Skip to content
Permalink
master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

bi-mcmc/src/net/thegeez/bi_mcmc/clusters.clj

Go to file
 
 
Cannot retrieve contributors at this time
270 lines (246 sloc) 12.1 KB
Raw Blame
Open in GitHub Desktop
(ns net.thegeez.bi-mcmc.clusters
(:require [incanter.charts :as charts]
[incanter.core :as inc-core]
[incanter.stats :as stats]
[net.thegeez.bi-mcmc.mcmc :as mcmc]
[net.thegeez.bi-mcmc.net :as net]
[net.thegeez.bi-mcmc.distributions :as dists]
[clojure.data.csv :as csv]
[clojure.java.io :as io]
[clojure.pprint :as pprint]))
;; chapter 3 from
;; http://nbviewer.ipython.org/github/CamDavidsonPilon/Probabilistic-Programming-and-Bayesian-Methods-for-Hackers/blob/master/Chapter3_MCMC/IntroMCMC.ipynb
;; from matkelcey
;; from pymc import *
;; from pymc.Matplot import plot
;; #data = [normal(100, 20) for _i in xrange(1000)]
;; data = map(float, open('data', 'r').readlines())
;; >>> min(data)
;; 40.3058046075
;; >>> max(data)
;; 172.745689773
;; mean = Uniform('mean', lower=min(data), upper=max(data))
;; precision = Uniform('precision', lower=0.0001, upper=1.0)
;; process = Normal('process', mu=mean, tau=precision, value=data, observed=True)
;; # do posterior sampling
;; m = pymc.MCMC([mean, precision, process],verbose=10)
;; #m.sample(iter=500, verbose=10)
;; m.sample(iter=10, verbose=10)
;; print(m.stats())
;; >>> m.stats()
;; {'precision': {'95% HPD interval': array([ 0.00307265, 0.03616355]), 'n': 500, 'quantiles': {2.5: 0.0030726520461263188, 25: 0.0056304478695005343, 50: 0.0056304478695005343, 75: 0.0056304478695005343, 97.5: 0.036163554933286587}, 'standard deviation': 0.024166450376244385, 'mc error': 0.0022530354801112954, 'mean': 0.012143625932209764}, 'mean': {'95% HPD interval': array([ 100.00015441, 117.23326122]), 'n': 500, 'quantiles': {2.5: 100.00015440990551, 25: 100.00015440990551, 50: 102.89806412433191, 75: 102.89806412433191, 97.5: 130.95207326510638}, 'standard deviation': 8.3417552405533666, 'mc error': 0.82388925092850029, 'mean': 103.90355067943328}}
;; >>>
(defonce data (let [file-path "data/matkelcey_simple.csv"]
(doall
(map (comp read-string first) (csv/read-csv (slurp (io/resource file-path)))))))
(def structure {:mean {:type :stochastic
:parents []
:children [:process]
:dist-fn (constantly
(dists/uniform-distribution 40.304 172.745)
#_(dists/normal-distribution 80.0 20.0)
)
:n 1
}
:std-dev {:type :stochastic
:n 1
:parents []
:children [:process]
:dist-fn (constantly
(dists/uniform-distribution 1.0 50.0)
#_(dists/uniform-distribution 19.9 20.1))}
:process {:type :stochastic
:parents [:mean :std-dev]
:children []
:dist-fn (fn [{:keys [mean std-dev] :as s}]
(dists/normal-distribution mean std-dev))
:n (count data)
:observed data}
})
(def structure-tau {:mean {:type :stochastic
:parents []
:children [:process]
:dist-fn (constantly
#_(dists/uniform-distribution 99.00 101.00)
(dists/uniform-distribution 40.304 172.745)
#_(dists/normal-distribution 80.0 20.0)
)
:n 1
}
:tau {:type :stochastic
:n 1
:parents []
:children [:std-dev]
:dist-fn (constantly
(dists/uniform-distribution 0.0001 1.0)
#_(dists/uniform-distribution 0.0024 0.0026))}
:std-dev {:type :deterministic
:n 1
:parents [:tau]
:children [:process]
:value-fn (fn [{:keys [tau]}]
(/ 1.0 (Math/sqrt tau)))}
:process {:type :stochastic
:parents [:mean :std-dev]
:children []
:dist-fn (fn [{:keys [mean std-dev] :as s}]
(dists/normal-distribution mean std-dev))
:n (count data)
:observed data}
})
(defn go []
(let [model (mcmc/mcmc structure #_-tau)]
(println "model" model)
(let [model (mcmc/sample model :iter 5000)]
(mcmc/pr-model model)
(def res model)
(pprint/pprint (mcmc/stats res)))
))
(defn go-tau []
(let [model (mcmc/mcmc structure-tau)]
(println "model" model)
(let [model (mcmc/sample model :iter 5000)]
(mcmc/pr-model model)
(def res model))
))
(comment
(let [model res
mean-vals (->> model
:samples
(map (comp :value :mean)))
std-vals (->> model
:samples
(map (comp :value :precision)))]
(doto (charts/xy-plot (range 500) mean-vals
:y-label "mean")
(charts/add-points (range 500) std-vals
:series-label "std")
#_(charts/add-function lin -3 3)
;; show biggest gains when p is low
#_(charts/add-function (fn [x]
(- (pax_p x)
(lin x))) 0 1)
#_(charts/add-points [from to] [p-from p-to])
#_(charts/add-points [from to] [logp-from logp-to])
inc-core/view))
(let [model res
mean-vals (->> model
:samples
(map (comp :value :mean)))]
(-> (charts/histogram mean-vals
:title "mean"
:nbins 30
:density true
)
#_(add-points [-10 20] [0.01 0.01])
inc-core/view))
(let [model res
mean-vals (->> model
:samples
(map (comp :value :mean)))
std-vals (->> model
:samples
(map (comp :value :precision)))
]
(stats/quantile mean-vals :probs [0.025 0.975]))
(let [model res
std-vals (->> model
:samples
(map (comp :value :precision)))
]
(stats/quantile std-vals :probs [0.025 0.975]))
(doto (charts/function-plot (fn [s]
(/ 1.0 (Math/sqrt s))) 0.001 50.0)
inc-core/view)
(doto (charts/function-plot (fn [t]
(/ 1.0 (* t t))) 0.001 1.0)
inc-core/view)
)
(defonce data-two (let [file-path "data/matkelcey_two.csv"]
(doall
(map (comp read-string first) (csv/read-csv (slurp (io/resource file-path)))))))
(def structure-two (let [n (count data-two)
min (apply min data-two)
max (apply max data-two)]
{:theta {:type :stochastic
:dist-fn (constantly
#_(dists/uniform-distribution 0.31 0.36)
(dists/uniform-distribution 0.0 1.0))
:n 1
}
:bern {:type :stochastic
:parents [:theta]
:dist-fn (fn [{theta :theta}]
(dists/binomial-distribution 1 theta))
:n (count data-two)
:step-scale 0.152003093 ;; log(1.0 - 0.1) / log 0.5
}
:mean1 {:type :stochastic
:dist-fn (constantly
(dists/uniform-distribution min max))
:n 1}
:mean2 {:type :stochastic
:dist-fn (constantly
(dists/uniform-distribution min max))
:n 1}
:std-dev {:type :stochastic
:dist-fn (constantly
#_(dists/uniform-distribution 18.0 22.0)
(dists/uniform-distribution 0.0 50.0))
:n 1
}
:mean {:type :deterministic
:parents [:bern :mean1 :mean2]
:value-fn (fn [{:keys [bern mean1 mean2]}]
;; (if (= bern 1) mean1 mean2)
(+ (* bern mean1)
(* (- 1 bern) mean2)))
:n n}
:process {:type :stochastic
:parents [:mean :std-dev]
:dist-fn (fn [{:keys [mean std-dev]}]
(dists/normal-distribution mean std-dev))
:n n
:observed data-two}}))
(defn chart-go-two [res]
(let [model res
mean1-vals (->> model
:samples
(map (comp :value :mean1)))
mean2-vals (->> model
:samples
(map (comp :value :mean2)))
std-vals (->> model
:samples
(map (comp :value :std-dev)))
theta-vals (->> model
:samples
(map (comp :value :theta)))]
(doto (charts/xy-plot (range) mean1-vals
:series-label "mean1"
:legend true)
(charts/add-points (range) mean2-vals
:series-label "mean2")
(charts/add-points (range) std-vals
:series-label "std")
inc-core/view)
(doto (charts/xy-plot (range) theta-vals
:series-label "theta"
:legend true)
inc-core/view)))
(defn go-two []
(let [model (mcmc/mcmc structure-two)]
(println "model" model)
(let [model (time (mcmc/sample model
:iter 100000 :burn 0 #_10000 :thin 1000
;; :iter 100 :burn 10 :thin 5
))]
(mcmc/pr-model model)
(def res model)
(pprint/pprint (mcmc/stats res))
(chart-go-two res)))
)
(comment
(chart-go-two res)
)
;; {:theta {:std-dev 0.013403034589704833, :mean 0.712831870202466, :quantiles {0.975 0.7333746636976199, 0.75 0.7210089328056395, 0.5 0.7143299420175615, 0.25 0.70497601670011, 0.025 0.68798796011754}, :mc-error 7.232812366720091E-4, :n 10000, :accepted 185, :rejected 9814}, :bern {:n 10000, :accepted 0, :rejected 9999}, :mean1 {:std-dev 2.9354356240079986, :mean 132.50650257579827, :quantiles {0.975 135.37761623300153, 0.75 133.3180231268265, 0.5 132.31519770405492, 0.25 131.72547871936558, 0.025 129.23042183191617}, :mc-error 0.12904597314788857, :n 10000, :accepted 106, :rejected 9893}, :mean2 {:std-dev 2.9444003351822747, :mean 136.96123383660253, :quantiles {0.975 141.73971252295715, 0.75 138.8813821712648, 0.5 137.1255005968757, 0.25 134.9752034278358, 0.025 132.8474174171206}, :mc-error 0.17057574621581284, :n 10000, :accepted 251, :rejected 9748}, :std-dev {:std-dev 0.9298255428806178, :mean 49.40049384311787, :quantiles {0.975 49.990456566240994, 0.75 49.79866575538405, 0.5 49.53601803303591, 0.25 49.21113425754308, 0.025 48.18724144586106}, :mc-error 0.050223705760840245, :n 10000, :accepted 209, :rejected 9790}}