Kusanthula Kosavuta kwa Linear Regression
Kubwerezabwereza kosavuta kwa mzere ndi njira yowerengera yomwe imagwiritsidwa ntchito pofufuza ubale pakati pa zosintha ziwiri zowerengera. Kusintha komwe tikuyesera kuneneratu kumatchedwa kusintha kodalira kapena kuyankha, pomwe kusintha komwe kumagwiritsidwa ntchito popanga kuneneratu kumatchedwa kusintha kodziyimira pawokha kapena koneneratu. Mu kubwerezabwereza kosavuta kwa mzere, timayesa kupeza mzere wowongoka bwino womwe umafotokoza ubale pakati pa zosintha ziwirizi.
Malingaliro Oyambira a Kuchepetsa Kosavuta kwa Linear
Kubwerezabwereza kosavuta kwa mzere kumachokera pa lingaliro lakuti pali ubale wolunjika pakati pa variable yodalira \(Y\) ndi variable yodziyimira payokha \(X\). Mtundu wamba wa chitsanzo chobwerezabwereza chosavuta ndi:
\[ Y = \beta_0 + \beta_1 X + \epsilon \]
Kumene:
– \( Y \) ndi chosinthika chodalira.
– \( X \) ndi chosinthika chodziyimira pawokha.
– \( \beta_0 \) ndi intercept, yomwe ndi mtengo wa \(Y\) pamene \(X = 0\).
– \( \beta_1 \) ndi slope kapena gradient, yomwe ndi kusintha kwapakati pa \(Y\) pa kusintha kulikonse kwa unit mu \(X\).
– \( \epsilon \) ndi cholakwika kapena mawu otsala omwe akuyimira kusiyana kwa \(Y\) komwe sikungathe kufotokozedwa ndi \(X\).
Cholinga cha kubwerezabwereza kosavuta kwa mzere ndikuyerekeza magawo \(\beta_0\) ndi \(\beta_1\) kuti chitsanzocho chigwiritsidwe ntchito kulosera mtengo wa \(Y\) wogwirizana ndi mtengo wa \(X\).
Njira Yocheperako ya Mabwalo
Njira imodzi yomwe imagwiritsidwa ntchito kwambiri poyika chitsanzo chosavuta cha mzere ndi njira ya Least Squares. Njirayi cholinga chake ndi kuchepetsa kuchuluka kwa masikweya a kusiyana kwa vertical pakati pa zomwe zawonedwa ndi zomwe zanenedweratu ndi chitsanzocho. Tiyerekeze kuti tili ndi zomwe zikuwonedwa ndi n zomwe zili ndi ma pairs \((x_i, y_i)\) pa \(i = 1, 2, …, n\). Ntchito yochepetsedwa ndi iyi:
\[ S(\beta_0, \beta_1) = \sum_{i=1}^{n} (y_i – (\beta_0 + \beta_1 x_i))^2 \]
Kuti tipeze \(\beta_0\) ndi \(\beta_1\) zomwe zimachepetsa ntchito iyi, timatenga magawo a \(S(\beta_0, \beta_1)\) poyerekeza ndi gawo lililonse ndikuyika ma derivatives awa ku zero. Kuwerengera kwa masamu kumatha kuchepetsedwa motere:
\[ \beta_1 = \frac{\sum_{i=1}^{n} (x_i – \bar{x})(y_i – \bar{y})}{\sum_{i=1}^{n} (x_i – \bar{x})^2} \]
\[ \beta_0 = \bar{y} – \beta_1 \bar{x} \]
Kumene:
– \(\bar{x}\) ndi avareji ya \(X\)
– \(\bar{y}\) ndi avareji ya \(Y\)
Pambuyo popeza magawo \(\beta_0\) ndi \(\beta_1\), chitsanzo chosavuta chowongolera mzere chingagwiritsidwe ntchito kulosera mtengo wa \(Y\) pa mtengo uliwonse wa \(X\).
Malingaliro mu Kubwereza Kosavuta kwa Linear
Kuti mupeze zotsatira zodalirika komanso zodalirika, kubwerezabwereza kosavuta kwa mzere kumatanthauza zinthu zingapo:
1. Mzere: Ubale pakati pa chosinthika chodalira ndi chosinthika chodziyimira pawokha uyenera kukhala wolunjika.
2. Kudziyimira pawokha: Kuonera zinthu kuyenera kukhala kodziyimira pawokha.
3. Homoscedasticity: Kusinthasintha kotsalira kuyenera kukhala kosalekeza pamitundu yonse ya ma values a variable yodziyimira payokha.
4. Kukhazikika kwa Zotsalira: Zotsalira (zolakwika) ziyenera kutsatira kugawa kwabwinobwino.
Ngati malingaliro awa sakwaniritsidwa, zotsatira za chitsanzo chosavuta chowongolera mzere sizidzakhala zodalirika ndipo sizingathe kuneneratu molondola.
Kuwunika kwa Chitsanzo cha Kubwerera M'mbuyo
Njira imodzi yowunikira momwe chitsanzo chosavuta chowongolera mzere chaneneratu ndikugwiritsa ntchito Coefficient of Determination (\(R^2\)). Coefficient of determination ikuwonetsa kuchuluka kwa kusinthasintha mu variable yodalira yomwe ingafotokozedwe ndi kusinthasintha mu zosintha zodziyimira pawokha.
\[ R^2 = \frac{\sum_{i=1}^{n} (\hat{y}_i – \bar{y})^2}{\sum_{i=1}^{n} (y_i – \bar{y})^2} \]
Kumene:
– \(\hat{y}_i\) ndi mtengo woloseredwa wa \(Y\).
– \(y_i\) ndi mtengo weniweni wa \(Y\).
– \(\bar{y}\) ndi avareji ya mitengo ya \(Y\).
Mtengo wa \(R^2\) umayambira pa 0 mpaka 1. Mtengo wa \(R^2\) womwe uli pafupi ndi 1 umasonyeza kuti chitsanzocho chingafotokoze kusiyana kwakukulu kwa variable yodalira.
Kukhazikitsa mu Chilankhulo cha Mapulogalamu
Kuti tigwiritse ntchito njira yosavuta yosinthira mzere, tingagwiritse ntchito mapulogalamu osiyanasiyana owerengera kapena zilankhulo zopanga mapulogalamu. Pansipa pali chitsanzo cha momwe tingagwiritsire ntchito Python pogwiritsa ntchito laibulale ya `scikit-learn`:
"`python
tumizani numpy ngati np
tumizani matplotlib.pyplot ngati plt
kuchokera ku sklearn.linear_model import LinearRegression
kuchokera ku sklearn.metrics import mean_squared_error, r2_score
Deta
X = np.array([[1], [2], [3], [4], [5]]).astype(np.float64)
y = np.array([1.5, 3.6, 3.5, 2.9, 5.5]).astype(np.float64)
lachitsanzo
chitsanzo = LinearRegression ()
model.fit (X, y)
Kuneneratu
y_pred = chitsanzo.kuneneratu (X)
Koefficient
beta_0 = model.intercept_
beta_1 = model.coef_[0]
sindikizani (f'Intercept: {beta_0}')
sindikizani (f'Slope: {beta_1}')
print(f'Pafupifupi cholakwika cha squared: {mean_squared_error(y, y_pred)}')
print(f'Koyefifiti yotsimikizira (R^2): {r2_score(y, y_pred)}')
Chiwembu cha deta ndi mzere wobwerera
plt.scatter(X, y, mtundu='buluu')
plt.plot(X, y_pred, mtundu='wofiira')
plt.xlabel('X')
plt.ylabel('Y')
onetsani ()
``
Mu chitsanzo pamwambapa, choyamba timalowetsa malaibulale ofunikira, kufotokoza deta \(X\) ndi \(Y\), kenako timagwiritsa ntchito chinthu cha `LinearRegression` kuchokera ku `scikit-learn` kuti tigwirizane ndi chitsanzo ndi deta. Chitsanzocho chikayikidwa, timapanga maulosi ndikuwerengera ma coefficients, komanso mean squared error ndi coefficient of determination. Pomaliza, timajambula deta ndi mzere wa regression.
Mapeto
Kubwerezabwereza kosavuta kwa mzere ndi chida champhamvu chowunikira ziwerengero chomwe chimagwiritsidwa ntchito pofotokoza ubale pakati pa zosintha ziwiri zowerengera. Ndi malingaliro oyambira okhudza kulumikizana, kudziyimira pawokha, homoscedasticity, ndi normality, titha kuneneratu mtengo wa chosinthika chodalira kutengera mtengo wa zosintha zodziyimira pawokha. Njira ya Least Squares imapereka njira yothandiza yolumikizira mzere wobwerera ndikupeza magawo oyenera. Kuwunika kwa chitsanzo kudzera mu coefficient of determination (R2) kumapereka chidziwitso cha momwe chitsanzo chathu chimagwirira ntchito bwino.
Ngakhale kuti njira yosavuta yosinthira mzere ili ndi zoletsa, monga kungotha kuthana ndi zosinthika ziwiri ndi malingaliro omwe ayenera kukwaniritsidwa, njira iyi ikadali maziko ofunikira mu ziwerengero ndi kusanthula deta, ndipo nthawi zambiri imagwiritsidwa ntchito ngati sitepe yoyamba kumvetsetsa ubale pakati pa zosinthika musanapite ku njira zovuta kwambiri.
