Probability and Statistics Experiments

with Python
by Thom Ives, Ph.D.

Find this on DagsHub too. On DagsHub, I am ThomIves, and this repo is

"Probability_and_Statistics_with_Python".

What is the motivation for such an approach? The approach of coding math from scratch
without libraries or modules? I like the way my dear friend and brother Manjunatha
Gummaraju says it best.

"Hand crafting (without libraries & automation) helps to get a firm grip on
the subject, nuances & its applications. It also helps probably to author new
innovative techniques from the ground up."

Calculating (Sample) Mean and Standard

Deviation
2
∑ xi ∑(xi − μ)
μ = σ = √
n n
In [ ]: def mean(X):
mu = sum(X) / len(X)

return mu

In [ ]: def standard_deviation(X, mu=None):

if not mu:
mu = mean(X)
sigma = (sum([(x - mu)**2 for x in X])/len(X))**0.5

return sigma

Normal Probability Distribution Function

2
1 x−μ

1 − ( )
2 σ
p = e
σ√2π

In [ ]: import matplotlib.pyplot as plt

def PDF(x, mean=0, std_dev=1):

# define e and pi explicitly
e = 2.718281828
pi = 3.1415927
# calculate in two steps
p = 1.0 / (std_dev * ((2 * pi) ** 0.5))
p *= e ** (-0.5 * ((x - mean)/std_dev)**2)

return p

X = [(x - 1000)/200 for x in list(range(2001))]

P = [PDF(x) for x in X]
plt.plot(X, P)
plt.title(label="Standard Normal Distribution")
plt.xlabel(xlabel="value")
plt.ylabel(ylabel="propability of value occurance")
plt.show()
 Cummulative Normal Distribution
Function
xright 2
1 x−μ
1 − ( )
2 σ
cdf = ∫ e
xlef t σ√2π

In [ ]: import matplotlib.pyplot as plt

def PDF(x, mean=0, std_dev=1):

# define e and pi explicitly
e = 2.718281828
pi = 3.1415927
# calculate in two steps
p = 1.0 / (std_dev * ((2 * pi) ** 0.5))
p *= e ** (-0.5 * ((x - mean)/std_dev)**2)

return p

def CDF(mean=0, std_dev=1, x_left=-5, x_right=5, width=0.0001):

CDF = 0
X = [] # for plotting only
CDF_y = [] # for plotting only

x = x_left + width / 2
while x < x_right:
X.append(x) # for plotting only
panel = PDF(x, mean, std_dev) * width # panel under PDF
CDF += panel # running sum of panels = integration
CDF_y.append(CDF) # for plotting only
x += width # current x value

return CDF, X, CDF_y

 total_integral, X, CDF_y = CDF()
P = [PDF(x) for x in X]
total_integral = round(total_integral, 5)
msg = f'Total integral of PDF = {total_integral}'
print(msg)
plt.plot(X, P)
plt.plot(X, CDF_y)
plt.show()

Total integral of PDF = 1.0

The Beta Distribution

1 α−1 β−1
f (x, α, β) = x (1 − x)
B(α, β)

1
α−1 β−1
B(α, β) = ∫ t (1 − t) dt
0

In [ ]: import matplotlib.pyplot as plt

class Beta_Distribution:
def __init__(self, alpha, beta, panels=10000):
self.alpha = alpha
self.beta = beta
self.panels = panels
self.__Beta_Function__()

def __Beta_Function__(self):
width = 1 / self.panels
X = [x/self.panels for x in range(self.panels)]
# makes total integral of beta_PDF sum to 1
self.B = sum(
[(x**(self.alpha - 1) * \
(1 - x)**(self.beta - 1)) * width
for x in X])
 def beta_PDF(self, x):
return x**(self.alpha - 1) * \
(1 - x)**(self.beta - 1) / self.B

X = [x/1000 for x in range(1000+1)]

bd = Beta_Distribution(5, 2)
Y1 = [bd.beta_PDF(x) for x in X]
Y_integral = round(sum([y*0.001 for y in Y1]), 3)
bd = Beta_Distribution(2, 5)
Y2 = [bd.beta_PDF(x) for x in X]

print(f"The total integral of beta_PDF is {Y_integral}")

plt.plot(X, Y1)
plt.plot(X, Y2)
plt.title(label="Skewed Distribution")
plt.xlabel(xlabel="values")
plt.ylabel(ylabel="probabilities")
plt.show()

The total integral of beta_PDF is 1.0

Student's T-Distribution
1

α−1 β−1
B(α, β) = ∫ t (1 − t)
0

ν+1
−
2 2
1 t
P DFt (t) = (1 + )
1 ν
√νB( , ) ν
2 2

In [ ]: class T_Distribution:
def __init__(self, dof=9):
self.beta = self.beta_function(0.5, dof/2)

self.front = 1 / (dof ** 0.5 * self.beta)

self.dof = dof
 self.power = -(dof + 1)/2

def beta_function(self, x, y):

pw = 1 / 1000000
beta = 0
t = pw / 2
while t < 1.0:
beta += t ** (x - 1) * (1 - t) ** (y - 1) * pw
t += pw

return beta

def PDFt(self, t):

# The t probability distribution method
f_of_t = self.front * (1 + t**2/self.dof) ** self.power

return f_of_t

def CDFt(self, t_left, t_right):

# The t cummulative distribution method
# We simply numerically integrate under the PDFt curve
panels = self.dof * 100
width = (t_right - t_left) / panels
cdf = 0
t = t_left
prob = self.PDFt(t)
# print(panels, width, prob)
for i in range(panels):
t += i * width
prob = self.PDFt(t)
cdf += prob * width

return cdf

In [ ]: import matplotlib.pyplot as plt

def PDF(x, mean=0, std_dev=1):

# define e and pi explicitly
e = 2.718281828
pi = 3.1415927
# calculate in two steps
p = 1.0 / (std_dev * ((2 * pi) ** 0.5))
p *= e ** (-0.5 * ((x - mean)/std_dev)**2)

return p

X = [(x - 1000)/250 for x in list(range(2001))]

P = [PDF(x) for x in X]
plt.plot(X, P)
for dof in [1, 2, 3, 5, 10, 30]:
t_dist = T_Distribution(dof=dof)
TP = [t_dist.PDFt(x) for x in X]
plt.plot(X, TP)
 plt.title(label="Test Distribution")
plt.xlabel(xlabel="value")
plt.ylabel(ylabel="propability of value occurance")
plt.show()

Basic Determination Of Significance Value

A Khan Academy Problem

In [ ]: X = [80]*5 + [82.5]*24 + [85]*72 + [87.5]*181 + [90]*281 + \

[92.5]*272 + [95]*136 + [97.5]*27 + [100]*2

mu = mean(X)
std = standard_deviation(X, mu=mu)
print(mu, std)

90.54 3.362796455332963

In [ ]: the_85_and_less = [x for x in X if x <= 85]

percentage_LE_85 = len(the_85_and_less)/len(X)
print(percentage_LE_85)

0.101

Basic Centerpoint Integration

w
We start t at 2
to use centerpoints for each panel.

There are other methods of numerical integration.

Centerpoint is pretty good at balancing areas above

and below the function being integrated.

In [ ]: import matplotlib.pyplot as plt
import math

w = 1/1000
f_of_t = math.sin

T = [w/2]
S = [0]
C = [-1]
for t in range(10000):
T.append(T[-1] + w) # Our time step
S.append(f_of_t(t*w)) # Our Function
C.append(f_of_t(t*w)*w + C[-1]) # Integrating

plt.plot(T, S)
plt.plot(T, C)
plt.show()

Null And Alternate Hypotheses

Distributions With Dynamic Significance
Level
For the LaTeX in MatPlotLib Inside Colab, See:
https://stackoverflow.com/a/62075348/996205

In [ ]: import matplotlib
from matplotlib import rc
import matplotlib.pyplot as plt
%matplotlib inline

rc('text', usetex=True)
matplotlib.rcParams['text.latex.preamble'] = [r'\usepackage{amsmath}']
!apt install texlive-fonts-recommended texlive-fonts-extra cm-super dvipng

In [ ]: import matplotlib.pyplot as plt

import math
import time

def PDF(x, mean=0, std_dev=1):

p = 1.0 / (std_dev * ((2 * math.pi) ** 0.5))
p *= math.e ** (-0.5 * ((x - mean)/std_dev)**2)
return p

pw = 1/1000
X = [(x + 0.5)*pw for x in range(22000)]
P1 = [PDF(x, 8, 2) for x in X]
P2 = [PDF(x, 14, 2) for x in X]

C1 = [] # C2 = []
sum1 = 0 # sum2 = 0
for i in range(len(X)):
sum1 += P1[i]*pw # sum2 += P2[i]*pw
C1.append(sum1) # C2.append(sum2)

SigLevels = [(v/2 + 90)/100 for v in range(17)]

SigLevels = [0.975]

for sl in SigLevels:
for i in range(len(X)):
if C1[i] > sl:
sig_i = i
break

plt.figure(figsize = (10,5))
plt.plot(X, P1)
plt.plot(X, P2)

plt.title(
label="Distributions For Null And Alternative Hypotheses", fontsize=
plt.xlabel(xlabel="Values", fontsize=14)
plt.ylabel(ylabel="Probability of Occurance", fontsize=14)

plt.fill_between(X[sig_i:], 0, P1[sig_i:], facecolor='green', alpha=0.5)

plt.fill_between(X[:sig_i], 0, P2[:sig_i], facecolor='orange', alpha=0.5

plt.text(8, 0.13, 'NULL', fontsize=18, ha='center')

plt.text(14, 0.13, 'ALT', fontsize=18, ha='center')

plt.text(8, 0.06, 'True\nNegative', fontsize=16, ha='center')

plt.text(14, 0.06, 'True\nPositive', fontsize=16, ha='center')

plt.text(1, 0.15, r'$\beta$ = percent'+'\nfalse negative',

fontsize=16, ha='left')
plt.text(18, 0.15, r'$\alpha$ = percent'+'\nfalse positive',
fontsize=16, ha='left')
this_text = f'significance\nlevel = {round(sl, 3)}'
plt.text(18, 0.10, this_text,
fontsize=16, ha='left')

plt.text(8, 0.1, r'$1 - \alpha$', fontsize=18, ha='center')

plt.text(14, 0.1, r'$1 - \beta$', fontsize=18, ha='center')
 plt.text(X[sig_i] + 0.25, 0.005, r'$\alpha$', fontsize=18)
plt.text(X[sig_i] - 0.75, 0.005, r'$\beta$', fontsize=18)

plt.savefig(f'hypo_{round(sl, 3)}.png')
plt.show()
time.sleep(1)
plt.figure().clear()

<Figure size 432x288 with 0 Axes>

Null And Alternate Hypotheses

Distributions With Dynamic Alternative
Mean
In [ ]: import matplotlib.pyplot as plt
import math
import time

def PDF(x, mean_=0, std_dev_=1):

p = 1.0 / (std_dev_ * ((2 * math.pi) ** 0.5))
p *= math.e ** (-0.5 * ((x - mean_)/std_dev_)**2)
return p

pw = 1/1000
X = [(x + 0.5)*pw for x in range(22000)]
P1 = [PDF(x, 8, 2) for x in X]

C1 = [] # C2 = []
sum1 = 0
for i in range(len(X)):
sum1 += P1[i]*pw
C1.append(sum1)
sig_level = 0.975
for i in range(len(X)):
if C1[i] > sig_level:
sig_i = i
break

for i in range(19):
mu_alt = 12.4 + i * 0.2
mu_alt = round(mu_alt, 1)

P2 = [PDF(x, mu_alt, 2) for x in X]

plt.figure(figsize = (10,5))
plt.plot(X, P1)
plt.plot(X, P2)

plt.title(
label="Distributions For Null And Alternative Hypotheses", fontsize=
plt.xlabel(xlabel="Values", fontsize=14)
plt.ylabel(ylabel="Probability of Occurance", fontsize=14)

plt.fill_between(X[sig_i:], 0, P1[sig_i:], facecolor='green', alpha=0.5)

plt.fill_between(X[:sig_i], 0, P2[:sig_i], facecolor='orange', alpha=0.5

plt.text(8, 0.13, 'NULL', fontsize=16, ha='center')

plt.text(mu_alt, 0.13, 'ALT', fontsize=16, ha='center')

plt.text(8, 0.06, 'True\nNegative', fontsize=14, ha='center')

plt.text(mu_alt, 0.06, 'True\nPositive', fontsize=14, ha='center')

plt.text(1, 0.15, r'$\beta$ = percent'+'\nfalse negative',

fontsize=14, ha='left')
plt.text(20, 0.15, r'$\alpha$ = percent'+'\nfalse positive',
fontsize=14, ha='left')
this_text = f'significance\nlevel = {round(sig_level, 3)}'
plt.text(20, 0.10, this_text,
fontsize=14, ha='left')
plt.text(20, 0.06, r'$\mu_{alt}$' + f' = {mu_alt}',
fontsize=16, ha='left')

plt.text(8, 0.1, r'$1 - \alpha$', fontsize=16, ha='center')

plt.text(mu_alt, 0.1, r'$1 - \beta$', fontsize=16, ha='center')

plt.text(X[sig_i] + 0.25, 0.005, r'$\alpha$', fontsize=18)

plt.text(X[sig_i] - 0.75, 0.005, r'$\beta$', fontsize=18)

plt.xlim([0, 26])
plt.savefig(f'hypos_{round(mu_alt, 1)}.png')
plt.show()
time.sleep(1)
plt.figure().clear()

ROC Curve From Dynamic Significance

Level
 ROC = True Positive Rate vs. False Positive Rate

In [ ]: import matplotlib.pyplot as plt

import math
import time

def PDF(x, mean=0, std_dev=1):

p = 1.0 / (std_dev * ((2 * math.pi) ** 0.5))
p *= math.e ** (-0.5 * ((x - mean)/std_dev)**2)
return p

pw = 1/1000
X = [(x + 0.5)*pw for x in range(22000)]
P1 = [PDF(x, 8, 2) for x in X]
P2 = [PDF(x, 11, 2) for x in X]

C1 = []
C2 = []
sum1 = 0
sum2 = 0
for i in range(len(X)):
sum1 += P1[i]*pw
sum2 += P2[i]*pw
C1.append(sum1)
C2.append(sum2)

ROC_X = []
ROC_Y = []
pts = 41
Sig_Levels = [(v * 100/(pts-1))/100 for v in range(0, pts)]

for sig_lev in Sig_Levels:

if sig_lev == 1:
sig_lev = 0.999
for i in range(len(X)):
if C1[i] > sig_lev:
sig_i = i
break

if sig_lev == 0.999:
sig_lev = 1

TP_Rate = 1 - C2[sig_i]
FP_Rate = 1 - C1[sig_i]

ROC_X.append(FP_Rate)
ROC_Y.append(TP_Rate)

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 5))

ax1.plot(X, P1)
ax1.plot(X, P2)

msg = "Receiver Operator Curve (ROC) From Sweeping Significance Level"

fig.suptitle(msg, fontsize=18)
 msg = "Significance Level On Null & Alt Distributions"
ax1.set_title(label=msg)
ax1.set_xlabel(xlabel="Values", fontsize=14)
ax1.set_ylabel(ylabel="Probability of Occurance", fontsize=14)

ax1.fill_between(X[sig_i:], 0, P1[sig_i:], facecolor='green', alpha=0.5)

ax1.fill_between(X[:sig_i], 0, P2[:sig_i], facecolor='orange', alpha=0.5

ax1.text(1, 0.15, r'$\beta$ = percent'+'\nfalse neg',

fontsize=16, ha='left')
ax1.text(14, 0.15, r'$\alpha$ = percent'+'\nfalse pos',
fontsize=16, ha='left')
this_text = f'significance\nlevel = {round(sig_lev, 3)}'
ax1.text(14, 0.10, this_text,
fontsize=16, ha='left')

ax1.text(X[sig_i] + 0.25, 0.005, r'$\alpha$', fontsize=18)

ax1.text(X[sig_i] - 0.75, 0.005, r'$\beta$', fontsize=18)

ax1.set_xlim([0, 21])

ax2.plot([0, 1], [0, 1])

ax2.plot(ROC_X, ROC_Y)
ax2.set_title(label="Receiver Operator Curve (ROC)")
ax2.set_xlabel(xlabel="False Positive Rate", fontsize=14)
ax2.set_ylabel(ylabel="True Positive Rate", fontsize=14)
ax2.set_xlim(0, 1)
ax2.set_ylim(0, 1)
plt.show()

fig.savefig(f'hypo_{round(sig_lev, 3)}.png')
time.sleep(0.2)

ROC Curve Changes Due To Separation Of

NULL And ALT Distributions' Means
ROC = True Positive Rate vs. False Positive Rate

In [ ]: import matplotlib.pyplot as plt

import math
import time

def PDF(x, mean=0, std_dev=1):

p = 1.0 / (std_dev * ((2 * math.pi) ** 0.5))
p *= math.e ** (-0.5 * ((x - mean)/std_dev)**2)
return p

pw = 1/1000
X = [(x + 0.5)*pw for x in range(22000)]
P1 = [PDF(x, 8, 2) for x in X]

C1 = []
sum1 = 0
for i in range(len(X)):
sum1 += P1[i]*pw
C1.append(sum1)

pts = 101
Sig_Levels = [(v * 100/(pts-1))/100 for v in range(0, pts)]

for i in range(71):
mu_alt = 8 + 0.1 * i
mu_alt = round(mu_alt, 1)

P2 = [PDF(x, mu_alt, 2) for x in X]

C2 = []
sum2 = 0
for i in range(len(X)):
sum2 += P2[i]*pw
C2.append(sum2)

ROC_X = []
ROC_Y = []
for sig_lev in Sig_Levels:
if sig_lev == 1:
sig_lev = 0.999
for i in range(len(X)):
if C1[i] > sig_lev:
sig_i = i
break

if sig_lev == 0.999:
sig_lev = 1

TP_Rate = 1 - C2[sig_i]
FP_Rate = 1 - C1[sig_i]

ROC_X.append(FP_Rate)
ROC_Y.append(TP_Rate)

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 5))

ax1.plot(X, P1)
ax1.plot(X, P2)

msg = "Receiver Operator Curve (ROC) From Sweeping Significance Level"

fig.suptitle(msg, fontsize=18)

msg = "Significance Level On Null & Alt Distributions"

ax1.set_title(label=msg)
ax1.set_xlabel(xlabel="Values", fontsize=14)
ax1.set_ylabel(ylabel="Probability of Occurance", fontsize=14)

ax1.text(8, 0.09, 'NULL\nHypothesis',

fontsize=14, ha='center')
ax1.text(mu_alt, 0.04, 'ALT\nHypothesis',
fontsize=14, ha='center')

ax1.set_xlim([0, 21])
 ax2.plot([0, 1], [0, 1])
ax2.plot(ROC_X, ROC_Y)
ax2.set_title(label="Receiver Operator Curve (ROC)")
ax2.set_xlabel(xlabel="False Positive Rate", fontsize=14)
ax2.set_ylabel(ylabel="True Positive Rate", fontsize=14)
ax2.set_xlim(0, 1)
ax2.set_ylim(0, 1)
plt.show()

fig.savefig(f'hypo_{round(mu_alt, 3)}.png')
# time.sleep(0.05)

ROC Curve Changes Due To Changes In

NULL And ALT Distributions' Standard
Deviations
ROC = True Positive Rate vs. False Positive Rate

In [ ]: import matplotlib.pyplot as plt

import math
import time

def PDF(x, mean=0, std_dev=1):

p = 1.0 / (std_dev * ((2 * math.pi) ** 0.5))
p *= math.e ** (-0.5 * ((x - mean)/std_dev)**2)
return p

pw = 1/1000
X = [(x + 0.5)*pw for x in range(22000)]

pts = 101
Sig_Levels = [(v * 100/(pts-1))/100 for v in range(0, pts)]

for i in range(26):
std_alt = 3.5 - 0.1 * i
std_alt = round(std_alt, 1)

P1 = [PDF(x, 8, std_alt) for x in X]

C1 = []
sum1 = 0
for i in range(len(X)):
sum1 += P1[i]*pw
C1.append(sum1)

P2 = [PDF(x, 12, std_alt) for x in X]

C2 = []
sum2 = 0
for i in range(len(X)):
sum2 += P2[i]*pw
C2.append(sum2)
ROC_X = []
ROC_Y = []
for sig_lev in Sig_Levels:
if sig_lev == 1:
sig_lev = 0.999
for i in range(len(X)):
if C1[i] > sig_lev:
sig_i = i
break

if sig_lev == 0.999:
sig_lev = 1

TP_Rate = 1 - C2[sig_i]
FP_Rate = 1 - C1[sig_i]

ROC_X.append(FP_Rate)
ROC_Y.append(TP_Rate)

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 5))

ax1.plot(X, P1)
ax1.plot(X, P2)

msg = "Receiver Operator Curve (ROC) From Sweeping Significance Level"

fig.suptitle(msg, fontsize=18)

msg = "Significance Level On Null & Alt Distributions"

ax1.set_title(label=msg)
ax1.set_xlabel(xlabel="Values", fontsize=14)
ax1.set_ylabel(ylabel="Probability of Occurance", fontsize=14)

ax1.text(8, 0.09, 'NULL\nHypothesis',

fontsize=14, ha='center')
ax1.text(12, 0.04, 'ALT\nHypothesis',
fontsize=14, ha='center')

ax1.set_xlim([0, 21])

ax2.plot([0, 1], [0, 1])

ax2.plot(ROC_X, ROC_Y)
ax2.set_title(label="Receiver Operator Curve (ROC)")
ax2.set_xlabel(xlabel="False Positive Rate", fontsize=14)
ax2.set_ylabel(ylabel="True Positive Rate", fontsize=14)
ax2.set_xlim(0, 1)
ax2.set_ylim(0, 1)
plt.show()

fig.savefig(f'hypo_{round(sig_lev, 3)}.png')
# time.sleep(0.05)

F1 Score
2
F1 =
1 1
+
Recall P recision
 F1 = Harmonic Mean Of Recall and Precision
________________________________________________________

Accuracy, Recall, Precision

TP + TN W hat W e Got Right

Accuracy = =
TP + FP + TN + FN All Cases

TP W hat P ositives W e Got Right

Recall = =
TP + FN All Actual P ositives

TP W hat P ositives W e Got Right

P recision = =
TP + FP All P ositive P redictions

LASSO In Logistic Regression to Compare

with Statistics Version
In [ ]: from sklearn.linear_model import LogisticRegression as LR
from sklearn.datasets import load_iris

X, y = X, y = load_iris(return_X_y=True)
lr_mod = LR(penalty='l1', solver='liblinear')
lr_mod.fit(X, y)
print(lr_mod.coef_)

[[ 0. 2.52235623 -2.83220134 0. ]
[ 0.32846823 -1.79370624 0.66582088 -1.57267348]
[-2.62263278 -2.50833176 3.26131365 4.61826807]]

Regressor Coefficient From Statistics Vs.

Machine Learning Methods
In [ ]: import matplotlib.pyplot as plt
import numpy as np

# Synthesize some data (i.e. create fake data)

X = np.random.uniform(0, 1, 1000)
Y = 2.0 * X
 Y_noise = np.max(Y) * 0.073
Y += np.random.normal(0, 0.073, 1000)

# Statistics Way To Create Model

X_std = np.std(X)
Y_std = np.std(Y)
r = np.corrcoef(X, Y)
Cs = Y_std / X_std * r[0, 1]
print(Cs)

# Machine Learning
mod_LR = LinearRegression(fit_intercept=False, copy_X=True)
mod_LR.fit(X.reshape(-1, 1), Y.reshape(-1, 1))
Cml = mod_LR.coef_[0, 0]
print(Cml)

# Visualize
plt.figure(figsize=(10, 5))
plt.plot(X, Cs*X+1) # + 1 separates the two exact plots
plt.plot(X, Cml*X)
plt.title('Models Determined By Stats And ML')
plt.legend(('Stats Way', 'ML Way'))
plt.show()

1.9918506427390517
1.9957689952415045

Regressor Coefficient AND "Intercept"

FROM Statistics
In [ ]: import matplotlib.pyplot as plt
import numpy as np

# Synthesize some data (i.e. create fake data)

X = np.random.uniform(0, 1, 1000)
 Y = 2.0 * X + 1
Y_noise = np.max(Y) * 0.073
Y += np.random.normal(0, 0.073, 1000)

# Statistics Way To Fit Model Coefficient

X_std = np.std(X)
Y_std = np.std(Y)
r = np.corrcoef(X, Y)
Cs = Y_std / X_std * r[0, 1]
print(Cs)

# Statistics Way To Calculate Intercept

X_mean = np.mean(X)
Y_mean = np.mean(Y)
b = Y_mean - Cs*X_mean
print(b)

1.9992493711567196
1.001206029449972

In [ ]: # Visualize
plt.figure(figsize=(10, 5))
plt.scatter(X, Y, color='magenta')
plt.plot(X, Cs*X+b, color='black')
plt.ylim((0, 3.5))
plt.title('Models Determined By Stats And ML')
plt.show()

Central Limit Theorem Principles

Sample Means Distribtion For Increasing Sample

Sizes
In [ ]: import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm

die_values = [1, 2, 3, 4, 5, 6]
sample_sizes = [2, 4, 8, 16, 32]

for experiment in range(1):

for sample_size in sample_sizes:
sample_means = []
for num_samples in range(1000):
die_cast = np.random.choice(
die_values, size=sample_size)
sample_mean = np.mean(die_cast)
sample_means.append(sample_mean)

experiment_mean = np.mean(sample_means)
experiment_std = np.std(sample_means)
x_min = min(sample_means)
x_max = max(sample_means)
x = np.arange(x_min, x_max, 0.01)
y = norm.pdf(x, experiment_mean, experiment_std)
plt.plot(x, y)

legend_texts = [f'Sample Size = {v}' for v in sample_sizes]

plt.legend(legend_texts)
plt.title("Distribution Changes By Sample Size")
plt.show()

Approaching The Central Limit Theorem

In [ ]: import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm
import os

cwd = os.getcwd()
if not os.path.isdir(f"{cwd}/images"):
 os.mkdir(f"{cwd}/images")

no_images = True
image_num = 0
if no_images:
die_values = [1, 2, 3, 4, 5, 6]
sample_sizes = [2, 4, 8, 16, 32]
num_add_samples_list = [2] + [1]*8 + [2]*5 + [10]*8 + [100]*9
sample_means_D = {k: [] for k in sample_sizes}
total_samples = 0

for num_samples in num_add_samples_list:

total_samples += num_samples
plt.figure(figsize=(12, 6))
for sample_size in sample_sizes:
for sample_num in range(num_samples):
die_cast = np.random.choice(
die_values, size=sample_size)
sample_mean = np.mean(die_cast)
sample_means_D[sample_size].append(sample_mean)

experiment_mean = np.mean(sample_means_D[sample_size])
experiment_std = np.std(sample_means_D[sample_size])
x_min = min(sample_means_D[sample_size])
x_max = max(sample_means_D[sample_size])
x = np.arange(x_min, x_max, 0.001)
y = norm.pdf(x, experiment_mean, experiment_std)
plt.plot(x, y)

legend_texts = [f'Sample Size = {v}' for v in sample_sizes]

plt.legend(legend_texts)
title = f"Distribution Of Means For {total_samples} "
title += "Samples For Various Sample Sizes"
plt.title(title)
plt.xlim([1, 6])
if total_samples == 2:
for i in range(5):
plt.savefig(f"{cwd}/images/{image_num:02d}.png")
image_num += 1
elif total_samples == 1000:
for i in range(5):
plt.savefig(f"{cwd}/images/{image_num:02d}.png")
image_num += 1
else:
plt.savefig(f"{cwd}/images/{image_num:02d}.png")
image_num += 1

plt.close()

# Run below on command line to create movie - needs ffmpeg

# ffmpeg -framerate 4 -pattern_type glob -i "*.png" output.avi
 Sample From Huge Population To See When
Central Limit Theorem Is Reached
In [ ]: import numpy as np
import matplotlib.pyplot as plt

die_values = [1, 2, 3, 4, 5, 6]
die_roles = [np.random.choice(die_values, size=1)[0] for _ in range(int(1e6)

# plt.hist(die_roles, bins=6, width=0.73)

# plt.show();

mean = round(np.mean(die_roles), 1)
print(f'Population mean is {mean}')

for sample_size in [2, 4, 8, 16, 32, 64]:

sample_means = []
samples = 0
while True:
samples += 1
roles = np.random.choice(die_roles, size=sample_size)
sample_mean = np.mean(roles)
sample_means.append(sample_mean)

running_mean = round(np.mean(sample_means), 2)
if running_mean == 3.50:
break

title = f'Mean sample means = {running_mean} for {samples} samples of {s

running_std = np.std(sample_means)
x = np.arange(1, 6, 0.001)
y = norm.pdf(x, running_mean, running_std)
plt.xlim([1, 6])
plt.plot(x, y)
plt.title(title)
plt.axvline(3.5)
plt.show();

The Ugly Approach To The Central Limit Theorem

In [ ]: import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm
import os

cwd = os.getcwd()
if not os.path.isdir(f"{cwd}/images"):
os.mkdir(f"{cwd}/images")

no_images = True
image_num = 0
if no_images:
die_values = [1, 2, 3, 4, 5, 6]
num_add_samples_list = [2] + [1]*8 + [2]*5 + [10]*8 + [100]*9
num_add_samples_list += [9000] + [90000] + [900000] + [1000000]
sample_means = []
total_samples = 0

for num_samples in num_add_samples_list:

total_samples += num_samples
for sample_num in range(num_samples):
die_casts = np.random.choice(
die_values, size=32)
sample_mean = np.mean(die_casts)
sample_means.append(sample_mean)

fig, ax1 = plt.subplots()

fig.set_figheight(6)
fig.set_figwidth(12)

color = 'tab:blue'
ax1.set_xlabel('Sample Means')
ax1.set_ylabel('Occurence Rate', color=color)
bins = len(set(sample_means))
ax1.hist(sample_means, bins=bins, density=True, stacked=True)
ax1.set_xlim([2, 5])
ax1.set_ylim([0, 4])

ax2 = ax1.twinx()

color = 'tab:red'
ax2.set_ylabel('Probability', color=color)
ax2.tick_params(axis='y', labelcolor=color)
running_mean = round(np.mean(sample_means), 2)
title = f'Mean of sample means = {running_mean} '
title += f'for {total_samples} samples of size 32'
running_std = np.std(sample_means)
x = np.arange(2, 5, 0.001)
y = norm.pdf(x, running_mean, running_std)
ax2.plot(x, y, color=color)
ax2.set_ylim([0, 2])

plt.title(title)
plt.axvline(3.5)

fig.tight_layout()

# if total_samples == 2:
# for i in range(5):
# plt.savefig(f"{cwd}/images/{image_num:02d}.png")
# image_num += 1
# elif total_samples > 1000000:
# for i in range(5):
# plt.savefig(f"{cwd}/images/{image_num:02d}.png")
# image_num += 1
# else:
# plt.savefig(f"{cwd}/images/{image_num:02d}.png")
# image_num += 1
 plt.show();

# plt.close()

# Run below on command line to create movie - needs ffmpeg

# ffmpeg -framerate 4 -pattern_type glob -i "*.png" output.avi
 Two Y Axes
In [ ]: import numpy as np
import matplotlib.pyplot as plt

# Create some mock data

t = np.arange(0.01, 10.0, 0.01)
data1 = np.exp(t)
data2 = np.sin(2 * np.pi * t)

fig, ax1 = plt.subplots()

color = 'tab:red'
ax1.set_xlabel('time (s)')
ax1.set_ylabel('exp', color=color)
ax1.plot(t, data1, color=color)
ax1.tick_params(axis='y', labelcolor=color)

ax2 = ax1.twinx() # instantiate a second axes that shares the same x-axis

color = 'tab:blue'
ax2.set_ylabel('sin', color=color) # we already handled the x-label with ax
ax2.plot(t, data2, color=color)
ax2.tick_params(axis='y', labelcolor=color)

fig.tight_layout() # otherwise the right y-label is slightly clipped

plt.show()

Binomial Variables

Rules Binomial Variables

1. Comprised Of Independent Trials
2. Each Trial Can Be Regarded As A Success Or Failure
3. There Are A FIXED Number Of Trials
4. Probability Of Success On Each Trial Is Constant (i.e. Rule 1 Repeats)

NOTE: Use The 10% Rule For Approximate Independence Of Trials When Resampling Is
Not Possible

Binomial Combinametrics
n k n−k
p(k of n) = ( ) p (1 − p)
k
 n n!
( ) =
k k!(n − k)!

n n + k − 1 (n + k − 1)! n(n + 1)(n + 2) ⋯ (n + k − 1)

(( )) = ( ) = = ,
k k k! (n − 1)! k!

n! k n−k
p(k of n) = p (1 − p)
k!(n − k)!

In [ ]: # Independent Trials Success Probability = 0.73

import math
import matplotlib.pyplot as plt

fact = math.factorial

Ps = 0.73
p_D = {}
n = 7
for k in range(n+1):
nCk = fact(n) / (fact(k) * fact(n-k))
p = nCk * Ps**k * (1 - Ps)**(n - k)
p_D[k] = round(p, 2)

plt.bar(p_D.keys(), p_D.values())
plt.show()