ChuaCheowHuan

from gym.spaces import Discrete

import numpy as np

import ray
from ray.rllib.env.multi_agent_env import MultiAgentEnv
from ray.rllib.models import ModelCatalog
from ray.rllib.agents.ppo import ddppo
from ray.rllib.agents.ppo import DDPPOTrainer
from ray.rllib.utils.framework import try_import_tf

ChuaCheowHuan

# -*- coding: utf-8 -*-
"""mod_pol.ipynb

Automatically generated by Colaboratory.

Original file is located at
"""

"""
# Commented out IPython magic to ensure Python compatibility.

ChuaCheowHuan

# -*- coding: utf-8 -*-
"""hyp_chg_MARL.ipynb

Automatically generated by Colaboratory.

Original file is located at
"""

# Commented out IPython magic to ensure Python compatibility.
"""

ChuaCheowHuan

def kl_mvn(m0, S0, m1, S1):
    """
    https://stackoverflow.com/questions/44549369/kullback-leibler-divergence-from-gaussian-pm-pv-to-gaussian-qm-qv

    The following function computes the KL-Divergence between any two 
    multivariate normal distributions 
    (no need for the covariance matrices to be diagonal)

    Kullback-Liebler divergence from Gaussian pm,pv to Gaussian qm,qv.
    Also computes KL divergence from a single Gaussian pm,pv to a set

ChuaCheowHuan

import tensorflow as tf
import numpy as np


latent_dim = 2

    
#class FSRNNCell_VU(tf.contrib.rnn.RNNCell):
class FSRNNCell_VU(tf.compat.v1.nn.rnn_cell.RNNCell):
    def __init__(self, fast_cells, slow_cell, input_keep_prob=1.0, keep_prob=1.0, training=True):

ChuaCheowHuan

#class FSRNNCell(tf.contrib.rnn.RNNCell):
class FSRNNCell(tf.compat.v1.nn.rnn_cell.RNNCell):
    def __init__(self, fast_cells, slow_cell, input_keep_prob=1.0, keep_prob=1.0, training=True):
        """Initialize the basic Fast-Slow RNN.
            Args:
              fast_cells: A list of RNN cells that will be used for the fast RNN.
                The cells must be callable, implement zero_state() and all have the
                same hidden size, like for example tf.contrib.rnn.BasicLSTMCell.
              slow_cell: A single RNN cell for the slow RNN.
              keep_prob: Keep probability for the non recurrent dropout. Any kind of

ChuaCheowHuan

import tensorflow as tf
import numpy as np

m1 = np.array([[[0.0], [1.1]], [[0.1], [1.2]], [[0.2], [1.3]]])
s1 = np.array([[[1.1]], [[1.2]], [[1.3]]])
m2 = np.array([[[1.0]], [[1.1]], [[1.2]]])
s2 = np.array([[[0.1]], [[0.2]], [[0.3]]])
print(m1.shape)

nd_1 = tf.distributions.Normal(loc=m1, scale=s1)

ChuaCheowHuan

#! /usr/bin/python
from __future__ import print_function
from random import *
import sys
sys.path.append('..')
from orderbook import OrderBook

order_book = OrderBook()

order_0 = {'type' : 'limit',

ChuaCheowHuan

Creating tmp8z_8haic_algo-1-vmd10_1 ...
Attaching to tmp8z_8haic_algo-1-vmd10_12mdone
algo-1-vmd10_1  | 2019-09-30 07:54:22,885 sagemaker-containers INFO     Imported framework sagemaker_tensorflow_container.training
algo-1-vmd10_1  | 2019-09-30 07:54:22,892 sagemaker-containers INFO     No GPUs detected (normal if no gpus installed)
algo-1-vmd10_1  | 2019-09-30 07:54:23,011 sagemaker-containers INFO     No GPUs detected (normal if no gpus installed)
algo-1-vmd10_1  | 2019-09-30 07:54:23,031 sagemaker-containers INFO     No GPUs detected (normal if no gpus installed)
algo-1-vmd10_1  | 2019-09-30 07:54:23,045 sagemaker-containers INFO     Invoking user script
algo-1-vmd10_1  |
algo-1-vmd10_1  | Training Env:
algo-1-vmd10_1  |