Module mcmc_logtools

OBSOLETE

Returns: tuple(dict(str:onelog), dict(str:str))

a dictionary mapping filenames onto "data" and header information. Header information is from the last file read. The "data" are onelog class instances that encapsulate log information from one log file.

Returns: tuple(dict(str:onelog), dict(str:str))

a dictionary mapping filenames onto "data" and header information. Header information is from the last file read. The "data" are onelog class instances that encapsulate log information from one log file.

Parameters:

• per_fn (dict(str: mcmc_newlogger.logline))

Parameters:

• per_fn (dict(str: mcmc_newlogger.logline))

Parameters:

• per_fn (dict(str: mcmc_newlogger.logline))

Parameters:

• per_fn (dict(str: mcmc_newlogger.logline))

This selects which measurements will be used. It looks after convergence, then throws out optimizations that haven't converged.

This selects which measurements will be used. It looks after convergence, then gives you the best few results from each run.

Parameters:

• per_fn (dict(str: mcmc_newlogger.logline))

This selects which measurements will be used.

Parameters:

• per_fn (dict(str: mcmc_newlogger.logline))

Return a summary of the properties of the selected indexers.

Parameters:

• selector (function(dict(filename: list(mcmc_newlogger.logline))) that yields mcmc_newlogger.logline) - A generator that selects some of the loglines in per_fn that meet certain selection criteria.

Parameters:

• per_fn - produced byLT.read_uid_many_files()

Compares the ASCII form of keys, for sorting purposes. Does a good attempt at ASCII ordering for strings and numeric ordering for numbers.

Compares the tuple form of keys, for sorting purposes. Does a good attempt at ASCII ordering for strings and numeric ordering for numbers.

Log(PosteriorMarginalLikelihood) is the posterior marginal likelihood of this model: P_posterior[Data|model] = integral( P[D|params,model] * P[params|data,model] * d_params). It's the average of P[D|params,model] over the posterior distribution of P[params|data,model]*Prior(params).

References are:

• Rampal S. Etienne, Han Olff (2005) Confronting different models of community structure to species-abundance data: a Bayesian model comparison Ecology Letters 8 (5) , 493-504 doi:10.1111/j.1461-0248.2005.00745.x

and that references

• M. Aitkin 1991: Posterior Bayes Factors, J. of the Royal Statistical Soc. B 53: 111-142
• P. W. Laud and J. G. Ibrahim 1995: Predictive Model Selection, J. of the Royal Statistical Soc. B 57: 247-262.
• F. De Santis and F. Spezzaferri 1997: Alternative Bayes Factors for Model Selection, Canadian J. of Statistics 25: 503-515
• S. K. Upadhyay and M. Peswani 2003: Choice between Weibull and Lognormal Models: a simulation based Bayesian Study Communications in Statistics: Theory and Methods 32: 318-405
• P. K. Vlachos and A. E. Gelfand 2003: On the calibration of Bayeseian model choice criteria, J. of Statistical Planning and Inference 111: 223-234
• R. E. Kass and A. E. Rafferty 1995: Bayes Factors, J. of the American Statistical Assoc. 90: 773-795

The other thing, Log(BayesWeightedBayes) is the average of P(D|params,M)*Prior(params) over the posterior distribution of P[params|data,model]*Prior(params). It has no real statistical backing, but it's a crude approximation for the Bayes evidence itself, the normalized P[params|data,model]*Prior(params).