mdp.py 54.8 KB
Newer Older
Steven Cordwell's avatar
Steven Cordwell committed
1
# -*- coding: utf-8 -*-
2 3
"""Markov Decision Process (MDP) Toolbox: ``mdp`` module
=====================================================
4

5 6
The ``mdp`` module provides classes for the resolution of descrete-time Markov
Decision Processes.
Steven Cordwell's avatar
Steven Cordwell committed
7

Steven Cordwell's avatar
Steven Cordwell committed
8 9 10 11 12
Available classes
-----------------
MDP
    Base Markov decision process class
FiniteHorizon
Steven Cordwell's avatar
Steven Cordwell committed
13
    Backwards induction finite horizon MDP
Steven Cordwell's avatar
Steven Cordwell committed
14 15 16 17 18 19 20 21 22 23 24 25 26 27
LP
    Linear programming MDP
PolicyIteration
    Policy iteration MDP
PolicyIterationModified
    Modified policy iteration MDP
QLearning
    Q-learning MDP
RelativeValueIteration
    Relative value iteration MDP
ValueIteration
    Value iteration MDP
ValueIterationGS
    Gauss-Seidel value iteration MDP
Steven Cordwell's avatar
Steven Cordwell committed
28 29 30

"""

31
# Copyright (c) 2011-2015 Steven A. W. Cordwell
32
# Copyright (c) 2009 INRA
33
#
Steven Cordwell's avatar
Steven Cordwell committed
34
# All rights reserved.
35
#
Steven Cordwell's avatar
Steven Cordwell committed
36 37
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
38
#
Steven Cordwell's avatar
Steven Cordwell committed
39 40 41 42 43 44 45 46
#   * Redistributions of source code must retain the above copyright notice,
#     this list of conditions and the following disclaimer.
#   * Redistributions in binary form must reproduce the above copyright notice,
#     this list of conditions and the following disclaimer in the documentation
#     and/or other materials provided with the distribution.
#   * Neither the name of the <ORGANIZATION> nor the names of its contributors
#     may be used to endorse or promote products derived from this software
#     without specific prior written permission.
47
#
Steven Cordwell's avatar
Steven Cordwell committed
48 49 50 51 52 53 54 55 56 57 58 59
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
# POSSIBILITY OF SUCH DAMAGE.

60 61
import math as _math
import time as _time
Steven Cordwell's avatar
Steven Cordwell committed
62

63 64
import numpy as _np
import scipy.sparse as _sp
Steven Cordwell's avatar
Steven Cordwell committed
65

66
import mdptoolbox.util as _util
67

68
_MSG_STOP_MAX_ITER = "Iterating stopped due to maximum number of iterations " \
Steven Cordwell's avatar
Steven Cordwell committed
69
    "condition."
70
_MSG_STOP_EPSILON_OPTIMAL_POLICY = "Iterating stopped, epsilon-optimal " \
Steven Cordwell's avatar
Steven Cordwell committed
71
    "policy found."
72
_MSG_STOP_EPSILON_OPTIMAL_VALUE = "Iterating stopped, epsilon-optimal value " \
Steven Cordwell's avatar
Steven Cordwell committed
73
    "function found."
74
_MSG_STOP_UNCHANGING_POLICY = "Iterating stopped, unchanging policy found."
Steven Cordwell's avatar
Steven Cordwell committed
75

Steven Cordwell's avatar
Steven Cordwell committed
76
class MDP(object):
77

Steven Cordwell's avatar
Steven Cordwell committed
78
    """A Markov Decision Problem.
79

Steven Cordwell's avatar
Steven Cordwell committed
80
    Let ``S`` = the number of states, and ``A`` = the number of acions.
81

Steven Cordwell's avatar
Steven Cordwell committed
82 83 84
    Parameters
    ----------
    transitions : array
85
        Transition probability matrices. These can be defined in a variety of
Steven Cordwell's avatar
Steven Cordwell committed
86
        ways. The simplest is a numpy array that has the shape ``(A, S, S)``,
87
        though there are other possibilities. It can be a tuple or list or
Steven Cordwell's avatar
Steven Cordwell committed
88 89 90 91 92 93 94
        numpy object array of length ``A``, where each element contains a numpy
        array or matrix that has the shape ``(S, S)``. This "list of matrices"
        form is useful when the transition matrices are sparse as
        ``scipy.sparse.csr_matrix`` matrices can be used. In summary, each
        action's transition matrix must be indexable like ``transitions[a]``
        where ``a`` ∈ {0, 1...A-1}, and ``transitions[a]`` returns an ``S`` ×
        ``S`` array-like object.
Steven Cordwell's avatar
Steven Cordwell committed
95
    reward : array
96 97
        Reward matrices or vectors. Like the transition matrices, these can
        also be defined in a variety of ways. Again the simplest is a numpy
Steven Cordwell's avatar
Steven Cordwell committed
98 99 100 101 102 103 104 105
        array that has the shape ``(S, A)``, ``(S,)`` or ``(A, S, S)``. A list
        of lists can be used, where each inner list has length ``S`` and the
        outer list has length ``A``. A list of numpy arrays is possible where
        each inner array can be of the shape ``(S,)``, ``(S, 1)``, ``(1, S)``
        or ``(S, S)``. Also ``scipy.sparse.csr_matrix`` can be used instead of
        numpy arrays. In addition, the outer list can be replaced by any object
        that can be indexed like ``reward[a]`` such as a tuple or numpy object
        array of length ``A``.
106 107 108 109
    discount : float
        Discount factor. The per time-step discount factor on future rewards.
        Valid values are greater than 0 upto and including 1. If the discount
        factor is 1, then convergence is cannot be assumed and a warning will
Steven Cordwell's avatar
Steven Cordwell committed
110 111
        be displayed. Subclasses of ``MDP`` may pass ``None`` in the case where
        the algorithm does not use a discount factor.
112 113 114 115
    epsilon : float
        Stopping criterion. The maximum change in the value function at each
        iteration is compared against ``epsilon``. Once the change falls below
        this value, then the value function is considered to have converged to
Steven Cordwell's avatar
Steven Cordwell committed
116 117 118
        the optimal value function. Subclasses of ``MDP`` may pass ``None`` in
        the case where the algorithm does not use an epsilon-optimal stopping
        criterion.
119 120 121
    max_iter : int
        Maximum number of iterations. The algorithm will be terminated once
        this many iterations have elapsed. This must be greater than 0 if
Steven Cordwell's avatar
Steven Cordwell committed
122 123
        specified. Subclasses of ``MDP`` may pass ``None`` in the case where
        the algorithm does not use a maximum number of iterations.
124

Steven Cordwell's avatar
Steven Cordwell committed
125 126 127
    Attributes
    ----------
    P : array
128
        Transition probability matrices.
Steven Cordwell's avatar
Steven Cordwell committed
129
    R : array
130 131
        Reward vectors.
    V : tuple
Steven Cordwell's avatar
Steven Cordwell committed
132 133 134
        The optimal value function. Each element is a float corresponding to
        the expected value of being in that state assuming the optimal policy
        is followed.
Steven Cordwell's avatar
Steven Cordwell committed
135
    discount : float
136
        The discount rate on future rewards.
Steven Cordwell's avatar
Steven Cordwell committed
137
    max_iter : int
138 139 140
        The maximum number of iterations.
    policy : tuple
        The optimal policy.
Steven Cordwell's avatar
Steven Cordwell committed
141
    time : float
142 143
        The time used to converge to the optimal policy.
    verbose : boolean
Steven Cordwell's avatar
Steven Cordwell committed
144
        Whether verbose output should be displayed or not.
145

Steven Cordwell's avatar
Steven Cordwell committed
146 147
    Methods
    -------
148
    run
Steven Cordwell's avatar
Steven Cordwell committed
149
        Implemented in child classes as the main algorithm loop. Raises an
150
        exception if it has not been overridden.
Steven Cordwell's avatar
Steven Cordwell committed
151 152 153 154
    setSilent
        Turn the verbosity off
    setVerbose
        Turn the verbosity on
155

Steven Cordwell's avatar
Steven Cordwell committed
156
    """
157

158
    def __init__(self, transitions, reward, discount, epsilon, max_iter):
159
        # Initialise a MDP based on the input parameters.
160

Steven Cordwell's avatar
Steven Cordwell committed
161 162
        # if the discount is None then the algorithm is assumed to not use it
        # in its computations
163 164 165 166
        if discount is not None:
            self.discount = float(discount)
            assert 0.0 < self.discount <= 1.0, "Discount rate must be in ]0; 1]"
            if self.discount == 1:
Steven Cordwell's avatar
Steven Cordwell committed
167
                print("WARNING: check conditions of convergence. With no "
Steven Cordwell's avatar
Steven Cordwell committed
168
                      "discount, convergence can not be assumed.")
Steven Cordwell's avatar
Steven Cordwell committed
169 170
        # if the max_iter is None then the algorithm is assumed to not use it
        # in its computations
171 172 173 174
        if max_iter is not None:
            self.max_iter = int(max_iter)
            assert self.max_iter > 0, "The maximum number of iterations " \
                                      "must be greater than 0."
Steven Cordwell's avatar
Steven Cordwell committed
175
        # check that epsilon is something sane
176 177 178
        if epsilon is not None:
            self.epsilon = float(epsilon)
            assert self.epsilon > 0, "Epsilon must be greater than 0."
Steven Cordwell's avatar
Steven Cordwell committed
179 180
        # we run a check on P and R to make sure they are describing an MDP. If
        # an exception isn't raised then they are assumed to be correct.
181
        _util.check(transitions, reward)
Steven Cordwell's avatar
Steven Cordwell committed
182
        # computePR will assign the variables self.S, self.A, self.P and self.R
183
        self._computePR(transitions, reward)
Steven Cordwell's avatar
Steven Cordwell committed
184 185 186 187
        # the verbosity is by default turned off
        self.verbose = False
        # Initially the time taken to perform the computations is set to None
        self.time = None
188 189
        # set the initial iteration count to zero
        self.iter = 0
Steven Cordwell's avatar
Steven Cordwell committed
190
        # V should be stored as a vector ie shape of (S,) or (1, S)
Steven Cordwell's avatar
Steven Cordwell committed
191
        self.V = None
Steven Cordwell's avatar
Steven Cordwell committed
192
        # policy can also be stored as a vector
Steven Cordwell's avatar
Steven Cordwell committed
193
        self.policy = None
194

195 196 197 198 199 200
    def __repr__(self):
        P_repr = "P: \n"
        R_repr = "R: \n"
        for aa in range(self.A):
            P_repr += repr(self.P[aa]) + "\n"
            R_repr += repr(self.R[aa]) + "\n"
201
        return(P_repr + "\n" + R_repr)
202

203
    def _bellmanOperator(self, V=None):
Steven Cordwell's avatar
Steven Cordwell committed
204
        # Apply the Bellman operator on the value function.
205
        #
Steven Cordwell's avatar
Steven Cordwell committed
206
        # Updates the value function and the Vprev-improving policy.
207
        #
Steven Cordwell's avatar
Steven Cordwell committed
208 209 210 211
        # Returns: (policy, value), tuple of new policy and its value
        #
        # If V hasn't been sent into the method, then we assume to be working
        # on the objects V attribute
212 213
        if V is None:
            # this V should be a reference to the data rather than a copy
214 215
            V = self.V
        else:
Steven Cordwell's avatar
Steven Cordwell committed
216
            # make sure the user supplied V is of the right shape
217
            try:
218 219
                assert V.shape in ((self.S,), (1, self.S)), "V is not the " \
                    "right shape (Bellman operator)."
220
            except AttributeError:
221
                raise TypeError("V must be a numpy array or matrix.")
222 223 224 225
        # Looping through each action the the Q-value matrix is calculated.
        # P and V can be any object that supports indexing, so it is important
        # that you know they define a valid MDP before calling the
        # _bellmanOperator method. Otherwise the results will be meaningless.
226
        Q = _np.empty((self.A, self.S))
Steven Cordwell's avatar
Steven Cordwell committed
227
        for aa in range(self.A):
Steven Cordwell's avatar
Steven Cordwell committed
228
            Q[aa] = self.R[aa] + self.discount * self.P[aa].dot(V)
Steven Cordwell's avatar
Steven Cordwell committed
229
        # Get the policy and value, for now it is being returned but...
230
        # Which way is better?
231
        # 1. Return, (policy, value)
232
        return (Q.argmax(axis=0), Q.max(axis=0))
Steven Cordwell's avatar
Steven Cordwell committed
233 234
        # 2. update self.policy and self.V directly
        # self.V = Q.max(axis=1)
235
        # self.policy = Q.argmax(axis=1)
236

237 238 239 240 241 242 243 244
    def _computeP(self, P):
        # Set self.P as a tuple of length A, with each element storing an S×S
        # matrix.
        self.A = len(P)
        try:
            if P.ndim == 3:
                self.S = P.shape[1]
            else:
245
                self.S = P[0].shape[0]
246 247 248
        except AttributeError:
            self.S = P[0].shape[0]
        # convert P to a tuple of numpy arrays
249
        self.P = tuple(P[aa] for aa in range(self.A))
250

251
    def _computePR(self, P, R):
Steven Cordwell's avatar
Steven Cordwell committed
252 253 254 255
        # Compute the reward for the system in one state chosing an action.
        # Arguments
        # ---------
        # Let S = number of states, A = number of actions
256 257
        #    P(SxSxA)  = transition matrix
        #        P could be an array with 3 dimensions or  a cell array (1xA),
Steven Cordwell's avatar
Steven Cordwell committed
258 259
        #        each cell containing a matrix (SxS) possibly sparse
        #    R(SxSxA) or (SxA) = reward matrix
260 261 262
        #        R could be an array with 3 dimensions (SxSxA) or  a cell array
        #        (1xA), each cell containing a sparse matrix (SxS) or a 2D
        #        array(SxA) possibly sparse
Steven Cordwell's avatar
Steven Cordwell committed
263 264 265 266
        # Evaluation
        # ----------
        #    PR(SxA)   = reward matrix
        #
267
        # We assume that P and R define a MDP i,e. assumption is that
268
        # _util.check(P, R) has already been run and doesn't fail.
269
        #
270 271
        # First compute store P, S, and A
        self._computeP(P)
Steven Cordwell's avatar
Steven Cordwell committed
272 273
        # Set self.R as a tuple of length A, with each element storing an 1×S
        # vector.
274
        try:
275
            if R.ndim == 1:
276
                r = _np.array(R).reshape(self.S)
277
                self.R = tuple(r for aa in range(self.A))
278
            elif R.ndim == 2:
279
                self.R = tuple(_np.array(R[:, aa]).reshape(self.S)
280
                               for aa in range(self.A))
Steven Cordwell's avatar
Steven Cordwell committed
281
            else:
282
                self.R = tuple(_np.multiply(P[aa], R[aa]).sum(1).reshape(self.S)
283
                               for aa in range(self.A))
284
        except AttributeError:
285
            if len(R) == self.A:
286
                self.R = tuple(_np.multiply(P[aa], R[aa]).sum(1).reshape(self.S)
287
                               for aa in range(self.A))
288
            else:
289
                r = _np.array(R).reshape(self.S)
290
                self.R = tuple(r for aa in range(self.A))
291

292
    def run(self):
Steven Cordwell's avatar
Steven Cordwell committed
293
        # Raise error because child classes should implement this function.
294
        raise NotImplementedError("You should create a run() method.")
295

Steven Cordwell's avatar
Steven Cordwell committed
296
    def setSilent(self):
297
        """Set the MDP algorithm to silent mode."""
Steven Cordwell's avatar
Steven Cordwell committed
298
        self.verbose = False
299

Steven Cordwell's avatar
Steven Cordwell committed
300
    def setVerbose(self):
301
        """Set the MDP algorithm to verbose mode."""
Steven Cordwell's avatar
Steven Cordwell committed
302
        self.verbose = True
Steven Cordwell's avatar
Steven Cordwell committed
303 304

class FiniteHorizon(MDP):
305

Steven Cordwell's avatar
Steven Cordwell committed
306
    """A MDP solved using the finite-horizon backwards induction algorithm.
307

Steven Cordwell's avatar
Steven Cordwell committed
308 309
    Parameters
    ----------
310 311 312 313 314 315 316 317 318
    transitions : array
        Transition probability matrices. See the documentation for the ``MDP``
        class for details.
    reward : array
        Reward matrices or vectors. See the documentation for the ``MDP`` class
        for details.
    discount : float
        Discount factor. See the documentation for the ``MDP`` class for
        details.
Steven Cordwell's avatar
Steven Cordwell committed
319 320 321 322
    N : int
        Number of periods. Must be greater than 0.
    h : array, optional
        Terminal reward. Default: a vector of zeros.
323

Steven Cordwell's avatar
Steven Cordwell committed
324 325
    Data Attributes
    ---------------
326
    V : array
Steven Cordwell's avatar
Steven Cordwell committed
327
        Optimal value function. Shape = (S, N+1). ``V[:, n]`` = optimal value
Steven Cordwell's avatar
Steven Cordwell committed
328
        function at stage ``n`` with stage in {0, 1...N-1}. ``V[:, N]`` value
329
        function for terminal stage.
Steven Cordwell's avatar
Steven Cordwell committed
330 331
    policy : array
        Optimal policy. ``policy[:, n]`` = optimal policy at stage ``n`` with
Steven Cordwell's avatar
Steven Cordwell committed
332
        stage in {0, 1...N}. ``policy[:, N]`` = policy for stage ``N``.
Steven Cordwell's avatar
Steven Cordwell committed
333 334
    time : float
        used CPU time
335

Steven Cordwell's avatar
Steven Cordwell committed
336 337 338
    Notes
    -----
    In verbose mode, displays the current stage and policy transpose.
339

Steven Cordwell's avatar
Steven Cordwell committed
340 341
    Examples
    --------
342 343 344
    >>> import mdptoolbox, mdptoolbox.example
    >>> P, R = mdptoolbox.example.forest()
    >>> fh = mdptoolbox.mdp.FiniteHorizon(P, R, 0.9, 3)
345
    >>> fh.run()
Steven Cordwell's avatar
Steven Cordwell committed
346 347 348 349 350 351 352 353
    >>> fh.V
    array([[ 2.6973,  0.81  ,  0.    ,  0.    ],
           [ 5.9373,  3.24  ,  1.    ,  0.    ],
           [ 9.9373,  7.24  ,  4.    ,  0.    ]])
    >>> fh.policy
    array([[0, 0, 0],
           [0, 0, 1],
           [0, 0, 0]])
354

Steven Cordwell's avatar
Steven Cordwell committed
355
    """
Steven Cordwell's avatar
Steven Cordwell committed
356

Steven Cordwell's avatar
Steven Cordwell committed
357
    def __init__(self, transitions, reward, discount, N, h=None):
358
        # Initialise a finite horizon MDP.
359
        self.N = int(N)
Steven Cordwell's avatar
Steven Cordwell committed
360
        assert self.N > 0, "N must be greater than 0."
Steven Cordwell's avatar
Steven Cordwell committed
361
        # Initialise the base class
362
        MDP.__init__(self, transitions, reward, discount, None, None)
Steven Cordwell's avatar
Steven Cordwell committed
363 364
        # remove the iteration counter, it is not meaningful for backwards
        # induction
365
        del self.iter
Steven Cordwell's avatar
Steven Cordwell committed
366
        # There are value vectors for each time step up to the horizon
367
        self.V = _np.zeros((self.S, N + 1))
Steven Cordwell's avatar
Steven Cordwell committed
368 369
        # There are policy vectors for each time step before the horizon, when
        # we reach the horizon we don't need to make decisions anymore.
370
        self.policy = _np.empty((self.S, N), dtype=int)
Steven Cordwell's avatar
Steven Cordwell committed
371 372
        # Set the reward for the final transition to h, if specified.
        if h is not None:
373
            self.V[:, N] = h
374

375
    def run(self):
Steven Cordwell's avatar
Steven Cordwell committed
376
        # Run the finite horizon algorithm.
377
        self.time = _time.time()
Steven Cordwell's avatar
Steven Cordwell committed
378
        # loop through each time period
379
        for n in range(self.N):
Steven Cordwell's avatar
Steven Cordwell committed
380
            W, X = self._bellmanOperator(self.V[:, self.N - n])
Steven Cordwell's avatar
Steven Cordwell committed
381 382 383
            stage = self.N - n - 1
            self.V[:, stage] = X
            self.policy[:, stage] = W
Steven Cordwell's avatar
Steven Cordwell committed
384
            if self.verbose:
Steven Cordwell's avatar
Steven Cordwell committed
385 386
                print(("stage: %s, policy: %s") % (
                    stage, self.policy[:, stage].tolist()))
Steven Cordwell's avatar
Steven Cordwell committed
387
        # update time spent running
388
        self.time = _time.time() - self.time
389
        # After this we could create a tuple of tuples for the values and
Steven Cordwell's avatar
Steven Cordwell committed
390
        # policies.
Steven Cordwell's avatar
Steven Cordwell committed
391 392 393
        #self.V = tuple(tuple(self.V[:, n].tolist()) for n in range(self.N))
        #self.policy = tuple(tuple(self.policy[:, n].tolist())
        #                    for n in range(self.N))
Steven Cordwell's avatar
Steven Cordwell committed
394 395

class LP(MDP):
396

397
    """A discounted MDP soloved using linear programming.
398

Steven Cordwell's avatar
Steven Cordwell committed
399
    This class requires the Python ``cvxopt`` module to be installed.
Steven Cordwell's avatar
Steven Cordwell committed
400 401 402

    Arguments
    ---------
403 404 405 406 407 408 409 410 411
    transitions : array
        Transition probability matrices. See the documentation for the ``MDP``
        class for details.
    reward : array
        Reward matrices or vectors. See the documentation for the ``MDP`` class
        for details.
    discount : float
        Discount factor. See the documentation for the ``MDP`` class for
        details.
Steven Cordwell's avatar
Steven Cordwell committed
412 413
    h : array, optional
        Terminal reward. Default: a vector of zeros.
414

Steven Cordwell's avatar
Steven Cordwell committed
415 416 417 418 419 420 421 422
    Data Attributes
    ---------------
    V : tuple
        optimal values
    policy : tuple
        optimal policy
    time : float
        used CPU time
423

Steven Cordwell's avatar
Steven Cordwell committed
424 425
    Examples
    --------
426
    >>> import mdptoolbox.example
427 428
    >>> P, R = mdptoolbox.example.forest()
    >>> lp = mdptoolbox.mdp.LP(P, R, 0.9)
429
    >>> lp.run()
430 431 432 433 434 435 436 437

    >>> import numpy, mdptoolbox
    >>> P = numpy.array((((0.5, 0.5), (0.8, 0.2)), ((0, 1), (0.1, 0.9))))
    >>> R = numpy.array(((5, 10), (-1, 2)))
    >>> lp = mdptoolbox.mdp.LP(P, R, 0.9)
    >>> lp.run()
    >>> #lp.policy #FIXME: gives (1, 1), should be (1, 0)

Steven Cordwell's avatar
Steven Cordwell committed
438
    """
Steven Cordwell's avatar
Steven Cordwell committed
439

Steven Cordwell's avatar
Steven Cordwell committed
440
    def __init__(self, transitions, reward, discount):
441
        # Initialise a linear programming MDP.
Steven Cordwell's avatar
Steven Cordwell committed
442
        # import some functions from cvxopt and set them as object methods
Steven Cordwell's avatar
Steven Cordwell committed
443 444
        try:
            from cvxopt import matrix, solvers
445 446
            self._linprog = solvers.lp
            self._cvxmat = matrix
Steven Cordwell's avatar
Steven Cordwell committed
447
        except ImportError:
448 449
            raise ImportError("The python module cvxopt is required to use "
                              "linear programming functionality.")
Steven Cordwell's avatar
Steven Cordwell committed
450
        # initialise the MDP. epsilon and max_iter are not needed
451
        MDP.__init__(self, transitions, reward, discount, None, None)
Steven Cordwell's avatar
Steven Cordwell committed
452
        # Set the cvxopt solver to be quiet by default, but ...
453
        # this doesn't do what I want it to do c.f. issue #3
454 455
        if not self.verbose:
            solvers.options['show_progress'] = False
456

457
    def run(self):
Steven Cordwell's avatar
Steven Cordwell committed
458
        #Run the linear programming algorithm.
459
        self.time = _time.time()
Steven Cordwell's avatar
Steven Cordwell committed
460
        # The objective is to resolve : min V / V >= PR + discount*P*V
461 462
        # The function linprog of the optimisation Toolbox of Mathworks
        # resolves :
Steven Cordwell's avatar
Steven Cordwell committed
463
        # min f'* x / M * x <= b
464 465 466 467
        # So the objective could be expressed as :
        # min V / (discount*P-I) * V <= - PR
        # To avoid loop on states, the matrix M is structured following actions
        # M(A*S,S)
468
        f = self._cvxmat(_np.ones((self.S, 1)))
469 470
        h = _np.array(self.R).reshape(self.S * self.A, 1, order="F")
        h = self._cvxmat(h, tc='d')
471
        M = _np.zeros((self.A * self.S, self.S))
Steven Cordwell's avatar
Steven Cordwell committed
472 473
        for aa in range(self.A):
            pos = (aa + 1) * self.S
474
            M[(pos - self.S):pos, :] = (
475
                self.discount * self.P[aa] - _sp.eye(self.S, self.S))
476
        M = self._cvxmat(M)
477
        # Using the glpk option will make this behave more like Octave
478
        # (Octave uses glpk) and perhaps Matlab. If solver=None (ie using the
479
        # default cvxopt solver) then V agrees with the Octave equivalent
Steven Cordwell's avatar
Steven Cordwell committed
480
        # only to 10e-8 places. This assumes glpk is installed of course.
481
        self.V = _np.array(self._linprog(f, M, -h)['x']).reshape(self.S)
Steven Cordwell's avatar
Steven Cordwell committed
482
        # apply the Bellman operator
483
        self.policy, self.V = self._bellmanOperator()
Steven Cordwell's avatar
Steven Cordwell committed
484
        # update the time spent solving
485
        self.time = _time.time() - self.time
486
        # store value and policy as tuples
487 488
        self.V = tuple(self.V.tolist())
        self.policy = tuple(self.policy.tolist())
Steven Cordwell's avatar
Steven Cordwell committed
489 490

class PolicyIteration(MDP):
491

492
    """A discounted MDP solved using the policy iteration algorithm.
493

Steven Cordwell's avatar
Steven Cordwell committed
494 495
    Arguments
    ---------
496 497 498 499 500
    transitions : array
        Transition probability matrices. See the documentation for the ``MDP``
        class for details.
    reward : array
        Reward matrices or vectors. See the documentation for the ``MDP`` class
501
        for details.
502 503 504
    discount : float
        Discount factor. See the documentation for the ``MDP`` class for
        details.
Steven Cordwell's avatar
Steven Cordwell committed
505 506 507
    policy0 : array, optional
        Starting policy.
    max_iter : int, optional
508 509
        Maximum number of iterations. See the documentation for the ``MDP``
        class for details. Default is 1000.
Steven Cordwell's avatar
Steven Cordwell committed
510 511 512 513
    eval_type : int or string, optional
        Type of function used to evaluate policy. 0 or "matrix" to solve as a
        set of linear equations. 1 or "iterative" to solve iteratively.
        Default: 0.
514

Steven Cordwell's avatar
Steven Cordwell committed
515 516 517
    Data Attributes
    ---------------
    V : tuple
518
        value function
Steven Cordwell's avatar
Steven Cordwell committed
519 520 521 522 523 524
    policy : tuple
        optimal policy
    iter : int
        number of done iterations
    time : float
        used CPU time
525

Steven Cordwell's avatar
Steven Cordwell committed
526 527
    Notes
    -----
528
    In verbose mode, at each iteration, displays the number
Steven Cordwell's avatar
Steven Cordwell committed
529
    of differents actions between policy n-1 and n
530

Steven Cordwell's avatar
Steven Cordwell committed
531 532
    Examples
    --------
533
    >>> import mdptoolbox, mdptoolbox.example
534
    >>> P, R = mdptoolbox.example.rand(10, 3)
535
    >>> pi = mdptoolbox.mdp.PolicyIteration(P, R, 0.9)
536
    >>> pi.run()
537

538 539
    >>> P, R = mdptoolbox.example.forest()
    >>> pi = mdptoolbox.mdp.PolicyIteration(P, R, 0.9)
540
    >>> pi.run()
541 542 543
    >>> expected = (26.244000000000014, 29.484000000000016, 33.484000000000016)
    >>> all(expected[k] - pi.V[k] < 1e-12 for k in range(len(expected)))
    True
Steven Cordwell's avatar
Steven Cordwell committed
544
    >>> pi.policy
545
    (0, 0, 0)
Steven Cordwell's avatar
Steven Cordwell committed
546
    """
547

548 549
    def __init__(self, transitions, reward, discount, policy0=None,
                 max_iter=1000, eval_type=0):
Steven Cordwell's avatar
Steven Cordwell committed
550 551 552
        # Initialise a policy iteration MDP.
        #
        # Set up the MDP, but don't need to worry about epsilon values
553
        MDP.__init__(self, transitions, reward, discount, None, max_iter)
Steven Cordwell's avatar
Steven Cordwell committed
554
        # Check if the user has supplied an initial policy. If not make one.
555
        if policy0 is None:
Steven Cordwell's avatar
Steven Cordwell committed
556
            # Initialise the policy to the one which maximises the expected
Steven Cordwell's avatar
Steven Cordwell committed
557
            # immediate reward
558
            null = _np.zeros(self.S)
Steven Cordwell's avatar
Steven Cordwell committed
559
            self.policy, null = self._bellmanOperator(null)
560
            del null
Steven Cordwell's avatar
Steven Cordwell committed
561
        else:
Steven Cordwell's avatar
Steven Cordwell committed
562 563
            # Use the policy that the user supplied
            # Make sure it is a numpy array
564
            policy0 = _np.array(policy0)
Steven Cordwell's avatar
Steven Cordwell committed
565
            # Make sure the policy is the right size and shape
566 567
            assert policy0.shape in ((self.S, ), (self.S, 1), (1, self.S)), \
                "'policy0' must a vector with length S."
Steven Cordwell's avatar
Steven Cordwell committed
568
            # reshape the policy to be a vector
Steven Cordwell's avatar
Steven Cordwell committed
569
            policy0 = policy0.reshape(self.S)
570 571
            # The policy can only contain integers between 0 and S-1
            msg = "'policy0' must be a vector of integers between 0 and S-1."
572
            assert not _np.mod(policy0, 1).any(), msg
573 574 575
            assert (policy0 >= 0).all(), msg
            assert (policy0 < self.S).all(), msg
            self.policy = policy0
Steven Cordwell's avatar
Steven Cordwell committed
576
        # set the initial values to zero
577
        self.V = _np.zeros(self.S)
Steven Cordwell's avatar
Steven Cordwell committed
578
        # Do some setup depending on the evaluation type
Steven Cordwell's avatar
Steven Cordwell committed
579 580 581 582 583
        if eval_type in (0, "matrix"):
            self.eval_type = "matrix"
        elif eval_type in (1, "iterative"):
            self.eval_type = "iterative"
        else:
Steven Cordwell's avatar
Steven Cordwell committed
584 585 586
            raise ValueError("'eval_type' should be '0' for matrix evaluation "
                             "or '1' for iterative evaluation. The strings "
                             "'matrix' and 'iterative' can also be used.")
587

588
    def _computePpolicyPRpolicy(self):
Steven Cordwell's avatar
Steven Cordwell committed
589 590 591 592 593
        # Compute the transition matrix and the reward matrix for a policy.
        #
        # Arguments
        # ---------
        # Let S = number of states, A = number of actions
594
        # P(SxSxA)  = transition matrix
Steven Cordwell's avatar
Steven Cordwell committed
595 596 597
        #     P could be an array with 3 dimensions or a cell array (1xA),
        #     each cell containing a matrix (SxS) possibly sparse
        # R(SxSxA) or (SxA) = reward matrix
598
        #     R could be an array with 3 dimensions (SxSxA) or
Steven Cordwell's avatar
Steven Cordwell committed
599
        #     a cell array (1xA), each cell containing a sparse matrix (SxS) or
600
        #     a 2D array(SxA) possibly sparse
Steven Cordwell's avatar
Steven Cordwell committed
601 602 603 604 605 606 607
        # policy(S) = a policy
        #
        # Evaluation
        # ----------
        # Ppolicy(SxS)  = transition matrix for policy
        # PRpolicy(S)   = reward matrix for policy
        #
608 609
        Ppolicy = _np.empty((self.S, self.S))
        Rpolicy = _np.zeros(self.S)
Steven Cordwell's avatar
Steven Cordwell committed
610
        for aa in range(self.A): # avoid looping over S
Steven Cordwell's avatar
Steven Cordwell committed
611 612
            # the rows that use action a.
            ind = (self.policy == aa).nonzero()[0]
613 614
            # if no rows use action a, then no need to assign this
            if ind.size > 0:
615 616 617 618
                try:
                    Ppolicy[ind, :] = self.P[aa][ind, :]
                except ValueError:
                    Ppolicy[ind, :] = self.P[aa][ind, :].todense()
619
                #PR = self._computePR() # an apparently uneeded line, and
Steven Cordwell's avatar
Steven Cordwell committed
620 621
                # perhaps harmful in this implementation c.f.
                # mdp_computePpolicyPRpolicy.m
622
                Rpolicy[ind] = self.R[aa][ind]
Steven Cordwell's avatar
Steven Cordwell committed
623 624 625 626
        # self.R cannot be sparse with the code in its current condition, but
        # it should be possible in the future. Also, if R is so big that its
        # a good idea to use a sparse matrix for it, then converting PRpolicy
        # from a dense to sparse matrix doesn't seem very memory efficient
627 628
        if type(self.R) is _sp.csr_matrix:
            Rpolicy = _sp.csr_matrix(Rpolicy)
Steven Cordwell's avatar
Steven Cordwell committed
629 630 631
        #self.Ppolicy = Ppolicy
        #self.Rpolicy = Rpolicy
        return (Ppolicy, Rpolicy)
632

633
    def _evalPolicyIterative(self, V0=0, epsilon=0.0001, max_iter=10000):
Steven Cordwell's avatar
Steven Cordwell committed
634 635 636 637 638
        # Evaluate a policy using iteration.
        #
        # Arguments
        # ---------
        # Let S = number of states, A = number of actions
639 640
        # P(SxSxA)  = transition matrix
        #    P could be an array with 3 dimensions or
Steven Cordwell's avatar
Steven Cordwell committed
641 642
        #    a cell array (1xS), each cell containing a matrix possibly sparse
        # R(SxSxA) or (SxA) = reward matrix
643
        #    R could be an array with 3 dimensions (SxSxA) or
Steven Cordwell's avatar
Steven Cordwell committed
644
        #    a cell array (1xA), each cell containing a sparse matrix (SxS) or
645
        #    a 2D array(SxA) possibly sparse
Steven Cordwell's avatar
Steven Cordwell committed
646 647 648 649 650
        # discount  = discount rate in ]0; 1[
        # policy(S) = a policy
        # V0(S)     = starting value function, optional (default : zeros(S,1))
        # epsilon   = epsilon-optimal policy search, upper than 0,
        #    optional (default : 0.0001)
651
        # max_iter  = maximum number of iteration to be done, upper than 0,
Steven Cordwell's avatar
Steven Cordwell committed
652
        #    optional (default : 10000)
653
        #
Steven Cordwell's avatar
Steven Cordwell committed
654 655 656 657 658 659 660 661 662 663
        # Evaluation
        # ----------
        # Vpolicy(S) = value function, associated to a specific policy
        #
        # Notes
        # -----
        # In verbose mode, at each iteration, displays the condition which
        # stopped iterations: epsilon-optimum value function found or maximum
        # number of iterations reached.
        #
664 665 666
        try:
            assert V0.shape in ((self.S, ), (self.S, 1), (1, self.S)), \
                "'V0' must be a vector of length S."
667
            policy_V = _np.array(V0).reshape(self.S)
668
        except AttributeError:
Steven Cordwell's avatar
Steven Cordwell committed
669
            if V0 == 0:
670
                policy_V = _np.zeros(self.S)
Steven Cordwell's avatar
Steven Cordwell committed
671
            else:
672
                policy_V = _np.array(V0).reshape(self.S)
673

674
        policy_P, policy_R = self._computePpolicyPRpolicy()
675

Steven Cordwell's avatar
Steven Cordwell committed
676
        if self.verbose:
Steven Cordwell's avatar
Steven Cordwell committed
677
            print('    Iteration\t\t    V variation')
678

Steven Cordwell's avatar
Steven Cordwell committed
679 680 681
        itr = 0
        done = False
        while not done:
682
            itr += 1
683

684
            Vprev = policy_V
685
            policy_V = policy_R + self.discount * policy_P.dot(Vprev)
686

687
            variation = _np.absolute(policy_V - Vprev).max()
Steven Cordwell's avatar
Steven Cordwell committed
688
            if self.verbose:
Steven Cordwell's avatar
Steven Cordwell committed
689
                print(('      %s\t\t      %s') % (itr, variation))
690

691 692
            # ensure |Vn - Vpolicy| < epsilon
            if variation < ((1 - self.discount) / self.discount) * epsilon:
Steven Cordwell's avatar
Steven Cordwell committed
693 694
                done = True
                if self.verbose:
695
                    print(_MSG_STOP_EPSILON_OPTIMAL_VALUE)
Steven Cordwell's avatar
Steven Cordwell committed
696 697 698
            elif itr == max_iter:
                done = True
                if self.verbose:
699
                    print(_MSG_STOP_MAX_ITER)
700

Steven Cordwell's avatar
Steven Cordwell committed
701
        self.V = policy_V
702

703
    def _evalPolicyMatrix(self):
Steven Cordwell's avatar
Steven Cordwell committed
704 705
        # Evaluate the value function of the policy using linear equations.
        #
706
        # Arguments
Steven Cordwell's avatar
Steven Cordwell committed
707 708
        # ---------
        # Let S = number of states, A = number of actions
709
        # P(SxSxA) = transition matrix
Steven Cordwell's avatar
Steven Cordwell committed
710 711 712
        #      P could be an array with 3 dimensions or a cell array (1xA),
        #      each cell containing a matrix (SxS) possibly sparse
        # R(SxSxA) or (SxA) = reward matrix
713
        #      R could be an array with 3 dimensions (SxSxA) or
Steven Cordwell's avatar
Steven Cordwell committed
714
        #      a cell array (1xA), each cell containing a sparse matrix (SxS) or
715
        #      a 2D array(SxA) possibly sparse
Steven Cordwell's avatar
Steven Cordwell committed
716 717 718 719 720 721 722
        # discount = discount rate in ]0; 1[
        # policy(S) = a policy
        #
        # Evaluation
        # ----------
        # Vpolicy(S) = value function of the policy
        #
723
        Ppolicy, Rpolicy = self._computePpolicyPRpolicy()
Steven Cordwell's avatar
Steven Cordwell committed
724
        # V = PR + gPV  => (I-gP)V = PR  => V = inv(I-gP)* PR
725 726
        self.V = _np.linalg.solve(
            (_sp.eye(self.S, self.S) - self.discount * Ppolicy), Rpolicy)
727

728
    def run(self):
Steven Cordwell's avatar
Steven Cordwell committed
729 730
        # Run the policy iteration algorithm.
        # If verbose the print a header
731
        if self.verbose:
Steven Cordwell's avatar
Steven Cordwell committed
732
            print('  Iteration\t\tNumber of different actions')
Steven Cordwell's avatar
Steven Cordwell committed
733
        # Set up the while stopping condition and the current time
Steven Cordwell's avatar
Steven Cordwell committed
734
        done = False
735
        self.time = _time.time()
Steven Cordwell's avatar
Steven Cordwell committed
736
        # loop until a stopping condition is reached
Steven Cordwell's avatar
Steven Cordwell committed
737
        while not done:
738
            self.iter += 1
739
            # these _evalPolicy* functions will update the classes value
Steven Cordwell's avatar
Steven Cordwell committed
740
            # attribute
Steven Cordwell's avatar
Steven Cordwell committed
741
            if self.eval_type == "matrix":
742
                self._evalPolicyMatrix()
Steven Cordwell's avatar
Steven Cordwell committed
743
            elif self.eval_type == "iterative":
744
                self._evalPolicyIterative()
Steven Cordwell's avatar
Steven Cordwell committed
745 746
            # This should update the classes policy attribute but leave the
            # value alone
747
            policy_next, null = self._bellmanOperator()
748
            del null
Steven Cordwell's avatar
Steven Cordwell committed
749 750
            # calculate in how many places does the old policy disagree with
            # the new policy
751
            n_different = (policy_next != self.policy).sum()
Steven Cordwell's avatar
Steven Cordwell committed
752
            # if verbose then continue printing a table
753
            if self.verbose:
Steven Cordwell's avatar
Steven Cordwell committed
754
                print(('    %s\t\t  %s') % (self.iter, n_different))
755
            # Once the policy is unchanging of the maximum number of
Steven Cordwell's avatar
Steven Cordwell committed
756
            # of iterations has been reached then stop
757
            if n_different == 0:
Steven Cordwell's avatar
Steven Cordwell committed
758
                done = True
759
                if self.verbose:
760
                    print(_MSG_STOP_UNCHANGING_POLICY)
761
            elif self.iter == self.max_iter:
762
                done = True
763
                if self.verbose:
764
                    print(_MSG_STOP_MAX_ITER)
765 766
            else:
                self.policy = policy_next
Steven Cordwell's avatar
Steven Cordwell committed
767
        # update the time to return th computation time
768
        self.time = _time.time() - self.time
Steven Cordwell's avatar
Steven Cordwell committed
769
        # store value and policy as tuples
770 771
        self.V = tuple(self.V.tolist())
        self.policy = tuple(self.policy.tolist())
Steven Cordwell's avatar
Steven Cordwell committed
772

773
class PolicyIterationModified(PolicyIteration):
774

775
    """A discounted MDP  solved using a modifified policy iteration algorithm.