(intermediate_dim, activation='relu')(init_drop)
z_mean = Dense(latent_dim)(h)
z_log_var = Dense(latent_dim)(h)
def sampling(args):
z_mean, z_log_var = args
epsilon = K.random_normal(shape=(latent_dim,), mean=0.,
std=epsilon_std)
return z_mean + K.exp(z_log_var / 2) * epsilon
# note that "output_shape" isn't necessary with the TensorFlow backend
z = Lambda(sampling, output_shape=(latent_dim,))([z_mean, z_log_var])
# we instantiate these layers separately so as to reuse them later
decoder_h = Dense(intermediate_dim, activation='relu')
decoder_mean = Dense(original_dim, activation='linear')
h_decoded = decoder_h(z)
x_decoded_mean = decoder_mean(h_decoded)
def vae_loss(x, x_decoded_mean):
xent_loss = original_dim * objectives.mae(x, x_decoded_mean)
kl_loss = - 0.5 * K.sum(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
return xent_loss + kl_loss
vae = Model(x, x_decoded_mean)
vae.compile(optimizer=Adam(lr=0.01), loss=vae_loss)
train_ratio = 0.95
train_num = int(data_num_.shape[0]*train_ratio)
x_train = data_num_[:train_num,:]
x_test = data_num_[train_num:,:]
vae.fit(x_train, x_train,
shuffle=True,
nb_epoch=nb_epoch,
batch_size=batch_size,
validation_data=(x_test, x_test))
vae.save('./models/deep_learning_'+str(cluster_num)+'.h5')
del vae
from keras.models import load_model
vae = load_model('./models/deep_learning_'+str(cluster_num)+'.h5')
It shows error:
NameError: name 'latent_dim' is not defined
For variational loss you are using many variable not known by Keras module. You need to pass them through custom_objects param of load_model function.
In your case:
vae.save('./vae_'+str(cluster_num)+'.h5')
vae.summary()
del vae
from keras.models import load_model
vae = load_model('./vae_'+str(cluster_num)+'.h5', custom_objects={'latent_dim': latent_dim, 'epsilon_std': epsilon_std, 'vae_loss': vae_loss})
vae.summary()
If you load model (.h5) file in your new py file, you can use load_model('/.h5', compile = False).
Because you do not need to any custom objects (i.e loss function or latent_dim, etc) in prediction step.
Related
I want to fix problem in PyTorch.
I wrote the following code that is learning sine functions as tutorial.
import torch
from torch import nn
from torch import optim
from torch.autograd import Variable as V
from torch.utils.data import TensorDataset, DataLoader
import numpy as np
# y=sin(x1)
numTrain = 512
numTest = 128
noiseScale = 0.01
PI2 = 3.1415 * 2
X_train = np.random.rand(numTrain,1) * PI2
y_train = np.sin(X_train) + np.random.randn(numTrain,1) * noiseScale + 1.5
X_test = np.random.rand(numTest,1) * PI2
y_test = np.sin(X_test) + np.random.randn(numTest,1) * noiseScale
# Construct DataSet
X_trainT = torch.Tensor(X_train)
y_trainT = torch.Tensor(y_train)
X_testT = torch.Tensor(X_test)
y_testT = torch.Tensor(y_test)
ds_train = TensorDataset(X_trainT, y_trainT)
ds_test = TensorDataset(X_testT, y_testT)
# Construct DataLoader
loader_train = DataLoader(ds_train, batch_size=64, shuffle=True)
loader_test = DataLoader(ds_test, batch_size=64, shuffle=False)
# Construct network
net = nn.Sequential(
nn.Linear(1,10),
nn.ReLU(),
nn.BatchNorm1d(10),
nn.Linear(10,5),
nn.ReLU(),
nn.BatchNorm1d(5),
nn.Linear(5,1),
)
optimizer = optim.Adam(net.parameters())
loss_fn = nn.SmoothL1Loss()
# Training
losses = []
net.train()
for epoc in range(100):
for data, target in loader_train:
y_pred = net(data)
loss = loss_fn(target,y_pred)
optimizer.zero_grad()
loss.backward()
optimizer.step()
losses.append(loss.data)
# evaluation
%matplotlib inline
from matplotlib import pyplot as plt
#plt.plot(losses)
plt.scatter(X_train, y_train)
net.eval()
sinsX = []
sinsY = []
for t in range(128):
x = t/128 * PI2
output = net(V(torch.Tensor([x])))
sinsX.append(x)
sinsY.append(output.detach().numpy())
plt.scatter(sinsX,sinsY)
Training is done without error, But the next line caused an error, "expected 2D or 3D input (got 1D input)"
output = net(V(torch.Tensor([x])))
This error doesn't occur if it is without BatchNorm1d().
I feel strange because the input is 1D.
How to fix it?
Thanks.
Update: How did I fix
arr = np.array([x])
output = net(V(torch.Tensor(arr[None,...])))
When working with 1D signals, pyTorch actually expects a 2D tensors: the first dimension is the "mini-batch" dimension. Therefore, you should evaluate your net on a batch with one 1D signal:
output - net(V(torch.Tensor([x[None, ...]]))
Make sure you set your net to "eval" mode before evaluating it:
net.eval()
I built a two layered LSTM model(keras model) for a movie review dataset from kaggle : Dataset
While training the model, every epoch was giving the same accuracy of 0.5098.
Then I thought it might not be learning the long distance dependencies.Then instead of LSTM I used bidirectional LSTM. But, still model's accuracy while training was 0.5098 for every epoch. I trained the model for 8 hours/35 epochs on CPU. Then I stopped training.
Code:
import pandas as pd
from sentiment_utils import *
import keras
import keras.backend as k
import numpy as np
train_data = pd.read_table('train.tsv')
X_train = train_data.iloc[:,2]
Y_train = train_data.iloc[:,3]
from sklearn.preprocessing import OneHotEncoder
Y_train = Y_train.reshape(Y_train.shape[0],1)
ohe = OneHotEncoder(categorical_features=[0])
Y_train = ohe.fit_transform(Y_train).toarray()
maxLen = len(max(X_train, key=len).split())
words_to_index, index_to_words, word_to_vec_map = read_glove_vectors("glove/glove.6B.50d.txt")
m = X_train.shape[0]
def read_glove_vectors(path):
with open(path, encoding='utf8') as f:
words = set()
word_to_vec_map = {}
for line in f:
line = line.strip().split()
cur_word = line[0]
words.add(cur_word)
word_to_vec_map[cur_word] = np.array(line[1:], dtype=np.float64)
i = 1
words_to_index = {}
index_to_words = {}
for w in sorted(words):
words_to_index[w] = i
index_to_words[i] = w
i = i + 1
return words_to_index, index_to_words, word_to_vec_map
def sentance_to_indices(X_train, words_to_index, maxLen, dash_index_list, keys):
m = X_train.shape[0]
X_indices = np.zeros((m, maxLen))
for i in range(m):
if i in dash_index_list:
continue
sentance_words = X_train[i].lower().strip().split()
j = 0
for word in sentance_words:
if word in keys:
X_indices[i, j] = words_to_index[word]
j += 1
return X_indices
def pretrained_embedding_layer(word_to_vec_map, words_to_index):
emb_dim = word_to_vec_map['pen'].shape[0]
vocab_size = len(words_to_index) + 1
emb_matrix = np.zeros((vocab_size, emb_dim))
for word, index in words_to_index.items():
emb_matrix[index, :] = word_to_vec_map[word]
emb_layer= keras.layers.embeddings.Embedding(vocab_size, emb_dim, trainable= False)
emb_layer.build((None,))
emb_layer.set_weights([emb_matrix])
return emb_layer
def get_model(input_shape, word_to_vec_map, words_to_index):
sentance_indices = keras.layers.Input(shape = input_shape, dtype='int32')
embedding_layer = pretrained_embedding_layer(word_to_vec_map, words_to_index)
embeddings = embedding_layer(sentance_indices)
X = keras.layers.Bidirectional(keras.layers.LSTM(128, return_sequences=True))(embeddings)
X = keras.layers.Dropout(0.5)(X)
X = keras.layers.Bidirectional(keras.layers.LSTM(128, return_sequences=True))(X)
X = keras.layers.Dropout(0.5)(X)
X = keras.layers.Bidirectional(keras.layers.LSTM(128, return_sequences=False))(X)
X = keras.layers.Dropout(0.5)(X)
X = keras.layers.Dense(5)(X)
X = keras.layers.Activation('softmax')(X)
model = keras.models.Model(sentance_indices, X)
return model
model = get_model((maxLen,), word_to_vec_map,words_to_index)
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
dash_index_list = []
for i in range(m):
if '-' in X_train[i]:
dash_index_list.append(i)
keys = []
for key in word_to_vec_map.keys():
keys.append(key)
X_train_indices = sentance_to_indices(X_train, words_to_index, maxLen, dash_index_list, keys)
model.fit(X_train_indices, Y_train, epochs = 50, batch_size = 32, shuffle=True)
I think the way you defined the model architecture doesn't make sense! Try looking at this example on IMDB movie reviews with LSTM on Keras github repo: Trains an LSTM model on the IMDB sentiment classification task.
I have been trying to implement a CNN on the CIFAR-10 dataset for a few days and my test set accuracy does not seem to go beyond the 10% and the error just hang around 69.07733. I have tweaking the model and few days but in vain. I haven't been able to spot out where I am going wrong. Please help me recognise the fault in the model. Here is the code for it:
import os
import sys
import pickle
import tensorflow as tf
import numpy as np
from matplotlib import pyplot as plt
data_root = './cifar-10-batches-py'
train_data = np.ndarray(shape=(50000,3072), dtype=np.float32)
train_labels = np.ndarray(shape=(50000), dtype=np.float32)
num_images = 0
test_data = np.ndarray(shape=(10000,3072),dtype = np.float32)
test_labels = np.ndarray(shape=(10000),dtype=np.float32)
meta_data = {}
for file in os.listdir(data_root):
file_path = os.path.join(data_root,file)
with open(file_path,'rb') as f:
temp = pickle.load(f,encoding ='bytes')
if file == 'batches.meta':
for i,j in enumerate(temp[b'label_names']):
meta_data[i] = j
if 'data_batch_' in file:
for i in range(10000):
train_data[num_images,:] = temp[b'data'][i]
train_labels[num_images] = temp[b'labels'][i]
num_images += 1
if 'test_batch' in file:
for i in range(10000):
test_data[i,:] = temp[b'data'][i]
test_labels[i] = temp[b'labels'][i]
'''
print('meta: \n',meta_data)
train_data = train_data.reshape(50000,3,32,32).transpose(0,2,3,1)
print('\ntrain data: \n',train_data.shape,'\nLabels: \n',train_labels[0])
print('\ntest data: \n',test_data[0].shape,'\nLabels: \n',train_labels[0])'''
#accuracy function acc = (no. of correct prediction/total attempts) * 100
def accuracy(predictions, labels):
return (100 * (np.sum(np.argmax(predictions,1)== np.argmax(labels, 1))/predictions.shape[0]))
#reformat the data
def reformat(data,labels):
data = data.reshape(data.shape[0],3,32,32).transpose(0,2,3,1).astype(np.float32)
labels = (np.arange(10) == labels[:,None]).astype(np.float32)
return data,labels
train_data, train_labels = reformat(train_data,train_labels)
test_data, test_labels = reformat(test_data, test_labels)
print ('Train ',train_data[0][1])
plt.axis("off")
plt.imshow(train_data[1], interpolation = 'nearest')
plt.savefig("1.png")
plt.show()
'''
print("Train: \n",train_data.shape,test_data[0],"\nLabels: \n",train_labels.shape,train_labels[:11])
print("Test: \n",test_data.shape,test_data[0],"\nLabels: \n",test_labels.shape,test_labels[:11])'''
image_size = 32
num_channels = 3
batch_size = 30
patch_size = 5
depth = 64
num_hidden = 256
num_labels = 10
graph = tf.Graph()
with graph.as_default():
#input data and labels
train_input = tf.placeholder(tf.float32,shape=(batch_size,image_size,image_size,num_channels))
train_output = tf.placeholder(tf.float32,shape=(batch_size,num_labels))
test_input = tf.constant(test_data)
#layer weights and biases
layer_1_weights = tf.Variable(tf.truncated_normal([patch_size,patch_size,num_channels,depth]))
layer_1_biases = tf.Variable(tf.zeros([depth]))
layer_2_weights = tf.Variable(tf.truncated_normal([patch_size,patch_size,depth,depth]))
layer_2_biases = tf.Variable(tf.constant(0.1, shape=[depth]))
layer_3_weights = tf.Variable(tf.truncated_normal([64*64, num_hidden]))
layer_3_biases = tf.Variable(tf.constant(0.1, shape=[num_hidden]))
layer_4_weights = tf.Variable(tf.truncated_normal([num_hidden, num_labels]))
layer_4_biases = tf.Variable(tf.constant(0.1, shape=[num_labels]))
def convnet(data):
conv_1 = tf.nn.conv2d(data, layer_1_weights,[1,1,1,1], padding = 'SAME')
hidden_1 = tf.nn.relu(conv_1+layer_1_biases)
norm_1 = tf.nn.lrn(hidden_1, 4, bias=1.0, alpha=0.001 / 9.0, beta=0.75)
pool_1 = tf.nn.max_pool(norm_1,[1,2,2,1],[1,2,2,1], padding ='SAME')
conv_2 = tf.nn.conv2d(pool_1,layer_2_weights,[1,1,1,1], padding = 'SAME')
hidden_2 = tf.nn.relu(conv_2+layer_2_biases)
norm_2 = tf.nn.lrn(hidden_2, 4, bias=1.0, alpha=0.001 / 9.0, beta=0.75)
pool_2 = tf.nn.max_pool(norm_2,[1,2,2,1],[1,2,2,1], padding ='SAME')
shape = pool_2.get_shape().as_list()
hidd2_trans = tf.reshape(pool_2,[shape[0],shape[1]*shape[2]*shape[3]])
hidden_3 = tf.nn.relu(tf.matmul(hidd2_trans,layer_3_weights) + layer_3_biases)
return tf.nn.relu(tf.matmul(hidden_3,layer_4_weights) + layer_4_biases)
logits = convnet(train_input)
loss = tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(labels=train_output, logits = logits))
optimizer = tf.train.AdamOptimizer(1e-4).minimize(loss)
train_prediction = tf.nn.softmax(logits)
test_prediction = tf.nn.softmax(convnet(test_input))
num_steps = 100000
with tf.Session(graph=graph) as session:
tf.global_variables_initializer().run()
print('Initialized \n')
for step in range(num_steps):
offset = (step * batch_size) % (train_labels.shape[0] - batch_size)
batch = train_data[offset:(offset+batch_size),:,:,:]
batch_labels = train_labels[offset:(offset+batch_size),:]
feed_dict ={train_input: batch, train_output: batch_labels}
_,l,prediction = session.run([optimizer, loss, train_prediction], feed_dict = feed_dict)
if (step % 500 == 0):
print("Loss at step %d: %f" %(step, l))
print("Accuracy: %f" %(accuracy(prediction, batch_labels)))
print("Test accuracy: %f" %(accuracy(session.run(test_prediction), test_labels)))
On a first glance I would say the initialization of the CNN is the culprit. A convnet is an optimization algorithm in a highly non-convex space and therefore depends a lot on careful initialization to not get stuck on local minima or saddle points. Look at xavier initialization for an example on how to fix that.
Example Code:
W = tf.get_variable("W", shape=[784, 256],
initializer=tf.contrib.layers.xavier_initializer())
Problem is your network is having very high depth(number of filters = 64 for both layers). Also, you are training the network from scratch. And your dataset of CIFAR10 (50000 images) is very little. Moreover, each CIFAR10 image is only 32x32x3 size.
Couple of alternatives what I can suggest you is to retrain a pre-trained model, i.e do transfer learning.
Other better alternative is to reduce the number of filters in each layer. In this way, you will be able to train the model from scratch and also it will be faster. (Assuming you don't have GPU).
Next you are making use of local response normalization. I would suggest you to remove this layer and do mean normalization in pre-processing step.
Next, if you feel the learning is not picking up at all, try increasing the learning rate a little and see.
Lastly, just to reduce some operation in your code, you are reshaping your tensor and then doing transpose in many places like this:
data.reshape(data.shape[0],3,32,32).transpose(0,2,3,1)
Why not directly reshape it to something like this?
data.reshape(data.shape[0], 32, 32, 3)
Hope the answer helps you.
I have a data set which contains a list of stock prices. I need to use the tensorflow and python to predict the close price.
Q1: I have the following code which takes the first 2000 records as training and 2001 to 20000 records as test but I don't know how to change the code to do the prediction of the close price of today and 1 day later??? Please advise!
#!/usr/bin/env python2
import numpy as np
import pandas as pd
import tensorflow as tf
import matplotlib.pyplot as plt
def feature_scaling(input_pd, scaling_meathod):
if scaling_meathod == 'z-score':
scaled_pd = (input_pd - input_pd.mean()) / input_pd.std()
elif scaling_meathod == 'min-max':
scaled_pd = (input_pd - input_pd.min()) / (input_pd.max() -
input_pd.min())
return scaled_pd
def input_reshape(input_pd, start, end, batch_size, batch_shift, n_features):
temp_pd = input_pd[start-1: end+batch_size-1]
output_pd = map(lambda y : temp_pd[y:y+batch_size], xrange(0, end-start+1, batch_shift))
output_temp = map(lambda x : np.array(output_pd[x]).reshape([-1]), xrange(len(output_pd)))
output = np.reshape(output_temp, [-1, batch_size, n_features])
return output
def target_reshape(input_pd, start, end, batch_size, batch_shift, n_step_ahead, m_steps_pred):
temp_pd = input_pd[start+batch_size+n_step_ahead-2: end+batch_size+n_step_ahead+m_steps_pred-2]
print temp_pd
output_pd = map(lambda y : temp_pd[y:y+m_steps_pred], xrange(0, end-start+1, batch_shift))
output_temp = map(lambda x : np.array(output_pd[x]).reshape([-1]), xrange(len(output_pd)))
output = np.reshape(output_temp, [-1,1])
return output
def lstm(input, n_inputs, n_steps, n_of_layers, scope_name):
num_layers = n_of_layers
input = tf.transpose(input,[1, 0, 2])
input = tf.reshape(input,[-1, n_inputs])
input = tf.split(0, n_steps, input)
with tf.variable_scope(scope_name):
cell = tf.nn.rnn_cell.BasicLSTMCell(num_units=n_inputs)
cell = tf.nn.rnn_cell.MultiRNNCell([cell]*num_layers)
output, state = tf.nn.rnn(cell, input, dtype=tf.float32) yi1
output = output[-1]
return output
feature_to_input = ['open price', 'highest price', 'lowest price', 'close price','turnover', 'volume','mean price']
feature_to_predict = ['close price']
feature_to_scale = ['volume']
sacling_meathod = 'min-max'
train_start = 1
train_end = 1000
test_start = 1001
test_end = 20000
batch_size = 100
batch_shift = 1
n_step_ahead = 1
m_steps_pred = 1
n_features = len(feature_to_input)
lstm_scope_name = 'lstm_prediction'
n_lstm_layers = 1
n_pred_class = 1
learning_rate = 0.1
EPOCHS = 1000
PRINT_STEP = 100
read_data_pd = pd.read_csv('./stock_price.csv')
temp_pd = feature_scaling(input_pd[feature_to_scale],sacling_meathod)
input_pd[feature_to_scale] = temp_pd
train_input_temp_pd = input_pd[feature_to_input]
train_input_nparr = input_reshape(train_input_temp_pd,
train_start, train_end, batch_size, batch_shift, n_features)
train_target_temp_pd = input_pd[feature_to_predict]
train_target_nparr = target_reshape(train_target_temp_pd, train_start, train_end, batch_size, batch_shift, n_step_ahead, m_steps_pred)
test_input_temp_pd = input_pd[feature_to_input]
test_input_nparr = input_reshape(test_input_temp_pd, test_start, test_end, batch_size, batch_shift, n_features)
test_target_temp_pd = input_pd[feature_to_predict]
test_target_nparr = target_reshape(test_target_temp_pd, test_start, test_end, batch_size, batch_shift, n_step_ahead, m_steps_pred)
tf.reset_default_graph()
x_ = tf.placeholder(tf.float32, [None, batch_size, n_features])
y_ = tf.placeholder(tf.float32, [None, 1])
lstm_output = lstm(x_, n_features, batch_size, n_lstm_layers, lstm_scope_name)
W = tf.Variable(tf.random_normal([n_features, n_pred_class]))
b = tf.Variable(tf.random_normal([n_pred_class]))
y = tf.matmul(lstm_output, W) + b
cost_func = tf.reduce_mean(tf.square(y - y_))
train_op = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost_func)
optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step=global_step)
init = tf.initialize_all_variables()
with tf.Session() as sess:
sess.run(init)
for ii in range(EPOCHS):
sess.run(train_op, feed_dict={x_:train_input_nparr, y_:train_target_nparr})
if ii % PRINT_STEP == 0:
cost = sess.run(cost_func, feed_dict={x_:train_input_nparr, y_:train_target_nparr})
print 'iteration =', ii, 'training cost:', cost
Very simply, prediction (a.k.a. scoring or inference) comes from running the input through only the forward pass, and collecting the score for each input vector. It's the same process flow as testing. The difference is the four stages of model use:
Train: learn from the training data set; adjust weights as needed.
Test: evaluate the model's performance; if accuracy has converged, stop training.
Validate: evaluate the accuracy of the trained model. If it doesn't meet acceptance criteria, change something and start over with the training.
Predict: you've passed validation -- release the model for use by the intended application.
All four steps follow the same forward logic flow; training includes back-propagation; the others do not. Simply follow the forward-only process, and you'll get the result form you need.
I worry about your data partition: only 10% for training, 90% for testing, and none for validation. A more typical split is 50-30-20, or something in that general area.
Q-1 : You should change your LSTM parameter to return a sequence of size two which will be prediction for that day and the day after.
Q-2 it's clearly that your model is underfitting the data, which is so obvious with your 10% train 90% test data ! You should more equilibrated ratio as suggested in the previous answer.
I'm having a difficult time visualizing what this Tensorflow class creates. I want to implement a LSTM RNN that handles 3D data.
class Grid3LSTMCell(GridRNNCell):
"""3D BasicLSTM cell
This creates a 2D cell which receives input and gives output in the first dimension.
The first dimension can optionally be non-recurrent if `non_recurrent_fn` is specified.
The second and third dimensions are LSTM.
"""
def __init__(self, num_units, tied=False, non_recurrent_fn=None,
use_peepholes=False, forget_bias=1.0):
super(Grid3LSTMCell, self).__init__(num_units=num_units, num_dims=3,
input_dims=0, output_dims=0, priority_dims=0, tied=tied,
non_recurrent_dims=None if non_recurrent_fn is None else 0,
cell_fn=lambda n, i: rnn_cell.LSTMCell(
num_units=n, input_size=i, forget_bias=forget_bias,
use_peepholes=use_peepholes),
non_recurrent_fn=non_recurrent_fn)
The class is found in `from tensorflow.contrib.grid_rnn.python.ops import grid_rnn_cell`.
This is difficult to explain, so I've provided a drawing. Here is what I want it to do...
However the comment sounds like it isn't doing this. The comment makes it sound like the RNN is still a flat RNN, where the first dimension is outputting to, what is commonly called, the outputs variable (see below). The second dimension is outputting to the next step in the RNN, and the third dimension is outputting to the next hidden layer.
outputs, states = rnn.rnn(lstm_cell, x, dtype=tf.float32)
If this is the case, what is the point in having the first and second dimensions? Aren't they essentially the same thing? The BasicLSTMCell sends the output to the next step into outputs -- in other words they are one in the same.
Clarity?
For reference, here is my example code...
import tensorflow as tf
from tensorflow.python.ops import rnn, rnn_cell
from tensorflow.contrib.grid_rnn.python.ops import grid_rnn_cell
import numpy as np
#define parameters
learning_rate = 0.01
batch_size = 2
n_input_x = 10
n_input_y = 10
n_input_z = 10
n_hidden = 128
n_classes = 2
n_output = n_input_x * n_classes
x = tf.placeholder("float", [n_input_x, n_input_y, n_input_z])
y = tf.placeholder("float", [n_input_x, n_input_y, n_input_z, n_classes])
weights = {}
biases = {}
for i in xrange(n_input_y * n_input_z):
weights[i] = tf.Variable(tf.random_normal([n_hidden, n_output]))
biases[i] = tf.Variable(tf.random_normal([n_output]))
#generate random data
input_data = np.random.rand(n_input_x, n_input_y, n_input_z)
ground_truth = np.random.rand(n_input_x, n_input_y, n_input_z, n_classes)
#build GridLSTM
def GridLSTM_network(x):
x = tf.reshape(x, [-1,n_input_x])
x = tf.split(0, n_input_y * n_input_z, x)
lstm_cell = grid_rnn_cell.Grid3LSTMCell(n_hidden)
outputs, states = rnn.rnn(lstm_cell, x, dtype=tf.float32)
output = []
for i in xrange(n_input_y * n_input_z):
output.append(tf.matmul(outputs[i], weights[i]) + biases[i])
return output
#initialize network, cost, optimizer and all variables
pred = GridLSTM_network(x)
# import pdb
# pdb.set_trace()
pred = tf.pack(pred)
pred = tf.transpose(pred,[1,0,2])
pred= tf.reshape(pred, [-1, n_input_x, n_input_y, n_input_z, n_classes])
temp_pred = tf.reshape(pred, [-1,n_classes])
temp_y = tf.reshape(y,[-1, n_classes])
cost = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(temp_pred, temp_y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
# Evaluate model
correct_pred = tf.equal(0,tf.cast(tf.sub(tf.nn.sigmoid(temp_pred),temp_y), tf.int32))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
# Initializing the variables
init = tf.initialize_all_variables()
# Launch the graph
with tf.Session() as sess:
sess.run(init)
step = 0
while 1:
print step
step = step + 1
# pdb.set_trace
sess.run(optimizer, feed_dict={x: input_data, y: ground_truth})