Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And...
Transcript of Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And...
![Page 1: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/1.jpg)
1 / 41
Week 2, Lecture 2 - Fitting And Regression
Aaron Meyer
![Page 2: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/2.jpg)
2 / 41
Outline
I Administrative IssuesI Project ProposalsI Fitting
I Ordinary Least SquaresI Non-Linear Least SquaresI Generalized Linear Model
I Some Examples
![Page 3: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/3.jpg)
3 / 41
Project Proposals
In this project, you have two options for the general route you cantake:
1. Reimplement analysis from the literature.2. New, exploratory analysis of existing data.
More details at:
https://aarmey.github.io/ml-for-bioe/final-project/
![Page 4: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/4.jpg)
4 / 41
Project ProposalsThe proposal should be less than two pages and describe the follow-ing items:
I Why the topic you chose is interestingI Demonstrate that your project fits the criteria aboveI What overall approach do you plan to take for the project and
whyI Demonstrate that your project can be finished within a monthI Estimate the difficulty of your project
We are available to discuss your ideas whenever you are ready,and you should discuss your idea with us prior to submittingyour proposal.
Recommend an early start—the earlier you finalize a proposal thesooner you can begin the project.
![Page 5: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/5.jpg)
5 / 41
Goal Of Fitting
I Fitting is the process of comparing a model to a compendiumof data
I After fitting, we will have a model that explains existing dataand can predict new data
![Page 6: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/6.jpg)
6 / 41
Process of Fitting
The process of fitting is nothing more than finding the maximumlikelihood distribution of models for a set of points.
The key factor is how one defines the problem—i.e. how the distri-bution is described.
![Page 7: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/7.jpg)
7 / 41
Caveats
I Any fitting result is highly dependent upon the correctness ofthe model
I Successful fitting requires concordance between the model anddataI Too little data and a model is underdeterminedI Unaccounted for variables can lead to systematic error
Since all models are wrong the scientist cannot obtain a“correct” one by excessive elaboration. On the contraryfollowing William of Occam he should seek an economi-cal description of natural phenomena. Just as the abilityto devise simple but evocative models is the signature ofthe great scientist so overelaboration and overparameter-ization is often the mark of mediocrity. ~George Box, JAmerican Stat Assoc, 1976
![Page 8: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/8.jpg)
8 / 41
Any Fitting Is Dependent On The Correctness Of TheModel
Sociologydoctoratesawarded(U
S)Worldwidenon-commercialspacelaun
ches
Worldwidenon-commercialspacelaunchescorrelateswith
Sociologydoctoratesawarded(US)
Sociologydoctoratesawarded(US) Worldwidenon-commercialspacelaunches
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
500Degreesawarded
550Degreesawarded
600Degreesawarded
650Degreesawarded
700Degreesawarded
40Launches
50Launches
30Launches
60Launches
tylervigen.com
Fitting does not happen in a vacuum!
![Page 9: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/9.jpg)
9 / 41
Ordinary Least Squares
I Probably the most widely used estimation technique.I Based on extending the maximum likelihood estimate of a
distribution.I Model assumes output quantity is linear combination of
inputs.
![Page 10: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/10.jpg)
10 / 41
Ordinary Least Squares
If we have a vector of n observations y, our predictions are going tofollow the form:
y = Xβ + ε
Here X is a n×p structure matrix, β is a p-dimensional vector withthe parameters of our model, and ε = (ε1, ε2, ...εn)′ is the noisepresent in the model.
![Page 11: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/11.jpg)
11 / 41
Ordinary Least Squares
y = Xβ + ε
ε is usually handled to be uncorrelated random components withconstant variance σ2:
ε ∼(0, σ2I
)
![Page 12: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/12.jpg)
12 / 41
Ordinary Least Squares
Single variable caseThe structure matrix is little more than the data, sometimes trans-formed, usually with an offset. So, another way to write:
y = Xβ + ε
would be:
y = m1x1 +m2x2 . . .+ b+ ε
![Page 13: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/13.jpg)
13 / 41
Ordinary Least Squares
Single variable case
y = mx + b+ ε
The values ofm and b that minimize the distance from y are optimal,and they don’t depend on ε.
![Page 14: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/14.jpg)
14 / 41
Ordinary Least Squares
Gauss and Markov in the early 1800s identified that the least squaresestimate of β, β̂, is:
β̂ = arg minβ‖y−Xβ‖2
![Page 15: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/15.jpg)
15 / 41
Ordinary Least Squares
β̂ = arg minβ‖y−Xβ‖2
can be directly calculated by:
β̂ = S−1X′y
where
S = X′X
![Page 16: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/16.jpg)
16 / 41
Ordinary Least Squares
β̂ is the maximum likelihood estimate of β, and has covariancematrix σ2S−1:
β̂ ∼(β, σ2S−1
)
In the normal case (when our assumptions hold), β̂ is an unbiasedestimator of β. Making these calculations tractable for larger datasets used to be a challenge but is now trivial.
![Page 17: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/17.jpg)
17 / 41
Ordinary Least Squares
Likelihood of Model
−n2 log(σ2)− 1
2σ2
n∑i=1
(yi − xiβ)2
therefore, only considering β (the only factor that influences predic-tions), we need to minimize:
n∑i=1
(yi − xiβ)2
Exactly how we calculate β!
![Page 18: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/18.jpg)
18 / 41
Ordinary Least Squares
What might be some advantages of a method such as this?
![Page 19: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/19.jpg)
19 / 41
Ordinary Least Squares
What are some of the assumptions?
What are the implications of these assumptions not holding?
What are some downsides?
![Page 20: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/20.jpg)
20 / 41
Application: Paternal de novo mutations
Questions:I Where do de novo mutations arise?I Are there factors that influence the rate of de novo mutations
from one generation to another?
![Page 21: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/21.jpg)
21 / 41
Application: Paternal de novo mutations
Figure: By Rdbickel - Own work, CC BY-SA 4.0,https://commons.wikimedia.org/w/index.php?curid=49599354
![Page 22: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/22.jpg)
22 / 41
Application: Paternal de novo mutations
Figure: Kong et al, Nature, 2012
![Page 23: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/23.jpg)
23 / 41
Application: Paternal de novo mutations
Figure: Kong et al, Nature, 2012
![Page 24: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/24.jpg)
24 / 41
Implementation
sklearn provides a very basic function for ordinary least squares.
I sklearn.linear_model.LinearRegressionI fit_intercept: Should an intercept value be fit?I normalize: Should the input variables be mean and variance
scaled?I No tests for significance/model performance included.I We’ll discuss evaluating the model in depth later.
Or there’s an even more bare function in numpy numpy.linalg.lstsq.
I Takes input variables a and b.I Solves the equation ax = b by computing a vector x that
minimizes the Euclidean 2-norm ‖b− ax‖2.
![Page 25: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/25.jpg)
25 / 41
Implementation
import sklearn as sk, matplotlib.pyplot as plt
lr = sk.linear_model.LinearRegression()boston = sk.datasets.load_boston()y = boston.target
lr.fit(boston.data, y)
predicted = lr.predict(boston.data)
fig, ax = plt.subplots()ax.scatter(y, predicted, edgecolors=(0, 0, 0))ax.plot([y.min(), y.max()], [y.min(), y.max()], 'k--', lw=4)ax.set_xlabel('Measured')ax.set_ylabel('Predicted')plt.show()
![Page 26: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/26.jpg)
26 / 41
Implementation
![Page 27: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/27.jpg)
27 / 41
Non-Linear Least Squares
Non-Linear Least Squares makes similar assumptions to ordinaryleast squares, but for arbitrary functions. Thus, instead of followingthe form:
y = Xβ + ε
Our input-output relationship is:
y = f(X, β) + ε
for the same construction of ε.
![Page 28: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/28.jpg)
28 / 41
Transformation
NNLSQ used to be mostly performed by transforming one’s datainto a linear model.
E.g. taking the ratio of variables, or log-transforming them.
This is now considered bad practice.
Why?
![Page 29: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/29.jpg)
29 / 41
Non-Linear Least Squares
AlgorithmsWe again need to solve for β to minimize the sum of squared error:I There are many methods to solve these problems, and finding
the true minimum is not a trivial task.I We’re not going to cover how these algorithms work in depth.
![Page 30: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/30.jpg)
30 / 41
Non-Linear Least Squares
AlgorithmsOne property we can take advantage of is that the gradient of theSSE w.r.t. β at the minimum is zero (ri is the residual of the ithpoint):
∂S
∂βj= 2
∑i
ri∂ri∂βj
= 0
I ∂ri∂βj
is a function of both the nonlinear function and the data.I This can be expanded out through a first-order Taylor
approximation.I Doing so essentially performs ordinary least squares around
the current point, for the linearized function.
![Page 31: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/31.jpg)
31 / 41
Non-Linear Least Squares
Algorithms
∂S
∂βj= 2
∑i
ri∂ri∂βj
= 0
I ∂ri∂βj
is a function of both the nonlinear function and the data.I This can be expanded out through a first-order Taylor
approximation.I Doing so essentially performs ordinary least squares around
the current point, for the linearized function.I ∂ri
∂βj= −Jij , where J is the Jacobian of the function.
I Many NNLSQ solvers require J for this reason: can beapproximated by finite differences.
I Probably the most common method, Gauss-Newton, uses thisproperty with Newton’s method.
![Page 32: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/32.jpg)
32 / 41
Non-Linear Least Squares
Algorithms - Key TakawaysI Unlike ordinary least squares, no guarantee about finding the
optimal solution.I Depending upon the data and model, there may be many
local minima.I Exactly equivalent to shifting normal distributions up and
down around one’s data.
![Page 33: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/33.jpg)
33 / 41
Implementation
SciPy’s scipy.optimize.least_squares is a very capable imple-mentation.
I The main necessary parameters are:I fun, the functionI x0, an initial guess for the parameter values
I Note that fun should return a vector of the residualsI So it should handle all the data itself
![Page 34: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/34.jpg)
34 / 41
NNLSQ Example - Binding DataLet’s say we’re looking at a protein-protein interaction such as this:
plt.semilogx(X, Y,'.');plt.xlabel('Concentration [nM]')plt.ylabel('Binding')
![Page 35: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/35.jpg)
35 / 41
NNLSQ Example - Binding Data
We can predict the amount of binding we’d observe from a single-sitebinding model:
def klotz1(k1, lig):return (k1*lig)/(1 + k1*lig)
![Page 36: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/36.jpg)
36 / 41
NNLSQ Example - Binding Data
plt.semilogx(X,klotz1(1.,X),'.')plt.xlabel('Concentration [nM]')plt.ylabel('Binding')
![Page 37: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/37.jpg)
37 / 41
NNLSQ Example - Binding Data
SciPy asks for the residuals at each fitting point, so we need toconvert a prediction to that:
def ls_obj_k1(k1, ligs, data):return(data - klotz1(k1,ligs))
![Page 38: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/38.jpg)
38 / 41
NNLSQ Example - Binding Datasp.optimize.least_squares(ls_obj_k1, 1., args=(X,Y))# --------active_mask: array([ 0.])
cost: 0.0086776496708916573fun: array([ 4.79e-05, 9.00e-05, -1.09e-04,8.04e-04, -9.67e-04, 3.85e-03,4.61e-03, 2.34e-03, 2.36e-02,9.64e-03, -2.48e-02, 1.93e-02,
-4.93e-02, 5.54e-02, -3.66e-02,2.97e-03, 3.39e-02, -8.74e-02])
grad: array([ -9.57228474e-09])jac: array([[-0.00099809],
[-0.00199235],[-0.0039695 ],
# ...[-0.03119024],[-0.01608763],[-0.00817133]])
message: '`gtol` termination condition is satisfied.'nfev: 4njev: 4
optimality: 9.5722847420895082e-09status: 1
success: Truex: array([ 0.95864059])
![Page 39: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/39.jpg)
39 / 41
Generalized Linear Model
What if the error term isn’t Gaussian?I In many cases linear regression can be inappropriate
I E.g. A measurement that is Poisson distributed
![Page 40: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/40.jpg)
40 / 41
Questions
I Would you expect the confidence interval of your model to belarger or smaller than the confidence interval of prediction?
I Given the binding data presented here, do you think a leastsquares model is most appropriate?
I How might you test whether your data fits the model you’vespecified?
![Page 41: Week 2, Lecture 2 - Fitting And Regression · 2021. 2. 23. · Week 2, Lecture 2 - Fitting And Regression Author: Aaron Meyer Created Date: 20210223042403Z ...](https://reader036.fdocuments.net/reader036/viewer/2022071500/611e88e5bfe3b84497714d74/html5/thumbnails/41.jpg)
41 / 41
Further Reading
I Computer Age Statistical Inference, Chapter 8I sklearn: Linear Models