熊本深红vs长崎航海预测

python
prob_home=25  # predicted probability for home (maybe 0.25)
prob_draw=28
prob_away=47
odds_home=3.0
odds_draw=3.45
odds_away=2.25
exp_home=prob_home/100*odds_home
exp_draw=prob_draw/100*odds_draw
exp_away=prob_away/100*odds_away
exp_home, exp_draw, exp_away

(0.75, 0.9660000000000002, 1.0574999999999999)

python
from PIL import Image
import matplotlib.pyplot as plt

img = Image.open('/mnt/data/image.png')
plt.imshow(img)
plt.axis('off')

(-0.5, 1010.5, 217.5, -0.5)

python
import numpy as np
from math import exp, factorial

def poisson_prob(k, lam):
    return exp(-lam) * lam**k / factorial(k)

def score_prob(lam_home, lam_away, max_goals=6):
    probs = {}
    for i in range(max_goals+1):
        for j in range(max_goals+1):
            probs[(i,j)] = poisson_prob(i, lam_home) * poisson_prob(j, lam_away)
    # allocate probability mass for >max_goals as others
    leftover = 1 - sum(probs.values())
    probs[(max_goals, max_goals)] += leftover  # put leftover in 6-6 unrealistic but for completeness
    return probs

lam_home = 1.05
lam_away = 1.55
probs = score_prob(lam_home, lam_away, 5)
sum(probs.values())

1.0

python
# Map odds table
odds_score = {
    (1,0):10.5,
    (2,0):17,
    (2,1):9.5,
    (3,0):30,
    (3,1):20,
    (3,2):26,
    (4,0):100,
    (4,1):75,
    (4,2):90,
    (5,0):350,
    (5,1):300,
    (5,2):350,
    'home_other':90,
    (0,0):14,
    (1,1):7,
    (2,2):12,
    (3,3):50,
    'draw_other':350,
    (0,1):9,
    (0,2):12,
    (1,2):8,
    (0,3):21,
    (1,3):16,
    (2,3):21,
    (0,4):50,
    (1,4):35,
    (2,4):60,
    (0,5):150,
    (1,5):100,
    (2,5):175,
    'away_other':60
}

# compute probability for each specified score (within 0-5 goals)
max_g = 5
evs = {}
for key, odd in odds_score.items():
    if isinstance(key, tuple):
        prob = probs.get(key, 0)
    else:
        # others aggregated categories; compute probability of the rest cases
        if key == 'home_other':
            # home win with score >5 for home goals perhaps or 6 rel? But easier skip
            continue
    evs[key] = prob*odd
# Sort by EV
evs_sorted = sorted(evs.items(), key=lambda x: x[1], reverse=True)[:15]
evs_sorted

[('draw_other', 3.112883606700184),
 ((0, 2), 1.0706536299596228),
 ((0, 0), 1.039830095000674),
 ((0, 1), 1.0361164160899574),
 ((0, 3), 0.9680493237551591),
 ((0, 4), 0.8931407451312477),
 ((1, 1), 0.8461617398067985),
 ((0, 5), 0.8306208929720605),
 ((1, 0), 0.8188661998130309),
 ((1, 3), 0.7744394590041273),
 ((1, 2), 0.749457540971736),
 ((2, 0), 0.6960362698410764),
 ((1, 4), 0.6564584476714671),
 ((2, 1), 0.6028902396123441),
 ((2, 4), 0.5908126029043205)]

python
# total goals distribution 0-6+
tg_probs = {k:0 for k in range(0,7)}
for (i,j), p in probs.items():
    tot = i+j
    if tot<=6:
        tg_probs[tot]+=p
    else:
        tg_probs[6]+=p  # >6 goals aggregated

odds_tg = {0:14, 1:5.20, 2:3.60, 3:3.70, 4:5.30, 5:9.50, 6:17, 'gt6':25}
tg_ev={}
for g, odd in odds_tg.items():
    if g=='gt6':
        prob = tg_probs[6]
    else:
        prob = tg_probs[g]
    tg_ev[g]=prob*odd
sorted(tg_ev.items(), key=lambda x: x[1], reverse=True)

[('gt6', 1.2259287996740635),
 (0, 1.039830095000674),
 (1, 1.004178777457794),
 (2, 0.9037608997120146),
 (6, 0.8336315837783632),
 (3, 0.8050166532619982),
 (4, 0.749535775807455),
 (5, 0.698623911790345)]

python
lam_home_half = lam_home / 2
lam_away_half = lam_away / 2

# enumerating up to 5 each
def joint_probs(lam_h, lam_a, max_g=5):
    d={}
    for i in range(max_g+1):
        for j in range(max_g+1):
            d[(i,j)] = poisson_prob(i, lam_h) * poisson_prob(j, lam_a)
    leftover=1-sum(d.values())
    d[(max_g,max_g)]+=leftover
    return d
halftime_probs = joint_probs(lam_home_half, lam_away_half)
fulltime_probs = probs

# compute HT/FT categories: draw-draw, draw-away, etc
htft_odds = {
    ('home','home'):5.00,
    ('home','draw'):15,
    ('home','away'):22,
    ('draw','home'):7.00,
    ('draw','draw'):5.85,
    ('draw','away'):5.40,
    ('away','home'):25,
    ('away','draw'):15,
    ('away','away'):3.50
}

htft_prob={}
for (i_h,i_a), pht in halftime_probs.items():
    if i_h>i_a:
        ht_res='home'
    elif i_h<i_a:
        ht_res='away'
    else:
        ht_res='draw'
    for (f_h,f_a), pft in fulltime_probs.items():
        ft_res='home' if f_h>f_a else ('away' if f_h<f_a else 'draw')
        htft_prob[(ht_res, ft_res)] = htft_prob.get((ht_res, ft_res),0) + pht*pft

# compute EV
htft_ev={}
for k, odd in htft_odds.items():
    htft_ev[k]=htft_prob.get(k,0)*odd
sorted(htft_ev.items(), key=lambda x: x[1], reverse=True)

[(('away', 'home'), 2.428918325634561),
 (('home', 'away'), 2.3985119714327428),
 (('away', 'draw'), 1.479647514594696),
 (('draw', 'away'), 1.0325579545587926),
 (('home', 'draw'), 0.8797475337953233),
 (('draw', 'home'), 0.7092056702752813),
 (('away', 'away'), 0.6417819003972105),
 (('draw', 'draw'), 0.6017611310335055),
 (('home', 'home'), 0.2888302633823692)]

python
htft_prob[('draw','away')]

0.19121443602940602

python
htft_prob[('away','away')]

0.18336625725634587

python
probs[(0,2)]

0.08922113582996857

python
probs[(0,1)]

0.1151240462322175

python
stake_draw_away=5
prob_draw_away=htft_prob[('draw','away')]
odds_draw_away=5.4

ev_draw_away = stake_draw_away*(prob_draw_away*odds_draw_away -1)
ev_draw_away

0.16278977279396312

python
bets=[('draw/away',odds_draw_away,prob_draw_away),
      ('0-2',12,probs[(0,2)]),
      ('0-1',9,probs[(0,1)])]

for name,odds,prob in bets:
    b=odds-1
    q=1-prob
    f=(b*prob - q)/b
    print(name,f)

python
from PIL import Image
import pytesseract

img = Image.open('/mnt/data/image.png')
# Let's crop region? But entire image small.
import matplotlib.pyplot as plt
plt.imshow(img); plt.axis('off')

(-0.5, 1010.5, 217.5, -0.5)

python
text = pytesseract.image_to_string(img)
print(text)

python
# crop bottom 
w,h = img.size
crop = img.crop((0,int(h*0.45),w,int(h*0.75)))
plt.imshow(crop); plt.axis('off')

(-0.5, 1010.5, 64.5, -0.5)

python
crop2 = img.crop((260,90,1050,150))
plt.imshow(crop2); plt.axis('off')

(-0.5, 789.5, 59.5, -0.5)

python
crop3 = img.crop((250,130,1050,210))
plt.imshow(crop3); plt.axis('off')

(-0.5, 799.5, 79.5, -0.5)

python
import math, itertools
lam = 2.6
p3 = math.exp(-lam)*lam**3/math.factorial(3)
p3

0.21757206844918872

python
import numpy as np
lamH = 1.05
lamA = 1.55
max_goals = 6
# compute distribution
P = {}
for h in range(max_goals+1):
    p_h = math.exp(-lamH)*lamH**h/math.factorial(h)
    for a in range(max_goals+1):
        p_a = math.exp(-lamA)*lamA**a/math.factorial(a)
        P[(h,a)] = p_h * p_a

# probability of totals
prob_totals={}
for (h,a),p in P.items():
    tot=h+a
    prob_totals[tot]=prob_totals.get(tot,0)+p
# compute
prob_totals_sorted = sorted(prob_totals.items())
prob_totals_sorted[:10]

[(0, 0.07427357821433386),
 (1, 0.19311130335726806),
 (2, 0.2510446943644485),
 (3, 0.2175720684491887),
 (4, 0.14142184449197265),
 (5, 0.0735393591358258),
 (6, 0.031867055625524505),
 (7, 0.011498842141149207),
 (8, 0.0034179800989606163),
 (9, 0.0008322331380651117)]

python
lam=2.6
prob_0=math.exp(-lam)
prob_0

0.07427357821433388

python
# compute cumulative up to 4
cum=0
for k in range(0,5):
    cum += math.exp(-lam)* lam**k / math.factorial(k)
cum

0.8774234888772119

python
prob_5plus=1-cum
prob_5plus

0.12257651112278811

python
lam=2.68
prob0=math.exp(-lam)
cum=0
for k in range(0,5):
    cum += math.exp(-lam)* lam**k / math.factorial(k)
prob5plus=1-cum
prob0, prob5plus

(0.06856315415427791, 0.13413021657530066)

python
lam=2.65
prob0=math.exp(-lam)
cum=sum(math.exp(-lam)*lam**k/math.factorial(k) for k in range(0,5))
prob5plus=1-cum
prob0, prob5plus

(0.0706512130604296, 0.12974188952722832)

python
round(prob0*100,1), round(prob5plus*100,1)

(7.1, 13.0)

python
# open the image file
img = Image.open('/mnt/data/image.png')
plt.imshow(img); plt.axis('off')

(-0.5, 1010.5, 217.5, -0.5)

python
# crop total goals row
# row physically around center vert maybe 110? Let's inspect pixel dims
print(img.size)

python
# attempt cropping
total_row = img.crop((0, 100, 1011, 160))
plt.imshow(total_row); plt.axis('off')

(-0.5, 1010.5, 59.5, -0.5)

python
# enlarge
plt.figure(figsize=(10,2))
plt.imshow(total_row.resize((total_row.width*3, total_row.height*3), Image.NEAREST))
plt.axis('off')

(-0.5, 3032.5, 179.5, -0.5)

python
lam_home=1.0
lam_away=1.6
max_goals=6
# compute probabilities of scorelines up to 6, else rest
prob_matrix=[]
for i in range(max_goals+1):
    row=[]
    for j in range(max_goals+1):
        prob=math.exp(-lam_home)*lam_home**i/math.factorial(i)*math.exp(-lam_away)*lam_away**j/math.factorial(j)
        row.append(prob)
    prob_matrix.append(row)
prob_matrix=np.array(prob_matrix)
home_win=prob_matrix[np.triu_indices_from(prob_matrix, k=1)].sum()
draw=np.trace(prob_matrix)
away_win=prob_matrix[np.tril_indices_from(prob_matrix, k=-1)].sum()
home_win,draw,away_win

(0.5121018491134687, 0.24999863247782664, 0.23648062718819937)

python
odds=[2.75,3.3,2.2]
inv=[1/o for o in odds]
total=sum(inv)
prob=[i/total for i in inv]
prob

[0.3243243243243243, 0.27027027027027023, 0.4054054054054054]

python
probs={}
for i in range(5):
    for j in range(5):
        p=math.exp(-lam_home)*lam_home**i/math.factorial(i)*math.exp(-lam_away)*lam_away**j/math.factorial(j)
        probs[(i,j)]=p
# sort top scores
top=sorted(probs.items(), key=lambda x: x[1], reverse=True)[:5]
top

[((0, 1), 0.11883772514293421),
 ((1, 1), 0.11883772514293421),
 ((0, 2), 0.09507018011434738),
 ((1, 2), 0.09507018011434738),
 ((0, 0), 0.07427357821433388)]

python
from PIL import Image
import matplotlib.pyplot as plt

img = Image.open('/mnt/data/e9fc2ee9ca24562b2e9a9ad478f0195.jpg')
plt.imshow(img)
plt.axis('off')

(-0.5, 1079.5, 1919.5, -0.5)

python
# Crop first ticket (top-left)
w, h = img.size
# crop roughly top-left quarter
crop1 = img.crop((0, 0, w/2, h/2))
plt.imshow(crop1)
plt.axis('off')

(-0.5, 539.5, 959.5, -0.5)

python
crop1b = crop1.crop((0, h*0.25, w*0.8, h*0.55))
plt.imshow(crop1b)
plt.axis('off')

(-0.5, 863.5, 575.5, -0.5)

python
crop1c = crop1.crop((0, h*0.55, w*0.8, h))
plt.imshow(crop1c)
plt.axis('off')

(-0.5, 863.5, 863.5, -0.5)

python
crop2 = img.crop((w/2, 0, w, h/2))
plt.imshow(crop2)
plt.axis('off')

(-0.5, 539.5, 959.5, -0.5)

python
crop2b = crop2.crop((0, h*0.25, w*0.8, h*0.55))
plt.imshow(crop2b)
plt.axis('off')

(-0.5, 863.5, 575.5, -0.5)

python
crop3 = img.crop((0, h/2, w/2, h))
plt.imshow(crop3)
plt.axis('off')

(-0.5, 539.5, 959.5, -0.5)

python
crop3b = crop3.crop((0, 0, w*0.8, h*0.3))
plt.imshow(crop3b)
plt.axis('off')

(-0.5, 863.5, 575.5, -0.5)

python
crop4 = img.crop((w/2, h/2, w, h))
plt.imshow(crop4)
plt.axis('off')

(-0.5, 539.5, 959.5, -0.5)

python
crop4b = crop4.crop((0, 0, w*0.8, h*0.3))
plt.imshow(crop4b)
plt.axis('off')

(-0.5, 863.5, 575.5, -0.5)

python
stake = {'ticket1':4, 'ticket2':2, 'ticket3':2, 'ticket4':2}
odds = {'ticket1':3.7*2, # cause 2x multiplier 3.7*2=7.4? Wait 1yuan × 3.7 =3.7 returns includes stake? Chinese sports lottery "固定奖金" refers to profit, not including stake? Actually '金额为每1元投注对应的奖金金额', which is payout (including stake) of 3.7? Likely gross payout per 1 yuan. So 2 yuan ticket each 'bet' returns 2*3.7=7.4; with 2倍 it's 14.8. So compute 14.8.
 'ticket2':2*7, 'ticket3':2*6.3, 'ticket4':2*9.5}
odds

{'ticket1': 7.4, 'ticket2': 14, 'ticket3': 12.6, 'ticket4': 19.0}

python
import math, itertools

λ_home=1.0
λ_away=1.6

def poisson(k, lam):
    return math.exp(-lam)*lam**k/math.factorial(k)

max_goals=6  # integrate
prob=0
EV= -10  # stake negative? We'll accumulate expected returns - stake part?

# compute expected gain: returns minus stake
EV=0
for home in range(0, max_goals+1):
    for away in range(0, max_goals+1):
        p=poisson(home, λ_home)*poisson(away, λ_away)
        # gross return:
        ret=0
        total=home+away
        if total==3:
            ret+=odds['ticket1']
        if total==1:
            ret+=odds['ticket3']
        if home==1 and away==1:
            ret+=odds['ticket2']
        if home==0 and away==1:
            ret+=odds['ticket4']
        # Contribution to expected value
        EV += p*(ret)
# subtract stake 10 (since always)
EV_net = EV - 10
EV_net

-0.425086034933603

python
profit_positive=0
profit_zero=0
loss=0
for home in range(0, max_goals+1):
    for away in range(0, max_goals+1):
        p=poisson(home, λ_home)*poisson(away, λ_away)
        total=home+away
        ret=0
        if total==3: ret+=odds['ticket1']
        if total==1: ret+=odds['ticket3']
        if (home==1 and away==1): ret+=odds['ticket2']
        if (home==0 and away==1): ret+=odds['ticket4']
        net = ret - 10
        if net>0: profit_positive += p
        elif net==0: profit_zero += p
        else: loss += p
profit_positive, profit_zero, loss

(0.529521096949391, 0, 0.46906001183010365)