generator.py

import tensorflow as tf
from tensorflow.python.ops import tensor_array_ops, control_flow_ops


class Generator(object):
    def __init__(self, num_emb, batch_size, emb_dim, hidden_dim,
                 sequence_length, start_token,
                 learning_rate=1e-3, reward_gamma=0.95, name="generator", dropout_rate=0.5):
        self.num_emb = num_emb
        # self.batch_size = batch_size
        self.emb_dim = emb_dim
        self.hidden_dim = hidden_dim
        self.sequence_length = sequence_length
        self.learning_rate = tf.Variable(float(learning_rate), trainable=False)
        self.reward_gamma = reward_gamma
        self.g_params = []
        self.d_params = []
        self.temperature = 1.0
        self.create_recurrent_unit = self.create_recurrent_unit_LSTM
        self.name = name
        self.dropout_keep_rate = tf.Variable(float(1.0), trainable=False)
        self.dropout_on = self.dropout_keep_rate.assign(dropout_rate)
        self.dropout_off = self.dropout_keep_rate.assign(1.0)
        self.expected_reward = tf.Variable(tf.zeros([self.sequence_length]))

        with tf.variable_scope('generator'):
            self.g_embeddings = tf.Variable(self.init_matrix([self.num_emb, self.emb_dim]))
            self.g_params.append(self.g_embeddings)
            self.g_recurrent_unit = self.create_recurrent_unit(self.g_params)  # maps h_tm1 to h_t for generator
            self.g_output_unit = self.create_output_unit(self.g_params)  # maps h_t to o_t (output token logits)

        # placeholder definition
        self.x = tf.placeholder(tf.int32, shape=[None, self.sequence_length]) # sequence of tokens generated by generator
        self.rewards = tf.placeholder(tf.float32, shape=[None, self.sequence_length, self.num_emb]) # get from rollout policy and discriminator
        batch_size_var = tf.Variable(batch_size, trainable=False, dtype=tf.int32)
        self.batch_size = tf.shape(self.x)[0]
        self.start_token = tf.ones(shape=[batch_size_var], dtype=tf.int32) * start_token
        batch_size = tf.cast(self.batch_size, tf.float32)
        # processed for batch
        with tf.device("/cpu:0"):
            one_hot_x = tf.one_hot(self.x, self.num_emb, 1.0, 0.0)
            self.processed_x = tf.transpose(one_hot_x, perm=[1, 0, 2])  # seq_length x batch_size x vocab_size


        # Initial states
        self.h0 = tf.zeros([self.batch_size, self.hidden_dim])
        self.h0 = tf.stack([self.h0, self.h0])

        gen_o = tensor_array_ops.TensorArray(dtype=tf.float32, size=self.sequence_length,
                                             dynamic_size=False, infer_shape=True)
        gen_x = tensor_array_ops.TensorArray(dtype=tf.int32, size=self.sequence_length,
                                             dynamic_size=False, infer_shape=True)

        def _g_recurrence(i, x_t, h_tm1, gen_o, gen_x):
            h_t = self.g_recurrent_unit(x_t, h_tm1)  # hidden_memory_tuple
            o_t = self.g_output_unit(h_t)  # batch x vocab , logits not prob
            log_prob = tf.nn.log_softmax(o_t)
            next_token = tf.cast(tf.reshape(tf.multinomial(log_prob, 1), [-1]), tf.int32)
            x_tp1 = tf.one_hot(next_token, self.num_emb, 1.0, 0.0)
            x_tp1 = tf.matmul(x_tp1, self.g_embeddings)

            gen_o = gen_o.write(i, tf.reduce_sum(tf.multiply(tf.one_hot(next_token, self.num_emb, 1.0, 0.0),
                                                             tf.nn.softmax(o_t)), 1))  # [batch_size] , prob
            gen_x = gen_x.write(i, next_token)  # indices, batch_size
            return i + 1, x_tp1, h_t, gen_o, gen_x

        _, _, _, self.gen_o, self.gen_x = control_flow_ops.while_loop(
            cond=lambda i, _1, _2, _3, _4: i < self.sequence_length,
            body=_g_recurrence,
            loop_vars=(tf.constant(0, dtype=tf.int32),
                       tf.nn.embedding_lookup(self.g_embeddings, self.start_token),
                       tf.zeros([2, batch_size_var, self.hidden_dim]),
                       gen_o, gen_x))

        self.gen_x = self.gen_x.stack()  # seq_length x batch_size
        self.gen_x = tf.transpose(self.gen_x, perm=[1, 0])  # batch_size x seq_length

        # supervised pretraining for generator
        g_predictions = tensor_array_ops.TensorArray(
            dtype=tf.float32, size=self.sequence_length,
            dynamic_size=False, infer_shape=True)
        log_predictions = tensor_array_ops.TensorArray(
            dtype=tf.float32, size=self.sequence_length,
            dynamic_size=False, infer_shape=True)

        ta_emb_x = tensor_array_ops.TensorArray(
            dtype=tf.float32, size=self.sequence_length)
        ta_emb_x = ta_emb_x.unstack(self.processed_x)

        def _pretrain_recurrence(i, x_t, h_tm1, g_predictions, log_predictions):
            h_t = self.g_recurrent_unit(x_t, h_tm1)
            o_t = self.g_output_unit(h_t)
            log_prob = tf.nn.log_softmax(o_t)
            o_t = tf.nn.softmax(o_t)
            g_predictions = g_predictions.write(i, o_t)  # batch x vocab_size
            x_tp1 = ta_emb_x.read(i)
            log_predictions = log_predictions.write(i, log_prob)
            x_tp1 = tf.matmul(x_tp1, self.g_embeddings)
            return i + 1, x_tp1, h_t, g_predictions, log_predictions

        _, _, _, self.g_predictions, self.log_predictions = control_flow_ops.while_loop(
            cond=lambda i, _1, _2, _3, _4: i < self.sequence_length,
            body=_pretrain_recurrence,
            loop_vars=(tf.constant(0, dtype=tf.int32),
                       tf.nn.embedding_lookup(self.g_embeddings, tf.ones(shape=[tf.shape(self.x)[0]], dtype=tf.int32) * start_token),
                       self.h0, g_predictions, log_predictions))

        self.g_predictions = tf.transpose(self.g_predictions.stack(), perm=[1, 0, 2])  # batch_size x seq_length x vocab_size
        self.g_prediction = tf.reduce_sum(self.g_predictions * tf.one_hot(self.x, self.num_emb, 1.0, 0.0), axis=-1)
        self.log_predictions = tf.transpose(self.log_predictions.stack(), perm=[1, 0, 2])
        self.self_log_likelihood = tf.cumsum(tf.reduce_sum(self.log_predictions * one_hot_x, axis=-1), exclusive=True, axis=-1)
        self.med_log_likelihood = tf.cumsum(tf.reduce_sum(self.rewards * one_hot_x, axis=-1), exclusive=True, axis=-1)
        # pretraining loss
        self.likelihood_loss = -tf.reduce_sum(
            tf.one_hot(tf.to_int32(tf.reshape(self.x, [-1])), self.num_emb, 1.0, 0.0) *
                tf.reshape(self.log_predictions, [-1, self.num_emb])
        ) / (self.sequence_length * batch_size)

        # training updates
        pretrain_opt = self.g_optimizer(self.learning_rate, beta1=0.9, beta2=0.95)


        self.likelihood_updates = pretrain_opt.minimize(self.likelihood_loss, var_list=self.g_params)

        #######################################################################################################
        #  Unsupervised Training
        #######################################################################################################
        self.g_loss = -tf.reduce_sum(
            self.g_predictions * (self.rewards - self.log_predictions)
        ) / batch_size / sequence_length
        g_opt = self.g_optimizer(self.learning_rate, beta1=0.9, beta2=0.95)

        self.g_updates = g_opt.minimize(self.g_loss, var_list=self.g_params)

    def generate(self, sess):
        outputs = sess.run(self.gen_x)
        return outputs

    def get_reward(self, sess, x):
        output = sess.run(self.log_predictions, feed_dict={self.x: x})
        return output

    def get_log_prob(self, sess, x):
        log_prob = sess.run(self.log_predictions, feed_dict={self.x: x})
        return log_prob

    def maximum_likelihood(self, sess, x):
        outputs = sess.run([self.likelihood_updates, self.likelihood_loss], feed_dict={self.x: x})
        return outputs

    def init_matrix(self, shape):
        return tf.random_normal(shape, stddev=0.1)

    def init_vector(self, shape):
        return tf.zeros(shape)

    def create_recurrent_unit_indRNN(self, params):
        self.Wi = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
        self.ui = tf.Variable(self.init_matrix([self.hidden_dim]))
        self.bi = tf.Variable(self.init_matrix([self.hidden_dim]))

        params.extend([
            self.Wi, self.ui, self.bi,
            ])

        def unit(x, hidden_memory_tm1):
            h, _ = tf.unstack(hidden_memory_tm1)
            h = tf.nn.relu(x @ self.Wi + h * self.ui + self.bi)
            return tf.stack([h, h])
        return unit

    def create_recurrent_unit_LSTM(self, params):
        # Weights and Bias for input and hidden tensor
        self.Wi = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
        self.Ui = tf.Variable(self.init_matrix([self.hidden_dim, self.hidden_dim]))
        self.bi = tf.Variable(self.init_matrix([self.hidden_dim]))

        self.Wf = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
        self.Uf = tf.Variable(self.init_matrix([self.hidden_dim, self.hidden_dim]))
        self.bf = tf.Variable(self.init_matrix([self.hidden_dim]))

        self.Wog = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
        self.Uog = tf.Variable(self.init_matrix([self.hidden_dim, self.hidden_dim]))
        self.bog = tf.Variable(self.init_matrix([self.hidden_dim]))

        self.Wc = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
        self.Uc = tf.Variable(self.init_matrix([self.hidden_dim, self.hidden_dim]))
        self.bc = tf.Variable(self.init_matrix([self.hidden_dim]))
        params.extend([
            self.Wi, self.Ui, self.bi,
            self.Wf, self.Uf, self.bf,
            self.Wog, self.Uog, self.bog,
            self.Wc, self.Uc, self.bc])

        def unit(x, hidden_memory_tm1):
            previous_hidden_state, c_prev = tf.unstack(hidden_memory_tm1)

            # Input Gate
            i = tf.sigmoid(
                tf.matmul(x, self.Wi) +
                tf.matmul(previous_hidden_state, self.Ui) + self.bi
            )

            # Forget Gate
            f = tf.sigmoid(
                tf.matmul(x, self.Wf) +
                tf.matmul(previous_hidden_state, self.Uf) + self.bf
            )

            # Output Gate
            o = tf.sigmoid(
                tf.matmul(x, self.Wog) +
                tf.matmul(previous_hidden_state, self.Uog) + self.bog
            )

            # New Memory Cell
            c_ = tf.nn.tanh(
                tf.matmul(x, self.Wc) +
                tf.matmul(previous_hidden_state, self.Uc) + self.bc
            )

            # Final Memory cell
            c = f * c_prev + i * c_

            # Current Hidden state
            current_hidden_state = o * tf.nn.tanh(c)

            return tf.stack([current_hidden_state, c])

        return unit

    def create_output_unit(self, params):
        self.Wo = tf.Variable(self.init_matrix([self.hidden_dim, self.num_emb]))
        self.bo = tf.Variable(self.init_matrix([self.num_emb]))
        params.extend([self.Wo, self.bo])

        def unit(hidden_memory_tuple):
            hidden_state, c_prev = tf.unstack(hidden_memory_tuple)
            # hidden_state : batch x hidden_dim
            # hidden_state = tf.nn.dropout(hidden_state, self.dropout_keep_rate)
            logits = tf.matmul(hidden_state, self.Wo) + self.bo
            # output = tf.nn.softmax(logits)
            return logits

        return unit

    def g_optimizer(self, *args, **kwargs):
        return tf.train.AdamOptimizer(*args, **kwargs)