Develop Reference

Do we use different weights in Bidirectional LSTM for each batch?

For example this is one of the function which we need to call for each batch. Here it looks like different parameters are used for each batch. Is that correct? If it is then, why? Shouldn't we be using same parameters for whole training set?
def bidirectional_lstm(input_data, num_layers=3, rnn_size=200, keep_prob=0.6):
output = input_data
for layer in range(num_layers):
with tf.variable_scope('encoder_{}'.format(layer)):
cell_fw = tf.contrib.rnn.LSTMCell(rnn_size, initializer=tf.random_uniform_initializer(-0.1, 0.1, seed=2))
cell_fw = tf.contrib.rnn.DropoutWrapper(cell_fw, input_keep_prob = keep_prob)
cell_bw = tf.contrib.rnn.LSTMCell(rnn_size, initializer=tf.random_uniform_initializer(-0.1, 0.1, seed=2))
cell_bw = tf.contrib.rnn.DropoutWrapper(cell_bw, input_keep_prob = keep_prob)
outputs, states = tf.nn.bidirectional_dynamic_rnn(cell_fw,
cell_bw,
output,
dtype=tf.float32)
output = tf.concat(outputs,2)
return output
for batch_i, batch in enumerate(get_batches(X_train, batch_size)):
embeddings = tf.nn.embedding_lookup(word_embedding_matrix, batch)
output = bidirectional_lstm(embeddings)
print(output.shape)

I have figured out the issue in there. It turns out that we do use the same parameter and above code will give an error in second iteration saying that bidirectional kernel already exists. To fix this, we need to set, reuse=AUTO_REUSE while defining scope variable. Therefore, the line
with tf.variable_scope('encoder_{}'.format(layer)):
will become
with tf.variable_scope('encoder_{}'.format(layer),reuse=AUTO_REUSE):
Now we are using the same layers for each batch.

How to slice a tensor with None dimension in Tensorflow

I want to slice a tensor in "None" dimension.
For example,
tensor = tf.placeholder(tf.float32, shape=[None, None, 10], name="seq_holder")
sliced_tensor = tensor[:,1:,:] # it works well!
but
# Assume that tensor's shape will be [3,10, 10]
tensor = tf.placeholder(tf.float32, shape=[None, None, 10], name="seq_holder")
sliced_seq = tf.slice(tensor, [0,1,0],[3, 9, 10]) # it doens't work!
It is same that i get a message when i used another place_holder to feed size parameter for tf.slice().
The second methods gave me "Input size (depth of inputs) must be accessible via shape inference" error message.
I'd like to know what's different between two methods and what is more tensorflow-ish way.
[Edited]
Whole code is below
import tensorflow as tf
import numpy as np
print("Tensorflow for tests!")
vec_dim = 5
num_hidden = 10
# method 1
input_seq1 = np.random.random([3,7,vec_dim])
# method 2
input_seq2 = np.random.random([5,10,vec_dim])
shape_seq2 = [5,9,vec_dim]
# seq: [batch, seq_len]
seq = tf.placeholder(tf.float32, shape=[None, None, vec_dim], name="seq_holder")
# Method 1
sliced_seq = seq[:,1:,:]
# Method 2
seq_shape = tf.placeholder(tf.int32, shape=[3])
sliced_seq = tf.slice(seq,[0,0,0], seq_shape)
cell = tf.contrib.rnn.GRUCell(num_units=num_hidden)
init_state = cell.zero_state(tf.shape(seq)[0], tf.float32)
outputs, last_state = tf.nn.dynamic_rnn(cell, sliced_seq, initial_state=init_state)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
# method 1
# states = sess.run([sliced_seq], feed_dict={seq:input_seq1})
# print(states[0].shape)
# method 2
states = sess.run([sliced_seq], feed_dict={seq:input_seq2, seq_shape:shape_seq2})
print(states[0].shape)

Your problem is exactly described by issue #4590
The problem is that tf.nn.dynamic_rnn needs to know the size of the last dimension in the input (the "depth"). Unfortunately, as the issue points out, currently tf.slice cannot infer any output size if any of the slice ranges are not fully known at graph construction time; therefore, sliced_seq ends up having a shape (?, ?, ?).
In your case, the first issue is that you are using a placeholder of three elements to determine the size of the slice; this is not the best approach, since the last dimension should never change (even if you later pass vec_dim, it could cause errors). The easiest solution would be to turn seq_shape into a placeholder of size 2 (or even two separate placeholders), and then do the slicing like:
sliced_seq = seq[:seq_shape[0], :seq_shape[1], :]
For some reason, the NumPy-style indexing seems to have better shape inference capabilities, and this will preserve the size of the last dimension in sliced_seq.

How to feed back RNN output to input in tensorflow

In case where suppose I have a trained RNN (e.g. language model), and I want to see what it would generate on its own, how should I feed its output back to its input?
I read the following related questions:
TensorFlow using LSTMs for generating text
TensorFlow LSTM Generative Model
Theoretically it is clear to me, that in tensorflow we use truncated backpropagation, so we have to define the max step which we would like to "trace". Also we reserve a dimension for batches, therefore if I'd like to train a sine wave, I have to feed [None, num_step, 1] inputs.
The following code works:
tf.reset_default_graph()
n_samples=100
state_size=5
lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(state_size, forget_bias=1.)
def_x = np.sin(np.linspace(0, 10, n_samples))[None, :, None]
zero_x = np.zeros(n_samples)[None, :, None]
X = tf.placeholder_with_default(zero_x, [None, n_samples, 1])
output, last_states = tf.nn.dynamic_rnn(inputs=X, cell=lstm_cell, dtype=tf.float64)
pred = tf.contrib.layers.fully_connected(output, 1, activation_fn=tf.tanh)
Y = np.roll(def_x, 1)
loss = tf.reduce_sum(tf.pow(pred-Y, 2))/(2*n_samples)
opt = tf.train.AdamOptimizer().minimize(loss)
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
# Initial state run
plt.show(plt.plot(output.eval()[0]))
plt.plot(def_x.squeeze())
plt.show(plt.plot(pred.eval().squeeze()))
steps = 1001
for i in range(steps):
p, l, _= sess.run([pred, loss, opt])
The state size of the LSTM can be varied, also I experimented with feeding sine wave into the network and zeros, and in both cases it converged in ~500 iterations. So far I have understood that in this case the graph consists n_samples number of LSTM cells sharing their parameters, and it is only up to me that I feed input to them as a time series. However when generating samples the network is explicitly depending on its previous output - meaning that I cannot feed the unrolled model at once. I tried to compute the state and output at every step:
with tf.variable_scope('sine', reuse=True):
X_test = tf.placeholder(tf.float64)
X_reshaped = tf.reshape(X_test, [1, -1, 1])
output, last_states = tf.nn.dynamic_rnn(lstm_cell, X_reshaped, dtype=tf.float64)
pred = tf.contrib.layers.fully_connected(output, 1, activation_fn=tf.tanh)
test_vals = [0.]
for i in range(1000):
val = pred.eval({X_test:np.array(test_vals)[None, :, None]})
test_vals.append(val)
However in this model it seems that there is no continuity between the LSTM cells. What is going on here?
Do I have to initialize a zero array with i.e. 100 time steps, and assign each run's result into the array? Like feeding the network with this:
run 0: input_feed = [0, 0, 0 ... 0]; res1 = result
run 1: input_feed = [res1, 0, 0 ... 0]; res2 = result
run 1: input_feed = [res1, res2, 0 ... 0]; res3 = result
etc...
What to do if I want to use this trained network to use its own output as its input in the following time step?

If I understood you correctly, you want to find a way to feed the output of time step t as input to time step t+1, right? To do so, there is a relatively easy work around that you can use at test time:
Make sure your input placeholders can accept a dynamic sequence length, i.e. the size of the time dimension is None.
Make sure you are using tf.nn.dynamic_rnn (which you do in the posted example).
Pass the initial state into dynamic_rnn.
Then, at test time, you can loop through your sequence and feed each time step individually (i.e. max sequence length is 1). Additionally, you just have to carry over the internal state of the RNN. See pseudo code below (the variable names refer to your code snippet).
I.e., change the definition of the model to something like this:
lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(state_size, forget_bias=1.)
X = tf.placeholder_with_default(zero_x, [None, None, 1]) # [batch_size, seq_length, dimension of input]
batch_size = tf.shape(self.input_)[0]
initial_state = lstm_cell.zero_state(batch_size, dtype=tf.float32)
def_x = np.sin(np.linspace(0, 10, n_samples))[None, :, None]
zero_x = np.zeros(n_samples)[None, :, None]
output, last_states = tf.nn.dynamic_rnn(inputs=X, cell=lstm_cell, dtype=tf.float64,
initial_state=initial_state)
pred = tf.contrib.layers.fully_connected(output, 1, activation_fn=tf.tanh)
Then you can perform inference like so:
fetches = {'final_state': last_state,
'prediction': pred}
toy_initial_input = np.array([[[1]]]) # put suitable data here
seq_length = 20 # put whatever is reasonable here for you
# get the output for the first time step
feed_dict = {X: toy_initial_input}
eval_out = sess.run(fetches, feed_dict)
outputs = [eval_out['prediction']]
next_state = eval_out['final_state']
for i in range(1, seq_length):
feed_dict = {X: outputs[-1],
initial_state: next_state}
eval_out = sess.run(fetches, feed_dict)
outputs.append(eval_out['prediction'])
next_state = eval_out['final_state']
# outputs now contains the sequence you want
Note that this can also work for batches, however it can be a bit more complicated if you sequences of different lengths in the same batch.
If you want to perform this kind of prediction not only at test time, but also at training time, it is also possible to do, but a bit more complicated to implement.

You can use its own output (last state) as the next-step input (initial state).
One way to do this is to:
use zero-initialized variables as the input state at every time step
each time you completed a truncated sequence and got some output state, update the state variables with this output state you just got.
The second can be done by either:
fetching the states to python and feeding them back next time, as done in the ptb example in tensorflow/models
build an update op in the graph and add a dependency, as done in the ptb example in tensorpack.

I know I'm a bit late to the party but I think this gist could be useful:
https://gist.github.com/CharlieCodex/f494b27698157ec9a802bc231d8dcf31
It lets you autofeed the input through a filter and back into the network as input. To make shapes match up processing can be set as a tf.layers.Dense layer.
Please ask any questions!
Edit:
In your particular case, create a lambda which performs the processing of the dynamic_rnn outputs into your character vector space. Ex:
# if you have:
W = tf.Variable( ... )
B = tf.Variable( ... )
Yo, Ho = tf.nn.dynamic_rnn( cell , inputs , state )
logits = tf.matmul(W, Yo) + B
...
# use self_feeding_rnn as
process_yo = lambda Yo: tf.matmul(W, Yo) + B
Yo, Ho = self_feeding_rnn( cell, seed, initial_state, processing=process_yo)

TensorFlow: Remember LSTM state for next batch (stateful LSTM)

Given a trained LSTM model I want to perform inference for single timesteps, i.e. seq_length = 1 in the example below. After each timestep the internal LSTM (memory and hidden) states need to be remembered for the next 'batch'. For the very beginning of the inference the internal LSTM states init_c, init_h are computed given the input. These are then stored in a LSTMStateTuple object which is passed to the LSTM. During training this state is updated every timestep. However for inference I want the state to be saved in between batches, i.e. the initial states only need to be computed at the very beginning and after that the LSTM states should be saved after each 'batch' (n=1).
I found this related StackOverflow question: Tensorflow, best way to save state in RNNs?. However this only works if state_is_tuple=False, but this behavior is soon to be deprecated by TensorFlow (see rnn_cell.py). Keras seems to have a nice wrapper to make stateful LSTMs possible but I don't know the best way to achieve this in TensorFlow. This issue on the TensorFlow GitHub is also related to my question: https://github.com/tensorflow/tensorflow/issues/2838
Anyone good suggestions for building a stateful LSTM model?
inputs = tf.placeholder(tf.float32, shape=[None, seq_length, 84, 84], name="inputs")
targets = tf.placeholder(tf.float32, shape=[None, seq_length], name="targets")
num_lstm_layers = 2
with tf.variable_scope("LSTM") as scope:
lstm_cell = tf.nn.rnn_cell.LSTMCell(512, initializer=initializer, state_is_tuple=True)
self.lstm = tf.nn.rnn_cell.MultiRNNCell([lstm_cell] * num_lstm_layers, state_is_tuple=True)
init_c = # compute initial LSTM memory state using contents in placeholder 'inputs'
init_h = # compute initial LSTM hidden state using contents in placeholder 'inputs'
self.state = [tf.nn.rnn_cell.LSTMStateTuple(init_c, init_h)] * num_lstm_layers
outputs = []
for step in range(seq_length):
if step != 0:
scope.reuse_variables()
# CNN features, as input for LSTM
x_t = # ...
# LSTM step through time
output, self.state = self.lstm(x_t, self.state)
outputs.append(output)

I found out it was easiest to save the whole state for all layers in a placeholder.
init_state = np.zeros((num_layers, 2, batch_size, state_size))
...
state_placeholder = tf.placeholder(tf.float32, [num_layers, 2, batch_size, state_size])
Then unpack it and create a tuple of LSTMStateTuples before using the native tensorflow RNN Api.
l = tf.unpack(state_placeholder, axis=0)
rnn_tuple_state = tuple(
[tf.nn.rnn_cell.LSTMStateTuple(l[idx][0], l[idx][1])
for idx in range(num_layers)]
)
RNN passes in the API:
cell = tf.nn.rnn_cell.LSTMCell(state_size, state_is_tuple=True)
cell = tf.nn.rnn_cell.MultiRNNCell([cell]*num_layers, state_is_tuple=True)
outputs, state = tf.nn.dynamic_rnn(cell, x_input_batch, initial_state=rnn_tuple_state)
The state - variable will then be feeded to the next batch as a placeholder.

Tensorflow, best way to save state in RNNs? was actually my original question. The code bellow is how I use the state tuples.
with tf.variable_scope('decoder') as scope:
rnn_cell = tf.nn.rnn_cell.MultiRNNCell \
([
tf.nn.rnn_cell.LSTMCell(512, num_proj = 256, state_is_tuple = True),
tf.nn.rnn_cell.LSTMCell(512, num_proj = WORD_VEC_SIZE, state_is_tuple = True)
], state_is_tuple = True)
state = [[tf.zeros((BATCH_SIZE, sz)) for sz in sz_outer] for sz_outer in rnn_cell.state_size]
for t in range(TIME_STEPS):
if t:
last = y_[t - 1] if TRAINING else y[t - 1]
else:
last = tf.zeros((BATCH_SIZE, WORD_VEC_SIZE))
y[t] = tf.concat(1, (y[t], last))
y[t], state = rnn_cell(y[t], state)
scope.reuse_variables()
Rather than using tf.nn.rnn_cell.LSTMStateTuple I just create a lists of lists which works fine. In this example I am not saving the state. However you could easily have made state out of variables and just used assign to save the values.

Tensorflow convert array of tensors into single tensor

I'm building a LSTM RNN with Tensorflow that performs pixel-wise classification (or, maybe a better way to put it is, pixel-wise prediction?)
Bear with me as I explain the title.
The network looks like the following drawing...
The idea goes like this... an input image of size (200,200) is the input into a LSTM RNN of size (200,200,200). Each sequence output from the LSTM tensor vector (the pink boxes in the LSTM RNN) is fed into a MLP, and then the MLP makes a single output prediction -- ergo pixel-wise prediction (you can see how one input pixel generates one output "pixel"
The code looks like this (not all of the code, just parts that are needed):
...
n_input_x = 200
n_input_y = 200
x = tf.placeholder("float", [None, n_input_x, n_input_y])
y = tf.placeholder("float", [None, n_input_x, n_input_y])
def RNN(x):
x = tf.transpose(x, [1, 0, 2])
x = tf.reshape(x, [-1, n_input_x])
x = tf.split(0, n_steps, x)
lstm_cell = rnn_cell.BasicLSTMCell(n_hidden, forget_bias=1.0, state_is_tuple=True)
outputs, states = rnn.rnn(lstm_cell, x, dtype=tf.float32)
output_matrix = []
for i in xrange(200):
temp_vector = []
for j in xrange(200):
lstm_vector = outputs[j]
pixel_pred = multilayer_perceptron(lstm_vector, mlp_weights, mlp_biases)
temp_vector.append(pixel_pred)
output_matrix.append(temp_vector)
print i
return output_matrix
temp = RNN(x)
pred = tf.placeholder(temp, [None, n_input_x, n_input_y])
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(pred, y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
...
I have confirmed that the output of RNN -- that is, what is stored in temp is a 200x200 array of <tf.Tensor 'Softmax_39999:0' shape=(?, 1) dtype=float32>
As you can see, I place temp in a tf.placeholder of the same shape (None for the batch size ... or do I need this?)... and the program just exits as if it completed running. Ideally what I want to see when I debug and print pred is something like <tf.Tensor shape=(200,200)>
When I debug, the first time I execute pred = tf.placeholder(temp, [None, n_input_x, n_input_y]) I get TypeError: TypeErro...32>]].",) and then it returns and I try again, and it says Exception AttributeError: "'NoneType' object has no attribute 'path'" in <function _remove at 0x7f1ab77c26e0> ignored
EDIT I also now realize that I need to place the lines
lstm_cell = rnn_cell.BasicLSTMCell(n_hidden, forget_bias=1.0, state_is_tuple=True)
outputs, states = rnn.rnn(lstm_cell, x, dtype=tf.float32)
Inside the first loop so that new 2D LSTM RNN are generated, however I'm getting an error about variable reusing ValueError: Variable RNN/BasicLSTMCell/Linear/Matrix does not exist, disallowed. Did you mean to set reuse=None in VarScope?
So in other words, is it isn't auto incrementing the RNN tensor name?

A more convenient way to report shapes is with tf.shape(). In your case:
size1 = tf.shape(temp)
sess = tf.Session()
size1_fetched = sess.run(size1, feed_dict = your_feed_dict)
That way, the size1_fetched is something like you would get from NumPy. Moreover, also your sizes for that particular feed_dict are given. For example, your [None, 200, 200] Tensor would be [64, 200, 200]
Another question: why do you have the placeholder in between your flow-graph? Will you later on feed pre-defined images feature-maps?