def general_logistic_regression():
'''Example General Logistic Recression,
Example 7.4.1, p. 135'''
# Get the data
inFile = r'GLM_data/Table 7.5 Embryogenic anthers.xls'
df = get_data(inFile)
# Define the variables so that they match Dobson
df['n_y'] = df['n'] - df['y']
df['newstor'] = df['storage'] - 1
df['x'] = np.log(df['centrifuge'])
# Model 1
model1 = smf.glm('n_y + y ~ newstor*x',
data=df,
family=sm_families.Binomial()).fit()
print(model1.summary())
# Model 2
model2 = smf.glm('n_y + y ~ newstor+x',
data=df,
family=sm_families.Binomial()).fit()
print(model2.summary())
# Model 3
model3 = smf.glm('n_y + y ~ x', data=df,
family=sm_families.Binomial()).fit()
print(model3.summary())
def logistic_regression():
'''Logistic regression example
chapter 7.3, p 130
[tbd]: the cloglog values are inconsistent with those mentioned in the book.
This is probably due to the specific definitions of "loglog" and "cloglog"
in the respective languages.
'''
inFile = r'GLM_data/Table 7.2 Beetle mortality.xls'
df = get_data(inFile)
# adjust the unusual column names in the Excel file
colNames = [name.split(',')[1].lstrip() for name in df.columns.values]
df.columns = colNames
# fit the model
df['tested'] = df['n']
df['killed'] = df['y']
df['survived'] = df['tested'] - df['killed']
model = smf.glm('survived + killed ~ x',
data=df,
family=sm_families.Binomial()).fit()
print(model.summary())
print('-' * 65)
print('Equivalent solution:')
model = smf.glm('I(n - y) + y ~ x', data=df,
family=sm_families.Binomial()).fit()
print(model.summary())
# The fitted number of survivors can be obtained by
fits = df['n'] * (1 - model.fittedvalues)
print('Fits Logit:')
print(fits)
# The fits for other link functions are:
model_probit = smf.glm('I(n - y) + y ~ x',
data=df,
family=sm_families.Binomial(
sm_families.links.probit)).fit()
print(model_probit.summary())
fits_probit = df['n'] * (1 - model_probit.fittedvalues)
print('Fits Probit:')
print(fits_probit)
model_cll = smf.glm('I(n - y) + y ~ x',
data=df,
family=sm_families.Binomial(
sm_families.links.cloglog)).fit()
print(model_cll.summary())
fits_cll = df['n'] * (1 - model_cll.fittedvalues)
print('Fits Extreme Value:')
print(fits_cll)
def __init__(self,
endog,
exog,
exog_vc,
ident,
vcp_p=1,
fe_p=2,
fep_names=None,
vcp_names=None,
vc_names=None):
super(BinomialBayesMixedGLM, self).__init__(endog,
exog,
exog_vc=exog_vc,
ident=ident,
vcp_p=vcp_p,
fe_p=fe_p,
family=families.Binomial(),
fep_names=fep_names,
vcp_names=vcp_names,
vc_names=vc_names)
if not np.all(np.unique(endog) == np.r_[0, 1]):
msg = "endog values must be 0 and 1, and not all identical"
raise ValueError(msg)
def from_formula(cls,
formula,
vc_formulas,
data,
vcp_p=1,
fe_p=2,
vc_names=None):
fam = families.Binomial()
x = _BayesMixedGLM.from_formula(formula,
vc_formulas,
data,
family=fam,
vcp_p=vcp_p,
fe_p=fe_p,
vc_names=vc_names)
return BinomialBayesMixedGLM(endog=x.endog,
exog_fe=x.exog_fe,
exog_vc=x.exog_vc,
ident=x.ident,
vcp_p=x.vcp_p,
fe_p=x.fe_p,
fep_names=x.fep_names,
vcp_names=x.vcp_names,
vc_names=x.vc_names)
def setup_class(cls):
df = data_bin
res = GLM(df['constrict'], df[['const', 'log_rate', 'log_volumne']],
family=families.Binomial()).fit(attach_wls=True, atol=1e-10)
cls.infl1 = res.get_influence()
cls.infl0 = MLEInfluence(res)
def setup_class(cls):
cls.idx = slice(None) # params sequence same as Stata
#res1ul = Logit(data.endog, data.exog).fit(method="newton", disp=0)
cls.res2 = reslogit.results_constraint2_robust
mod1 = GLM(spector_data.endog,
spector_data.exog,
family=families.Binomial())
# not used to match Stata for HC
# nobs, k_params = mod1.exog.shape
# k_params -= 1 # one constraint
cov_type = 'HC0'
cov_kwds = {'scaling_factor': 32 / 31}
# looks like nobs / (nobs - 1) and not (nobs - 1.) / (nobs - k_params)}
constr = 'x1 - x3 = 0'
cls.res1m = mod1.fit_constrained(constr,
cov_type=cov_type,
cov_kwds=cov_kwds,
atol=1e-10)
R, q = cls.res1m.constraints.coefs, cls.res1m.constraints.constants
cls.res1 = fit_constrained(mod1,
R,
q,
fit_kwds={
'atol': 1e-10,
'cov_type': cov_type,
'cov_kwds': cov_kwds
})
cls.constraints_rq = (R, q)
def fit_logistic(X_hold,Y_hold,Firth=False,resBase=None,LRtest=True):
"""
Fits a logistic regression model using standard (when Firth = False) or Firth's method (when Firth = True).
resBase is the result of a previous call to a regression that is used to store data for Firth's method.
LRtest indicates if the likelihood ratio test should be reported.
"""
if not Firth:
res = GLM(Y_hold, X_hold, family=families.Binomial()).fit()#XXX Confirm this with logistic using older XXXX
# AICc adjustment
res.aicc = statsmodels.tools.eval_measures.aicc(res.llf, nobs=res.nobs, df_modelwc=res.df_model+1)
# Correct BIC
res.bic = statsmodels.tools.eval_measures.bic(res.llf, nobs=res.nobs, df_modelwc=res.df_model+1)
else:
if resBase is None:
sys.stderr.write('resBase must be provided to do Firth regression\n')
sys.exit(1)
elif type(resBase) is not statsmodels.genmod.generalized_linear_model.GLMResultsWrapper:
sys.stderr.write('resBase must be type statsmodels.genmod.generalized_linear_model.GLMResultsWrapper\n')
sys.exit(2)
else:
res = resBase
#Do Firth's logistic regression
(rint, rbeta, rbse, rfitll, pi) = fit_firth(Y_hold, X_hold, start_vec = None)
if LRtest:
# LRT
null_X = np.delete(arr=X_hold,obj=range(int(np.size(X_hold)/len(X_hold)))[1:int(np.size(X_hold)/len(X_hold))],axis=1)
(null_intercept, null_beta, null_bse, null_fitll, null_pi) = fit_firth(Y_hold, null_X, start_vec = None)
lrstat = -2.*(null_fitll - rfitll)
lrt_pvalue = 1.
if lrstat > 0.: # non-convergence
lrt_pvalue = stats.chi2.sf(lrstat, 1)
res.llnull = null_fitll
res.lrstat = lrstat
res.lrt_pval = lrt_pvalue
# AICc adjustment for Firth model
aicc = statsmodels.tools.eval_measures.aicc(rfitll, nobs=len(Y_hold), df_modelwc=np.shape(X_hold)[1])
# AIC
aic = statsmodels.tools.eval_measures.aic(rfitll, nobs=len(Y_hold), df_modelwc=np.shape(X_hold)[1])
# BIC
bic = statsmodels.tools.eval_measures.bic(rfitll, nobs=len(Y_hold), df_modelwc=np.shape(X_hold)[1])
#Store parameters, standard errors, likelihoods, and statistics
rint = np.array([rint])
rbeta = np.array(rbeta)
res.params = np.concatenate([rint,rbeta])
res.bse = rbse
res.llf = rfitll
res.aicc = aicc
res.aic = aic
res.bic = bic
#Get Wald p vals for parameters
res.pvalues = 1. - chi2.cdf(x=(res.params/res.bse)**2, df=1)
#Add predicted y
res.predict = pi
return res
def iterate_logistic(X_hold,Y_hold, fixed_columns = [0], Firth=False):
"""
Fits logistic regression to the provided data while using the fixed_columns in the regression.
Firth specifies if Firth regression should be used.
Returns matrices of fitted betas, pvalues, aic, aicc (second order aic), and bic
"""
l = np.size(fixed_columns)+1
k = np.shape(X_hold)[1]
betas = np.zeros([k,l])
pvalues = np.zeros([k,l])
aic = np.zeros([k,1])
aicc = np.zeros([k,1])
bic = np.zeros([k,1])
# Fit constant
if Firth:
null_X = np.delete(arr=X_hold,obj=range(int(np.size(X_hold)/len(X_hold)))[1:int(np.size(X_hold)/len(X_hold))],axis=1)
(null_intercept, null_beta, null_bse, null_fitll, null_pi) = fit_firth(Y_hold, null_X, start_vec = None)
#Using this as a way to return a model in the same class as GLM.
res = GLM(Y_hold, null_X, family=families.Binomial()).fit()
# AICc adjustment for Firth model
res.aicc = statsmodels.tools.eval_measures.aicc(null_fitll, nobs=res.nobs, df_modelwc=res.df_model+1)
# AIC
res.aic = statsmodels.tools.eval_measures.aic(null_fitll, nobs=res.nobs, df_modelwc=res.df_model+1)
# BIC
res.bic = statsmodels.tools.eval_measures.bic(null_fitll, nobs=res.nobs, df_modelwc=res.df_model+1)
#Store parameters, standard errors, likelihoods, and statistics
res.params = np.array([null_intercept])
#Get Wald p vals for parameters
res.pvalues = 1. - chi2.cdf(x=(res.params/null_bse)**2, df=1)
else:
res = fit_logistic(X_hold[:,0],Y_hold)
betas[0,:] = res.params
pvalues[0,:] = res.pvalues
aic[0] = res.aic
aicc[0] = res.aicc
bic[0] = res.bic
#Set variable for use later
resBase = copy.deepcopy(res)
NAN = ~np.isnan(X_hold).any(axis=0)
for i in range(1,k):
if NAN[i]:
if i not in fixed_columns:
columns = fixed_columns.copy()
columns.append(i)
res = fit_logistic(X_hold[:,columns],Y_hold, Firth=Firth, resBase=resBase,LRtest=False)
betas[i,:] = res.params
pvalues[i,:] = res.pvalues
aic[i] = res.aic
aicc[i] = res.aicc
bic[i] = res.bic
return betas, pvalues,aic,aicc,bic
def setup_class(cls):
endog_bin = (endog > endog.mean()).astype(int)
cls.cov_type = 'cluster'
mod1 = GLM(endog_bin, exog, family=families.Binomial())
cls.res1 = mod1.fit(cov_type='cluster', cov_kwds=dict(groups=group))
mod1 = smd.Logit(endog_bin, exog)
cls.res2 = mod1.fit(cov_type='cluster', cov_kwds=dict(groups=group))
def __init__(self, endog, exog_fe, exog_vc, ident, vcp_p=1,
fe_p=2, fep_names=None, vcp_names=None,
vc_names=None):
super(BinomialBayesMixedGLM, self).__init__(
endog=endog, exog_fe=exog_fe, exog_vc=exog_vc,
ident=ident, vcp_p=vcp_p, fe_p=fe_p,
family=families.Binomial(),
fep_names=fep_names, vcp_names=vcp_names,
vc_names=vc_names)
def setup_class(cls):
df = data_bin
mod = GLM(df['constrict'], df[['const', 'log_rate', 'log_volumne']],
family=families.Binomial())
res = mod.fit(method="newton", tol=1e-10)
from statsmodels.discrete.discrete_model import Logit
mod2 = Logit(df['constrict'], df[['const', 'log_rate', 'log_volumne']])
res2 = mod2.fit(method="newton", tol=1e-10)
cls.infl1 = res.get_influence()
cls.infl0 = res2.get_influence()
def setup_class(cls):
endog_bin = (endog > endog.mean()).astype(int)
cls.cov_type = 'cluster'
mod1 = GLM(endog_bin, exog, family=families.Binomial(link=links.probit()))
cls.res1 = mod1.fit(method='newton',
cov_type='cluster', cov_kwds=dict(groups=group))
mod1 = smd.Probit(endog_bin, exog)
cls.res2 = mod1.fit(cov_type='cluster', cov_kwds=dict(groups=group))
cls.rtol = 1e-6
def senility_and_WAIS():
'''Another example of logistic regression.
chapter 7.8, p 143
[tbd]: I don't understand how the "Binomial model" (grouped response)
is supposed to work, in either language'''
inFile = r'GLM_data/Table 7.8 Senility and WAIS.xls'
df = get_data(inFile)
# ungrouped
model = smf.glm('s ~ x', data=df, family=sm_families.Binomial()).fit()
print(model.summary())
def setup_class(cls):
yi = np.array([0, 2, 14, 19, 30])
ni = 40 * np.ones(len(yi))
xi = np.arange(1, len(yi) + 1)
exog = np.column_stack((np.ones(len(yi)), xi))
endog = np.column_stack((yi, ni - yi))
res = GLM(endog, exog, family=families.Binomial()).fit()
cls.infl1 = res.get_influence()
cls.infl0 = MLEInfluence(res)
cls.cd_rtol = 5e-5
def setup_class(cls):
cls.idx = slice(None) # params sequence same as Stata
#res1ul = Logit(data.endog, data.exog).fit(method="newton", disp=0)
cls.res2 = reslogit.results_constraint2
mod1 = GLM(spector_data.endog, spector_data.exog,
family=families.Binomial())
constr = 'x1 - x3 = 0'
cls.res1m = mod1.fit_constrained(constr, atol=1e-10)
R, q = cls.res1m.constraints.coefs, cls.res1m.constraints.constants
cls.res1 = fit_constrained(mod1, R, q, fit_kwds={'atol': 1e-10})
cls.constraints_rq = (R, q)
def setup_class(cls):
cls.idx = slice(None) # params sequence same as Stata
cls.res2 = reslogit.results_constraint2
mod1 = GLM(spector_data.endog, spector_data.exog,
family=families.Binomial())
constr = 'x1 - x3 = 0'
cls.res1m = mod1.fit_constrained(constr, atol=1e-10)
# patsy compatible constraints
R, q = cls.res1m.constraints.coefs, cls.res1m.constraints.constants
cls.res1 = fit_constrained(mod1, R, q, fit_kwds={'atol': 1e-10})
cls.constraints_rq = (R, q)
def setup_class(cls):
cls.idx = slice(None)
# params sequence same as Stata, but Stata reports param = nan
# and we have param = value = 0
cls.res2 = reslogit.results_constraint1
mod1 = GLM(spector_data.endog, spector_data.exog,
family=families.Binomial())
constr = 'x1 = 2.8'
cls.res1m = mod1.fit_constrained(constr)
R, q = cls.res1m.constraints
cls.res1 = fit_constrained(mod1, R, q)
def reply_analysis_report(data_input_path, data_output_path):
reply_analysis = prepare_data_reply_analysis(data_input_path,
data_output_path)
score = reply_analysis.assign(
tweet_negative_score=lambda df: df.apply(
lambda x: x["tweet_score"]
if x["tweet_label"] == "NEGATIVE" else 1 - x["tweet_score"],
axis=1),
trump_negative_score=lambda df: df.apply(
lambda x: x["trump_score"]
if x["trump_label"] == "NEGATIVE" else 1 - x["trump_score"],
axis=1))
logits = scipy.special.logit(
score[["negative_score_retweet", "negative_score_trump"]])
logits.plot(kind="scatter",
x="negative_score_trump",
y="negative_score_retweet",
alpha=0.1)
logits.save("plots/logit_sentiment_score.png")
print(
"Naive Sentiment Score Calculation",
scipy.stats.pearsonr(logits["negative_score_trump"],
logits["negative_score_retweet"]))
tmp_data = reply_analysis[~reply_analysis["trump_label"].isnull()]
x = sm.add_constant(tmp_data[[
# "created_at_trump_day", "created_at_trump_month", "created_at_trump_year",
"followers_count_norm",
"friends_count_norm",
"listed_count_norm",
"statuses_count_norm"
]])
res = GLM(tmp_data['trump_label'].astype("category").cat.codes,
x,
family=families.Binomial()).fit(attach_wls=True, atol=1e-10)
print(res.summary())
print(
"ATE",
backdoor_binary_respose_ate(reply_analysis, "trump_label", "NEGATIVE",
"tweet_label", "NEGATIVE",
"trump_created_at", "tweet_created_at",
datetime.timedelta(minutes=0), "1D"))
def probit_fit(x, resp):
'''
Probit fit with 95% CIs
'''
# binomial GLM with probit link
model = GLM(resp,
add_constant(x),
family=families.Binomial(),
link=families.links.probit())
mod_result = model.fit(disp=0)
xt = np.linspace(np.min(x), np.max(x), 100)
r_hat = mod_result.predict(add_constant(xt))
pred_summ = mod_result.get_prediction(
add_constant(xt)).summary_frame(alpha=0.05)
ci_5, ci_95 = pred_summ['mean_ci_lower'], pred_summ['mean_ci_upper']
return mod_result.params, r_hat, (xt, ci_5, ci_95)
def test_influence_glm_bernoulli():
# example uses Finney's data and is used in Pregibon 1981
df = data_bin
results_sas = np.asarray(results_sas_df)
res = GLM(df['constrict'], df[['const', 'log_rate', 'log_volumne']],
family=families.Binomial()).fit(attach_wls=True, atol=1e-10)
infl = res.get_influence(observed=False)
k_vars = 3
assert_allclose(infl.dfbetas, results_sas[:, 5:8], atol=1e-4)
assert_allclose(infl.d_params, results_sas[:, 5:8] * res.bse.values, atol=1e-4)
assert_allclose(infl.cooks_distance[0] * k_vars, results_sas[:, 8], atol=6e-5)
assert_allclose(infl.hat_matrix_diag, results_sas[:, 4], atol=6e-5)
c_bar = infl.cooks_distance[0] * 3 * (1 - infl.hat_matrix_diag)
assert_allclose(c_bar, results_sas[:, 9], atol=6e-5)
def from_formula(cls, formula, vc_formulas, data, vcp_p=1, fe_p=2):
fam = families.Binomial()
x = _BayesMixedGLM.from_formula(
formula, vc_formulas, data, family=fam, vcp_p=vcp_p, fe_p=fe_p)
# Copy over to the intended class structure
mod = BinomialBayesMixedGLM(
x.endog,
x.exog,
exog_vc=x.exog_vc,
ident=x.ident,
vcp_p=x.vcp_p,
fe_p=x.fe_p,
fep_names=x.fep_names,
vcp_names=x.vcp_names,
vc_names=x.vc_names)
mod.data = x.data
return mod
def setup_class(cls):
vs = Independence()
family = families.Binomial()
np.random.seed(987126)
Y = 1 * (np.random.normal(size=100) < 0)
X1 = np.random.normal(size=100)
X2 = np.random.normal(size=100)
X3 = np.random.normal(size=100)
groups = np.random.randint(0, 4, size=100)
D = pd.DataFrame({"Y": Y, "X1": X1, "X2": X2, "X3": X3})
mod1 = GEE.from_formula("Y ~ X1 + X2 + X3",
groups,
D,
family=family,
cov_struct=vs)
cls.result1 = mod1.fit()
mod2 = GLM.from_formula("Y ~ X1 + X2 + X3", data=D, family=family)
cls.result2 = mod2.fit(disp=False)
def test_score_binomial(miss_frac):
np.random.seed(23424)
n, p = 100, 5
for d in range(1, 5):
# Generate the data
icept = np.linspace(3, 5, p)
fac = np.random.normal(size=(p, d))
fac, _, _ = np.linalg.svd(fac, 0)
sc = np.random.normal(size=(n, d))
lp = np.dot(sc, fac.T) + icept
mu = 1 / (1 + np.exp(-lp))
endog = (np.random.uniform(size=(n, p)) < mu).astype(np.float64)
valid = (np.random.uniform(size=(n, p)) > miss_frac).astype(np.bool)
pca = GPCA(endog, d, family=families.Binomial(), valid=valid)
par = np.concatenate((icept, fac.ravel()))
grad = pca.score(par)
ngrad = nd.Gradient(pca.loglike)(par)
assert_allclose(grad, ngrad, rtol=1e-4, atol=1e-4)
def __init__(self):
self.setup_class() # why does nose do it properly
from statsmodels.genmod.generalized_linear_model import GLM
from statsmodels.genmod import families
self.mod = lambda y, x: GLM(y, x, family=families.Binomial())
self.y = self.y_bin
def exercise7d1():
'''Logistic regression example
chapter 7.3, p 130
[tbd]: the cloglog values are inconsistent with those mentioned in the book.
This is probably due to the specific definitions of "loglog" and "cloglog"
in the respective languages.
'''
inFile = r'GLM_data/Table 7.11 Hiroshima deaths.xls'
df = get_data(inFile)
df['radBin'] = np.array([0,1,10,50,100,200])
# adjust the unusual column names in the Excel file
colNames = df.columns.values
colNames[2] = 'other'
colNames[3] = 'total'
df.columns = colNames
df['n'] = df['total']
df['y'] = df['leukemia']
model = smf.glm('other + leukemia ~ radBin', data=df, family=sm_families.Binomial()).fit()
print(model.summary())
print('-'*65)
print('Equivalent solution:')
model = smf.glm('I(n - y) + y ~ radBin', data=df, family=sm_families.Binomial()).fit()
print(model.summary())
# The fitted number of survivors can be obtained by
fits = df['n']*(1-model.fittedvalues)
print('Fits Logit:')
print(fits)
# The fits for other link functions are:
model_probit = smf.glm('I(n - y) + y ~ radBin', data=df, family=sm_families.Binomial(sm_families.links.probit)).fit()
print(model_probit.summary())
fits_probit = df['n']*(1-model_probit.fittedvalues)
print('Fits Probit:')
print(fits_probit)
model_cll = smf.glm('I(n - y) + y ~ radBin', data=df, family=sm_families.Binomial(sm_families.links.cloglog)).fit()
print(model_cll.summary())
fits_cll = df['n']*(1-model_cll.fittedvalues)
print('Fits Extreme Value:')
print(fits_cll)
x = np.arange(201)
y = np.exp(model.params[0]+model.params[1]*x)/(1+np.exp(model.params[0]+model.params[1]*x))
yUp = np.exp(model.params[0]+2*model.bse[0]+(model.params[1]+2*model.bse[1])*x)/(1+np.exp(model.params[0]+2*model.bse[0]+(model.params[1]+2*model.bse[1])*x))
yLo = np.exp(model.params[0]-2*model.bse[0]+(model.params[1]-2*model.bse[1])*x)/(1+np.exp(model.params[0]-2*model.bse[0]+(model.params[1]-2*model.bse[1])*x))
plt.plot(df['radBin'],df['leukemia']/df['total'],'.')
plt.plot(x,1-y,linewidth=0.5,color='black',label='Logit Fit')
plt.plot(x,1-yUp,'--',linewidth=0.5,color='red',label='2 Sigma Bounds')
plt.plot(x,1-yLo,'--',linewidth=0.5,color='red')
plt.title('Logit Fit')
plt.xlabel('Radiation Dose')
plt.ylabel('Probability of death from Leukemia')
plt.legend()
plt.figure()
y = norm.cdf(model_probit.params[0]+model_probit.params[1]*x)
yUp = norm.cdf(model_probit.params[0]+2*model_probit.bse[0]+(model_probit.params[1]+2*model_probit.bse[1])*x)
yLo = norm.cdf(model_probit.params[0]-2*model_probit.bse[0]+(model_probit.params[1]-2*model_probit.bse[1])*x)
plt.plot(df['radBin'],df['leukemia']/df['total'],'.')
plt.plot(x,1-y,linewidth=0.5,color='black',label='Probit Fit')
plt.plot(x,1-yUp,'--',linewidth=0.5,color='red',label='2 Sigma Bounds')
plt.plot(x,1-yLo,'--',linewidth=0.5,color='red')
plt.title('Probit Fit')
plt.xlabel('Radiation Dose')
plt.ylabel('Probability of death from Leukemia')
plt.legend()
plt.figure()
y = 1-np.exp(-(np.exp(model_cll.params[0]+model_cll.params[1]*x)))
yUp = 1-np.exp(-(np.exp(model_cll.params[0]+2*model_cll.bse[0]+(model_cll.params[1]+2*model_cll.bse[1])*x)))
yLo = 1-np.exp(-(np.exp(model_cll.params[0]-2*model_cll.bse[0]+(model_cll.params[1]-2*model_cll.bse[1])*x)))
plt.plot(df['radBin'],df['leukemia']/df['total'],'.')
plt.plot(x,1-y,linewidth=0.5,color='black',label='CLL Fit')
plt.plot(x,1-yUp,'--',linewidth=0.5,color='red',label='2 Sigma Bounds')
plt.plot(x,1-yLo,'--',linewidth=0.5,color='red')
plt.title('CLL Fit')
plt.xlabel('Radiation Dose')
plt.ylabel('Probability of death from Leukemia')
plt.legend()
pdb.set_trace()
def __init__(self, fam, nb_theta=None, mult_n=None):
if fam == "poi":
self.family = smf.Poisson()
elif fam == "nb":
if nb_theta is None:
raise GlmpcaError(
"Negative binomial dispersion parameter 'nb_theta' must be specified"
)
self.family = smf.NegativeBinomial(alpha=1 / nb_theta)
elif fam in ("mult", "bern"):
self.family = smf.Binomial()
if fam == "mult" and mult_n is None:
raise GlmpcaError(
"Multinomial sample size parameter vector 'mult_n' must be specified"
)
else:
raise GlmpcaError("unrecognized family type")
#variance function, determined by GLM family
vfunc = self.family.variance
#inverse link func, mu as a function of linear predictor R
ilfunc = self.family.link.inverse
#derivative of inverse link function, dmu/dR
hfunc = self.family.link.inverse_deriv
self.glmpca_fam = fam
if fam == "poi":
def infograd(Y, R):
M = ilfunc(R) #ilfunc=exp
return {"grad": (Y - M), "info": M}
elif fam == "nb":
def infograd(Y, R):
M = ilfunc(R) #ilfunc=exp
W = 1 / vfunc(M)
return {"grad": (Y - M) * W * M, "info": W * (M**2)}
self.nb_theta = nb_theta
elif fam == "mult":
def infograd(Y, R):
P = ilfunc(R) #ilfunc=expit, P very small probabilities
return {"grad": Y - (mult_n * P), "info": mult_n * vfunc(P)}
self.mult_n = mult_n
elif fam == "bern":
def infograd(Y, R):
P = ilfunc(R)
return {"grad": Y - P, "info": vfunc(P)}
else: #this is not actually used but keeping for future reference
#this is most generic formula for GLM but computationally slow
raise GlmpcaError("invalid fam")
def infograd(Y, R):
M = ilfunc(R)
W = 1 / vfunc(M)
H = hfunc(R)
return {"grad": (Y - M) * W * H, "info": W * (H**2)}
self.infograd = infograd
#create deviance function
if fam == "mult":
def dev_func(Y, R):
return mat_binom_dev(Y, ilfunc(R), mult_n)
else:
def dev_func(Y, R):
return self.family.deviance(Y, ilfunc(R))
self.dev_func = dev_func
def mod(y, x):
return GLM(y, x, family=families.Binomial())
from statsmodels.genmod.generalized_linear_model import GLM
from statsmodels.genmod import families
import statsmodels.stats.tests.test_influence
test_module = statsmodels.stats.tests.test_influence.__file__
cur_dir = cur_dir = os.path.abspath(os.path.dirname(test_module))
file_name = 'binary_constrict.csv'
file_path = os.path.join(cur_dir, 'results', file_name)
df = pd.read_csv(file_path, index_col=0)
res = GLM(
df['constrict'],
df[['const', 'log_rate', 'log_volumne']],
family=families.Binomial()).fit(
attach_wls=True, atol=1e-10)
print(res.summary())
# ## get the influence measures
#
# GLMResults has a `get_influence` method similar to OLSResults, that
# returns and instance of the GLMInfluence class. This class has methods and
# (cached) attributes to inspect influence and outlier measures.
#
# This measures are based on a one-step approximation to the the results
# for deleting one observation. One-step approximations are usually accurate
# for small changes but underestimate the magnitude of large changes. Event
# though large changes are underestimated, they still show clearly the
# effect of influential observations
#
plt.rc("figure", figsize=(16, 8))
plt.rc("font", size=14)
import statsmodels.stats.tests.test_influence
test_module = statsmodels.stats.tests.test_influence.__file__
cur_dir = cur_dir = os.path.abspath(os.path.dirname(test_module))
file_name = "binary_constrict.csv"
file_path = os.path.join(cur_dir, "results", file_name)
df = pd.read_csv(file_path, index_col=0)
res = GLM(
df["constrict"],
df[["const", "log_rate", "log_volumne"]],
family=families.Binomial(),
).fit(attach_wls=True, atol=1e-10)
print(res.summary())
# ## get the influence measures
#
# GLMResults has a `get_influence` method similar to OLSResults, that
# returns and instance of the GLMInfluence class. This class has methods and
# (cached) attributes to inspect influence and outlier measures.
#
# This measures are based on a one-step approximation to the the results
# for deleting one observation. One-step approximations are usually accurate
# for small changes but underestimate the magnitude of large changes. Event
# though large changes are underestimated, they still show clearly the
# effect of influential observations
#
if example == 1:
print "normal"
m = AdditiveModel(d)
m.fit(y)
x = np.linspace(-2,2,50)
print m
import scipy.stats, time
if example == 2:
print "binomial"
mod_name = 'Binomial'
f = families.Binomial()
#b = np.asarray([scipy.stats.bernoulli.rvs(p) for p in f.link.inverse(y)])
b = np.asarray([scipy.stats.bernoulli.rvs(p) for p in f.link.inverse(z)])
b.shape = y.shape
m = GAM(b, d, family=f)
toc = time.time()
m.fit(b)
tic = time.time()
print tic-toc
#for plotting
yp = f.link.inverse(y)
p = b
if example == 3:
print "Poisson"