ph!SrSSKJr SSKrSSKJr SSKJr SSK J r SSK J r J r JrJrJr SS KJrJr SS KJr S r\R."S \5r"S S\5r\"5r"SS\R6\5rSr\R<R>S5r \RCS5S5r"Sr#\RCS5S5r$\RCSSS9S5r%\RCS5S5r&\RCS5\"5S55r'\RCS SS9\"5S!55r(\RCS"SS9\RCS#SS9\"5S$555r)S%r*S&r+S'r,g)(z$ Functions for measuring distances. )ContextDecoratorN)emd)cdist)entropy) Directionconfig constantsutilsvalidate)flatten marginal_zero)Registry hamming_matricesc@^\rSrSrSrSrU4SjrSSjrSrSr U=r $) MeasureRegistryaStorage for measures registered with PyPhi. Users can define custom measures: Examples: >>> @measures.register('ALWAYS_ZERO') # doctest: +SKIP ... def always_zero(a, b): ... return 0 And use them by setting ``config.MEASURE = 'ALWAYS_ZERO'``. measuresc0>[TU]5 /UlgN)super__init__ _asymmetricself __class__s H/home/james-whalen/.local/lib/python3.13/site-packages/pyphi/distance.pyrMeasureRegistry.__init__*s c^^^UUU4SjnU$)zDecorator for registering a measure with PyPhi. Args: name (string): The name of the measure. Keyword Args: asymmetric (boolean): ``True`` if the measure is asymmetric. cj>T(aTRRT5 UTRT'U$r)rappendstore)func asymmetricnamers r register_func/MeasureRegistry.register..register_func7s-  ''-#DJJt Kr )rr'r&r(s``` rregisterMeasureRegistry.register.s  r cUR$)z%Return a list of asymmetric measures.r)rs rr&MeasureRegistry.asymmetric>sr r.)F) __name__ __module__ __qualname____firstlineno____doc__descrr+r&__static_attributes__ __classcell__rs@rrrs#  D   r rc,^\rSrSrSrU4SjrSrU=r$) np_suppressFzDecorator to suppress NumPy warnings about divide-by-zero and multiplication of ``NaN``. .. note:: This should only be used in cases where you are *sure* that these warnings are not indicative of deeper issues in your code. c">[TU]SSS9 g)Nignore)divideinvalid)rrrs rrnp_suppress.__init__Ns (;r r*)r0r1r2r3r4rr6r7r8s@rr:r:Fs<U[:a [U$[U5$)aReturn a matrix of Hamming distances for the possible states of |N| binary nodes. Args: N (int): The number of nodes under consideration Returns: np.ndarray: A |2^N x 2^N| matrix where the |ith| element is the Hamming distance between state |i| and state |j|. Example: >>> _hamming_matrix(2) array([[0., 1., 1., 2.], [1., 0., 2., 1.], [1., 2., 0., 1.], [2., 1., 1., 0.]]) )!_NUM_PRECOMPUTED_HAMMING_MATRICES_hamming_matrices_compute_hamming_matrix)Ns r_hamming_matrixrFSs#$ ,, ## "1 %%r c[R"[[R"U555n[ XS5U-$)a+Compute and store a Hamming matrix for |N| nodes. Hamming matrices have the following sizes:: N MBs == === 9 2 10 8 11 32 12 128 13 512 Given these sizes and the fact that large matrices are needed infrequently, we store computed matrices using the Joblib filesystem cache instead of adding computed matrices to the ``_hamming_matrices`` global and clogging up memory. This function is only called when |N| > ``_NUM_PRECOMPUTED_HAMMING_MATRICES``. Don't call this function directly; use |_hamming_matrix| instead. hamming)nparraylistr all_statesr)rEpossible_statess rrDrDjs4.hhtE$4$4a$9:;O 9 = AAr EMDcUR5Rn[U5[U5p[X[ U55$)zReturn the Earth Mover's Distance between two distributions (indexed by state, one dimension per node) using the Hamming distance between states as the transportation cost function. Singleton dimensions are sqeezed out. )squeezendimr rrF)d1d2rEs r hamming_emdrTs6 A R['"+ rq) **r cX^^[UU4Sj[TR555$)aCompute the EMD between two effect repertoires. Because the nodes are independent, the EMD between effect repertoires is equal to the sum of the EMDs between the marginal distributions of each node, and the EMD between marginal distribution for a node is the absolute difference in the probabilities that the node is OFF. Args: d1 (np.ndarray): The first repertoire. d2 (np.ndarray): The second repertoire. Returns: float: The EMD between ``d1`` and ``d2``. c3h># UH'n[[TU5[TU5- 5v M) g7fr)absr).0irRrSs r effect_emd..s3(&1=Q'-A*>>??&s/2)sumrangerQrRrSs``r effect_emdr_s' (bgg( ((r L1cN[R"X- 5R5$)zReturn the L1 distance between two distributions. Args: d1 (np.ndarray): The first distribution. d2 (np.ndarray): The second distribution. Returns: float: The sum of absolute differences of ``d1`` and ``d2``. )rIabsoluter\r^s rl1rcs ;;rw  # # %%r KLDT)r&cD[U5[U5p[XS5$)zReturn the Kullback-Leibler Divergence (KLD) between two distributions. Args: d1 (np.ndarray): The first distribution. d2 (np.ndarray): The second distribution. Returns: float: The KLD of ``d1`` from ``d2``. @)r rr^s rkldrgs R['"+ 23 r ENTROPY_DIFFERENCEch[U5[U5p[[USS9[USS9- 5$)z;Return the difference in entropy between two distributions.rf)base)r rWrr^s rentropy_differencerks1R['"+ wr$wr'<< ==r PSQ2cf[U5[U5pSn[U"U5U"U5- 5$)z|Compute the PSQ2 measure. Args: d1 (np.ndarray): The first distribution. d2 (np.ndarray): The second distribution. c [US-[R"[R"U[ U5-55-5$)N)r\rI nan_to_numloglen)ps rfpsq2..fs0AFbmmBFF1s1v:,>??@@r )r rW)rRrSrts rpsq2rvs2R['"+A ququ} r MP2Qc [U5[U5pS[U5- n[U[R"US-U- [R "X- 5-5-5$)zCompute the MP2Q measure. Args: p (np.ndarray): The unpartitioned repertoire q (np.ndarray): The partitioned repertoire rro)r rrr\rIrprq)rsq entropy_dists rmp2qr{sP 1:wqzqs1v:L |bmmQ!VqL266!%=,HII JJr KLMBLDc [U5[U5p[[U[R"[R "X- 55-55$)zCompute the KLM divergence.)r maxrWrIrprq)rsrys rklmrs; 1:wqzq s1r}}RVVAE]334 55r cU[R:Xa[nO1U[R:Xa[nO[ R "U5 [W"X5[R5$)aCompute the EMD between two repertoires for a given direction. The full EMD computation is used for cause repertoires. A fast analytic solution is used for effect repertoires. Args: direction (Direction): |CAUSE| or |EFFECT|. d1 (np.ndarray): The first repertoire. d2 (np.ndarray): The second repertoire. Returns: float: The EMD between ``d1`` and ``d2``, rounded to |PRECISION|. Raises: ValueError: If ``direction`` is invalid. ) rCAUSErTEFFECTr_r directionroundr PRECISION)rrRrSr%s rdirectional_emdrsO"IOO# i&& & 9% bv// 00r c[RS:Xa [XU5nO[[R"X5n[ U[R 5$)a)Compute the distance between two repertoires for the given direction. Args: direction (Direction): |CAUSE| or |EFFECT|. r1 (np.ndarray): The first repertoire. r2 (np.ndarray): The second repertoire. Returns: float: The distance between ``d1`` and ``d2``, rounded to |PRECISION|. rN)r MEASURErrrr)rr1r2dists rrepertoire_distancersB~~yb1'/ v'' ((r c[R[R5;a([ SR [R55e[[R"X5$)zCompute the distance between two repertoires of a system. Args: r1 (np.ndarray): The first repertoire. r2 (np.ndarray): The second repertoire. Returns: float: The distance between ``r1`` and ``r2``. zM{} is asymmetric and cannot be used as a system-level irreducibility measure.)r rrr& ValueErrorformat)rrs rsystem_repertoire_distancer!sO~~,,.. &&,fV^^&<> > FNN #B ++r )-r4 contextlibrnumpyrIpyemdrscipy.spatial.distancer scipy.statsrrr r r r distributionr rregistryrrB load_datarCrrerrstater:rF joblib_memorycacherDr+rTr_rcrgrkrvr{rrrrr*r rrs ((;;0%'!OO$6$EG& h& R   <"++/ <&. BB6 5 + +(& 4 & & 5T*  +   '(>)>  6    6d+  K, K 5T* 5T* 6++6 18)&,r