-
Notifications
You must be signed in to change notification settings - Fork 0
/
bootstrap_utils.py
212 lines (170 loc) · 7.61 KB
/
bootstrap_utils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
# %%
from matplotlib import pyplot as plt
import importlib
import numpy as np
import pandas as pd
import xarray as xr
from scipy.stats import ttest_ind_from_stats
import plotnine as p9
# from sklearn.linear_model import LinearRegression
# import statsmodels.api as sm
import functools
import cftime
import os
import glob
import re
import copy
import datetime
import numba
from numba import jit,prange
import sys
sys.path.append("../")
import pooled_stats
#%%
# Calculates the slope of a regression line on the input n x 2 array
@jit(nopython=True)
def calc_slope(dfmat):
x = dfmat[:,0]
y = dfmat[:,1]
if np.sum(np.isnan(y)) > 3:
slope = np.nan
# slope = np.sum(np.isnan(y))
else:
# X = np.vstack([x, np.ones(len(x))]).T
X = np.vstack((x, np.ones_like(y))).T
# Perform the regression, betas come in inverse order (i.e. the last element is the constant)
betas = np.linalg.inv(X.T @ X) @ X.T @ y.T
slope = betas[0]
return(slope)
def calc_slope_wrap(df,timeindex,regvarindex):
dfmat = df.iloc[:,[timeindex,regvarindex]].to_numpy().astype("float64")
# print(dfmat)
return(calc_slope(dfmat))
# Calculates mean and variance of the slope of a regression line on the input n x 2 array
# using a nsamp bootsrap
@jit(nopython=True, fastmath=False)
def calc_slope_bootstrap(dfmat, nsamp = 30):
# Remove rows with nans
# This doesnt work because the numba version of any doesn't support the axis argument
# dfmat = dfmat[~np.isnan(dfmat).any(axis=1)]]
# msk = np.array(np.ones(len(dfmat)),dtype=np.bool)
msk = np.empty(len(dfmat),np.bool_) # Note the underscore on the dtype for numba to work
for i in np.arange(len(dfmat)):
msk[i] = ~np.any(np.isnan(dfmat[i,:]))
dfmat = dfmat[msk]
xin = dfmat[:,0]
yin = dfmat[:,1]
ylen = len(yin)
# Preallocate a vector to contain outputs
# TODO: A running variance implementation would be more efficient
# e.g. https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Welford's_online_algorithm
slopes = np.empty(nsamp)
for i in np.arange(nsamp):
inds = np.random.choice(np.arange(ylen), ylen)
x = np.take(xin,inds)
y = np.take(yin,inds)
X = np.vstack((x, np.ones_like(y))).T
det = np.linalg.det(X.T @ X) # Determinant to catch singular matrices
if (np.sum(np.isnan(y)) > 3) | (det == 0):
slope = np.nan
# slope = np.sum(np.isnan(y))
else:
# Perform the regression, betas come in inverse order (i.e. the last element is the constant)
betas = np.linalg.inv(X.T @ X) @ X.T @ y.T
slope = betas[0]
slopes[i] = slope
# print(slopes)
out = np.array((np.mean(slopes), np.var(slopes)))
# print(out)
return(out)
@jit(nopython = True)
def calc_slope_bootstrap_wrap(df,timeindex,regvarindex, nsamp = 30):
dfmat = df.iloc[:,[timeindex,regvarindex]].to_numpy().astype("float64")
# print(dfmat)
return(calc_slope_bootstrap(dfmat))
# Loops calculation for several variables
def calc_slope_bootstrap_wrap_loop(df,timeindex,regvarindexes, nsamp = 30):
nregs = len(regvarindexes)
outmat = np.zeros([nregs,2])
for i in np.arange(nregs):
dfmat = df.iloc[:,[timeindex,regvarindexes[i]]].to_numpy().astype("float64")
# print((dfmat))
if np.all(np.isnan(dfmat[:,1])):
outmat[i,:] = np.nan
else:
outmat[i,:] = calc_slope_bootstrap(dfmat, nsamp = nsamp)
return(outmat)
# Calculate linear regression slopes between a variable (timevarname) and several
# other variables (regvarnames), estimating coefficient variances using bootstrap
def df_apply_reg_bootstrap(indf, timevarname, regvarnames, group_coords=[], nsamp = 30):
usevarnames = group_coords + [timevarname] + regvarnames
# Reset indexes
rindf = indf.reset_index()
# Get column indexes to use numba
timeindex = rindf.columns.get_loc(timevarname)
regvarindexes = [rindf.columns.get_loc(i) for i in regvarnames]
# Grouped DataFrame, if asked
#FIXME: Breaks if we don't group, since the .apply used is specifically the GroupBy one
if len(group_coords) > 0:
grindf = rindf.groupby(group_coords)
else:
grindf = rindf
# Apply the regressions. This will return a Series of n x 2 arrays,
# n being the number of regvars, with mean and variance in the columns
regseries = grindf.apply(
functools.partial(calc_slope_bootstrap_wrap_loop, timeindex = timeindex, regvarindexes = regvarindexes, nsamp = nsamp)).rename("slope")
# Unpack the series into a Dataframe with Series of [mean,variance]
regdf = pd.concat([regseries.apply(lambda x: x[i,:]).rename(regvarnames[i]) for i in range(len(regvarnames))], axis=1)
# Unpack each variable in a list of Dataframes with a mean Series and a variance Series with the _var suffix
dflist = [
pd.concat([regdf[regvarnames[i]].apply(lambda x: x[0]),
regdf[regvarnames[i]].apply(lambda x: x[1]).rename(regvarnames[i]+"_var")], axis=1)
for i in range(len(regvarnames))
]
# Finally, merge all Dataframes in one
regvardf = functools.reduce(lambda x, y: pd.merge(x, y, left_index=True, right_index=True), dflist)
return(regvardf)
# Calculate regressions between each variable in a Dataset
# and the dimension timevarname, estimating the coefficient's
# variance using bootstrap
def ds_apply_reg_bootstrap(inputds, timevarname, nsamp = 30):
# If input is a DataArray, make it a Dataset
if isinstance(inputds, xr.DataArray):
inputds = inputds.to_dataset()
regvarnames = list(inputds.data_vars)
# Calculating one variable at a time and concatenating
# the results saves a lot of memory on Pandas Dataframe wrangling
# It also allows Dataarrays with different dimensions
bigoutds = xr.Dataset()
for regvarname in regvarnames:
vards = inputds[[regvarname]]
orig_coords = list(vards.coords)
group_coords = [i for i in orig_coords if i != timevarname]
inputdf = vards.to_dataframe()
bootregdf = df_apply_reg_bootstrap(inputdf, timevarname, [regvarname], group_coords, nsamp = nsamp)
outds = xr.Dataset.from_dataframe(bootregdf)
bigoutds = bigoutds.merge(outds)
# inputdf = inputds.to_dataframe()
# bootregdf = df_apply_reg_bootstrap(inputdf, timevarname, regvarnames, group_coords, nsamp = nsamp)
# outds = xr.Dataset.from_dataframe(bootregdf)
return(bigoutds)
# # bigds = xr.open_dataset("tests/poi.nc")
# bigds = xr.open_dataset("tests/trend_test.nc")
# inputds = bigds[["tempmean","agestimate"]].isel(scenario=[1,3], member=0)
# # inputds = bigds[["tempmean","vpdmean"]].isel(scenario=[1,3], member=0)
# # testds = bigds[["tempmean"]]
# # # testds = bigds[["tempmean","vpdmean"]]
# # # # inputds = testds
# timevarname = "year"
# nsamp = 100
# # # outds = ds_apply_reg_bootstrap(inputds, timevarname, nsamp)
# # testoutds = ds_apply_reg_bootstrap(testds, timevarname, nsamp)
# testoutds = ds_apply_reg_bootstrap(inputds, timevarname, nsamp)
# # testoutds = ds_apply_reg_bootstrap(bigds, timevarname, nsamp)
# # %timeit -n 1 -r 1 testoutds = ds_apply_reg_bootstrap(inputds, timevarname, nsamp)
# # # %timeit -n 1 -r 1 testoutds = ds_apply_reg_bootstrap(testds, timevarname, nsamp)
# # # %timeit -n 1 -r 1 testoutds = ds_apply_reg_bootstrap(testds, timevarname, nsamp)
# # # outds["tempmean"].plot(row="scenario")
# testoutds["agestimate"].plot(col="scenario",row = "statmodel")
# # inputds.polyfit(dim = "year", deg =1)["tempmean_polyfit_coefficients"].sel(degree=1).plot(row="scenario")
# # %%