osprai/trunk/curvefitting.py

"""
CurveFitting: Perform the curve fitting for the data object.

Yuhang Wan
Last modified on 100427 (yymmdd) by YW

Typical Pipeline:(work together with "converter.py" and some model class
                    here take the basicmodel as example )
>----------load data-----------
>import converter as cv
>dat = cv.datafile_read("dat_example.obj") 
>----------load model----------
>import basicmodel as bm
>m = bm.basicmodel()
>m = m.load_model('model_example.obj')
>---------do the fitting-------
>import curvefitting as cf
>pfit, m2 = cf.fitting(dat, m)

"""
__version__ = "100427"

import numpy as np
import pylab as plt
import copy
from scipy.optimize import leastsq
import time
## import packageClamp_100225 as Pack
## import SPRdataclass_100225 as SPR
## import modelclass1_1 as Model


def initialize(data, parainfo):
    global pfloat_name, pfloat0, pfix_name
    # to seperate the parameters to fit and those fixed
    pfloat_name, pfloat0, pfix_name, pfix = [], [], [], []
    for i in parainfo:
        if i['fixed'] == 1: # for those fixed parameters
            pfix_name.append(i['name'])
            pfix.append(i['value'])
        elif i['fixed'] == 0:    # for those float parameters
            pfloat_name.append(i['name'])
            pfloat0.append(i['value'])
        else:   print 'unreadable status.\n\
                    please check the parameter infomation!'
    
    print '--'*30
    print 'The fixed parameters are:', pfix_name
    print 'The fixed value:'
    for i,j in enumerate(pfix_name):
        print j,':',pfix[i]
        
    print '\n','--'*30    
    print 'The float parameters are:', pfloat_name
    print 'The initial value:'
    for i,j in enumerate(pfloat_name):
        print j,':',pfloat0[i]

    print '\n','--'*30

    
    # format examination
    # to check if the input parameter matches the data to fit
    n_curve = len(data)/2
    for i in parainfo:
        ntmp = i['number']
        if ntmp != 1:
            if ntmp != n_curve:
                print 'the number of input curves doesn\'t match that of the \
                parameter\n\
                    Please check the data and the parameters!'
                break
            else:
                value = i['value']
                if ntmp != len(value):
                    print 'Please check the infomation of the parameters!'
                    break
                    
    pfloat0_1D, pfloat_name_1D = resizePfloat(pfloat0, pfloat_name, 0)
    return pfloat0_1D, pfix, pfloat_name_1D, pfix_name

def resizePfloat(p, pname, flag = 0):
    # resize the pfloat list
    # convertion between the two kinds of format:
    # 1D:       [p_a,p_b,p_c1,p_c2,p_c3,p_d]
    # mixed:    [p_a,p_b,[p_c1,p_c2,p_c3],p_d]
    
    # first examine the input
    if len(p) != len(pname):
        print 'The parameter list doesn\'t match the name list, please check again!'
        
    new_p, new_pname = [], []    

    if flag == 0:
    # from mixed list to one dim list:   
        for pi in p:
            if type(pi) == list:
                for j in pi:    
                    new_p.append(j)
                    new_pname.append(pname[p.index(pi)])
            else:
                new_p.append(pi)
                new_pname.append(pname[p.index(pi)])
        
    elif flag == 1:
    # from one dim list back to mixed list:
        ptmp = []
        for i,pni in enumerate(pname):
            n = pname.count(pni)
            if n == 1 :
                new_p.append(p[pname.index(pni)])
                new_pname.append(pni)
            else:                
                if pni not in new_pname:
                    new_pname.append(pni)
                    ptmp = p[i:i+n]
                    ptmp = list(ptmp)
                    new_p.append(ptmp)
                        
    return new_p, new_pname

def createdict(p,pname):
    '''Pack the name and value of all the parameters into a dictionary list.'''
    pdict = []
    if len(p) == len(pname):
        for i,j in enumerate(pname):
            dictmp = {}
            dictmp['name'] = j
            dictmp['value'] = p[i]
            pdict.append(dictmp)
        return pdict
    else:
        print 'check the input.\n'
        print pname
        print p
    return

## def keepsinglevalue(paratmp, i):
##     for n,p in enumerate(paratmp):
##         if type(p['value'])==list:
##             paralist[n]['value'] = p['value'][i]
##         else:
##             paralist[n]['value'] = p['value']
##     return paralist


def residuals(pfloat_1D, data, pfix):
    '''Error function.'''
    # the input of the float parameters are in one dimension
    e = 0
    pfloat, pfloat_name = resizePfloat(pfloat_1D, pfloat_name_1D, 1)
    
    p_all, pname_all = pfloat + pfix, pfloat_name + pfix_name
    paratmp = createdict(p_all, pname_all)
    paralist = copy.deepcopy(paratmp)    
    
    for i in range(len(data)/2):
        for n,p in enumerate(paratmp):
            if type(p['value'])==list:
                paralist[n]['value'] = p['value'][i]
            else:
                paralist[n]['value'] = p['value']
        
        t = data[i*2]
        y = data[i*2+1]
        e = e + (sprfunction(t,paralist) - y)**2
    return e
    

def lmafit(data,pfix):
    '''Fitting.
    '''
    # The initial values of the parameters need to be fit are converted 
    # into a one dimension list through "resizePfloat" function as required 
    # by the leastsq input.
    
    p_1D, success = leastsq(residuals, pfloat0_1D, args=(data,pfix), maxfev=10000)
    p, pfloat_name = resizePfloat(p_1D, pfloat_name_1D, 1)
    print 'Success:', success
    plotfit(data, p_1D, pfix)

    return p
    
        
def plotfit(data, pfit_1D, pfix):
    '''Plotting of real and fitted data.
    Creating titles with ssq error and parameters.
    Printing the result and fitted parameters.'''
    global pfit_name
    pfit, pfit_name = resizePfloat(pfit_1D, pfloat_name_1D, 1)
    txt = 'SumSqE: %1.8f' % sum(np.square(residuals(pfit_1D, data, pfix)))
    print txt
    plt.title(txt)
    plt.xlabel('Time (s)')
    plt.ylabel('Response (uRIU)')
    plt.grid(True)
    
    p_all, pname_all = pfit + pfix, pfit_name + pfix_name
    paratmp = createdict(p_all, pname_all)
    paralist = copy.deepcopy(paratmp)    
    
    for i in range(len(data)/2):
        for n,p in enumerate(paratmp):
            if type(p['value'])==list:
                paralist[n]['value'] = p['value'][i]
            else:
                paralist[n]['value'] = p['value']
        t = data[i*2]
        y = data[i*2+1]
        plt.plot(t, y, ',')  # Plot real data
        plt.plot(t, sprfunction(t, paralist))  # Plot fitted data
    plt.show()

    print '\n','-'*60
    print 'The fitted parameters are:', pfit_name
    print 'The fitted value:'
    for i,j in enumerate(pfit_name):
        print j,':',pfit[i]
    
    return
    
    
def updateparainfo(parainfo, pfit, pfit_name):
    new_parainfo = copy.deepcopy(parainfo)

    print '\n', '+'*25, "Comparison", '+'*25
##    print '-'*60
    print 'name\t\t initial value \t\t fitted value'
    for i in new_parainfo:
        for n,j in enumerate(pfit_name):
            if i['name'] == j:
                print j, '\t\t ', i['value'], ' \t\t ', pfit[n]
                i['value'] = pfit[n]
            
    return new_parainfo


def fitting(dataobj, mobj):
    '''The main function of curvefitting.
    Input the data object and the model object.
    '''
    modelobj = copy.deepcopy(mobj)
    time1 = time.time()
    global pfloat0_1D, pfix, pfloat_name_1D, pfix_name, sprfunction
    data = dataobj.data
    parainfo = modelobj.parainfo
    pfloat0_1D, pfix, pfloat_name_1D, pfix_name = initialize(data, parainfo)
    sprfunction = modelobj.function

    pfit = lmafit(data, pfix)
    parainfo = updateparainfo(parainfo, pfit, pfloat_name)
    
    time2 = time.time()
    print '-'*60
    print "Time elapsed:   %s seconds\n" %(time2-time1)
    print 'Do you want to update the model with fitted value?'
    if str.upper(raw_input('y/n : ')) == 'Y':        
        modelobj.updatemodel(parainfo)
        print 'The model is updated.'
    
    return pfit, modelobj
Revision:	25
Committed:	Wed Apr 28 20:22:24 2010 UTC (10 years, 2 months ago) by rjaynes
File size:	8430 byte(s)
Log Message:	Add py and obj files to allow modeling of more SPR experiments with converter and curvefitting modules. This is the work of Yuhang Wan and Rui Hou. 1. In "converter.py": Add the saving and reading function for the sprclass data object. Also add function "keyfile_read_fake" to provide default information for SPRit and ICM formats in case of the bug when do background_subtract. Fix the bugs in "background_subtract". Tested by DAM and ICM formats. 2. In model modules: "modelclass.py" is the parent class for all the other model classes that performs the theoretical simulating, loading and saving of the parameter or simulated data. Rui and I also add some other model modules like competing model, twostate model, parallel model, and the time variable concentrated models, where the simulated result is compared with Clamp's simulation to make sure the equations are correct. The basicmodel and basicmodel_varyC class are tested. 3. In "curvefitting.py": Add typical pipeline for operation. The examples are packed with the file. Add function to show the Elapsed time for each fitting.
Line	File contents
1	"""
2	CurveFitting: Perform the curve fitting for the data object.
3
4	Yuhang Wan
5	Last modified on 100427 (yymmdd) by YW
6
7	Typical Pipeline:(work together with "converter.py" and some model class
8	here take the basicmodel as example )
9	>----------load data-----------
10	>import converter as cv
11	>dat = cv.datafile_read("dat_example.obj")
12	>----------load model----------
13	>import basicmodel as bm
14	>m = bm.basicmodel()
15	>m = m.load_model('model_example.obj')
16	>---------do the fitting-------
17	>import curvefitting as cf
18	>pfit, m2 = cf.fitting(dat, m)
19
20	"""
21	__version__ = "100427"
22
23	import numpy as np
24	import pylab as plt
25	import copy
26	from scipy.optimize import leastsq
27	import time
28	## import packageClamp_100225 as Pack
29	## import SPRdataclass_100225 as SPR
30	## import modelclass1_1 as Model
31
32
33	def initialize(data, parainfo):
34	global pfloat_name, pfloat0, pfix_name
35	# to seperate the parameters to fit and those fixed
36	pfloat_name, pfloat0, pfix_name, pfix = [], [], [], []
37	for i in parainfo:
38	if i['fixed'] == 1: # for those fixed parameters
39	pfix_name.append(i['name'])
40	pfix.append(i['value'])
41	elif i['fixed'] == 0: # for those float parameters
42	pfloat_name.append(i['name'])
43	pfloat0.append(i['value'])
44	else: print 'unreadable status.\n\
45	please check the parameter infomation!'
46
47	print '--'*30
48	print 'The fixed parameters are:', pfix_name
49	print 'The fixed value:'
50	for i,j in enumerate(pfix_name):
51	print j,':',pfix[i]
52
53	print '\n','--'*30
54	print 'The float parameters are:', pfloat_name
55	print 'The initial value:'
56	for i,j in enumerate(pfloat_name):
57	print j,':',pfloat0[i]
58
59	print '\n','--'*30
60
61
62	# format examination
63	# to check if the input parameter matches the data to fit
64	n_curve = len(data)/2
65	for i in parainfo:
66	ntmp = i['number']
67	if ntmp != 1:
68	if ntmp != n_curve:
69	print 'the number of input curves doesn\'t match that of the \
70	parameter\n\
71	Please check the data and the parameters!'
72	break
73	else:
74	value = i['value']
75	if ntmp != len(value):
76	print 'Please check the infomation of the parameters!'
77	break
78
79	pfloat0_1D, pfloat_name_1D = resizePfloat(pfloat0, pfloat_name, 0)
80	return pfloat0_1D, pfix, pfloat_name_1D, pfix_name
81
82	def resizePfloat(p, pname, flag = 0):
83	# resize the pfloat list
84	# convertion between the two kinds of format:
85	# 1D: [p_a,p_b,p_c1,p_c2,p_c3,p_d]
86	# mixed: [p_a,p_b,[p_c1,p_c2,p_c3],p_d]
87
88	# first examine the input
89	if len(p) != len(pname):
90	print 'The parameter list doesn\'t match the name list, please check again!'
91
92	new_p, new_pname = [], []
93
94	if flag == 0:
95	# from mixed list to one dim list:
96	for pi in p:
97	if type(pi) == list:
98	for j in pi:
99	new_p.append(j)
100	new_pname.append(pname[p.index(pi)])
101	else:
102	new_p.append(pi)
103	new_pname.append(pname[p.index(pi)])
104
105	elif flag == 1:
106	# from one dim list back to mixed list:
107	ptmp = []
108	for i,pni in enumerate(pname):
109	n = pname.count(pni)
110	if n == 1 :
111	new_p.append(p[pname.index(pni)])
112	new_pname.append(pni)
113	else:
114	if pni not in new_pname:
115	new_pname.append(pni)
116	ptmp = p[i:i+n]
117	ptmp = list(ptmp)
118	new_p.append(ptmp)
119
120	return new_p, new_pname
121
122	def createdict(p,pname):
123	'''Pack the name and value of all the parameters into a dictionary list.'''
124	pdict = []
125	if len(p) == len(pname):
126	for i,j in enumerate(pname):
127	dictmp = {}
128	dictmp['name'] = j
129	dictmp['value'] = p[i]
130	pdict.append(dictmp)
131	return pdict
132	else:
133	print 'check the input.\n'
134	print pname
135	print p
136	return
137
138	## def keepsinglevalue(paratmp, i):
139	## for n,p in enumerate(paratmp):
140	## if type(p['value'])==list:
141	## paralist[n]['value'] = p['value'][i]
142	## else:
143	## paralist[n]['value'] = p['value']
144	## return paralist
145
146
147	def residuals(pfloat_1D, data, pfix):
148	'''Error function.'''
149	# the input of the float parameters are in one dimension
150	e = 0
151	pfloat, pfloat_name = resizePfloat(pfloat_1D, pfloat_name_1D, 1)
152
153	p_all, pname_all = pfloat + pfix, pfloat_name + pfix_name
154	paratmp = createdict(p_all, pname_all)
155	paralist = copy.deepcopy(paratmp)
156
157	for i in range(len(data)/2):
158	for n,p in enumerate(paratmp):
159	if type(p['value'])==list:
160	paralist[n]['value'] = p['value'][i]
161	else:
162	paralist[n]['value'] = p['value']
163
164	t = data[i*2]
165	y = data[i*2+1]
166	e = e + (sprfunction(t,paralist) - y)**2
167	return e
168
169
170
171	def lmafit(data,pfix):
172	'''Fitting.
173	'''
174	# The initial values of the parameters need to be fit are converted
175	# into a one dimension list through "resizePfloat" function as required
176	# by the leastsq input.
177
178	p_1D, success = leastsq(residuals, pfloat0_1D, args=(data,pfix), maxfev=10000)
179	p, pfloat_name = resizePfloat(p_1D, pfloat_name_1D, 1)
180	print 'Success:', success
181	plotfit(data, p_1D, pfix)
182
183	return p
184
185
186	def plotfit(data, pfit_1D, pfix):
187	'''Plotting of real and fitted data.
188	Creating titles with ssq error and parameters.
189	Printing the result and fitted parameters.'''
190	global pfit_name
191	pfit, pfit_name = resizePfloat(pfit_1D, pfloat_name_1D, 1)
192	txt = 'SumSqE: %1.8f' % sum(np.square(residuals(pfit_1D, data, pfix)))
193	print txt
194	plt.title(txt)
195	plt.xlabel('Time (s)')
196	plt.ylabel('Response (uRIU)')
197	plt.grid(True)
198
199	p_all, pname_all = pfit + pfix, pfit_name + pfix_name
200	paratmp = createdict(p_all, pname_all)
201	paralist = copy.deepcopy(paratmp)
202
203	for i in range(len(data)/2):
204	for n,p in enumerate(paratmp):
205	if type(p['value'])==list:
206	paralist[n]['value'] = p['value'][i]
207	else:
208	paralist[n]['value'] = p['value']
209	t = data[i*2]
210	y = data[i*2+1]
211	plt.plot(t, y, ',') # Plot real data
212	plt.plot(t, sprfunction(t, paralist)) # Plot fitted data
213	plt.show()
214
215	print '\n','-'*60
216	print 'The fitted parameters are:', pfit_name
217	print 'The fitted value:'
218	for i,j in enumerate(pfit_name):
219	print j,':',pfit[i]
220
221	return
222
223
224	def updateparainfo(parainfo, pfit, pfit_name):
225	new_parainfo = copy.deepcopy(parainfo)
226
227	print '\n', '+'25, "Comparison", '+'25
228	## print '-'*60
229	print 'name\t\t initial value \t\t fitted value'
230	for i in new_parainfo:
231	for n,j in enumerate(pfit_name):
232	if i['name'] == j:
233	print j, '\t\t ', i['value'], ' \t\t ', pfit[n]
234	i['value'] = pfit[n]
235
236	return new_parainfo
237
238
239
240	def fitting(dataobj, mobj):
241	'''The main function of curvefitting.
242	Input the data object and the model object.
243	'''
244	modelobj = copy.deepcopy(mobj)
245	time1 = time.time()
246	global pfloat0_1D, pfix, pfloat_name_1D, pfix_name, sprfunction
247	data = dataobj.data
248	parainfo = modelobj.parainfo
249	pfloat0_1D, pfix, pfloat_name_1D, pfix_name = initialize(data, parainfo)
250	sprfunction = modelobj.function
251
252	pfit = lmafit(data, pfix)
253	parainfo = updateparainfo(parainfo, pfit, pfloat_name)
254
255	time2 = time.time()
256	print '-'*60
257	print "Time elapsed: %s seconds\n" %(time2-time1)
258	print 'Do you want to update the model with fitted value?'
259	if str.upper(raw_input('y/n : ')) == 'Y':
260	modelobj.updatemodel(parainfo)
261	print 'The model is updated.'
262
263	return pfit, modelobj