16

While I am predicting the one sample from my data, it gives reshape error but my model has equal number of rows. Here is my code:

import pandas as pd
from sklearn.linear_model import LinearRegression
import numpy as np
x = np.array([2.0 , 2.4, 1.5, 3.5, 3.5, 3.5, 3.5, 3.7, 3.7])
y = np.array([196, 221, 136, 255, 244, 230, 232, 255, 267])

lr = LinearRegression()
lr.fit(x,y)

print(lr.predict(2.4))

The error is

if it contains a single sample.".format(array))
ValueError: Expected 2D array, got scalar array instead:
array=2.4.
Reshape your data either using array.reshape(-1, 1) if your data has a 
single feature or array.reshape(1, -1) if it contains a single sample.
cottontail
  • 10,268
  • 18
  • 50
  • 51
  • if you want to know why fitting a model requires 2d model -> [here](https://stackoverflow.com/questions/61367841/why-does-the-fit-method-in-sklearns-linearregression-only-accept-2d-array-for-t) – Dozzy Norm Apr 14 '22 at 01:08

2 Answers2

25

You should reshape your X to be a 2D array not 1D array. Fitting a model requires requires a 2D array. i.e (n_samples, n_features)

x = np.array([2.0 , 2.4, 1.5, 3.5, 3.5, 3.5, 3.5, 3.7, 3.7])
y = np.array([196, 221, 136, 255, 244, 230, 232, 255, 267])

lr = LinearRegression()
lr.fit(x.reshape(-1, 1), y)

print(lr.predict([[2.4]]))
Abhi
  • 4,068
  • 1
  • 16
  • 29
  • 3
    Thanks man. But what's that reshape does and why not using it causes error. –  Nov 01 '19 at 18:17
  • 3
    @razzOn2bull When fitting a model your X needs to be 2D array. i.e (n_samples, n_features). When you use `.reshape(1, -1)` it adds one dimension to the data. You can see this question ('https://stackoverflow.com/questions/18691084/what-does-1-mean-in-numpy-reshape') for more info regarding it. – Abhi Nov 01 '19 at 18:20
0

The error is basically saying to convert the flat feature array into a column array. reshape(-1, 1) does the job; also [:, None] can be used.

The second dimension of the feature array X must match the second dimension of whatever is passed to predict() as well. Since X is coerced into a 2D array, the array passed to predict() should be 2D as well.

x = np.array([2.0 , 2.4, 1.5, 3.5, 3.5, 3.5, 3.5, 3.7, 3.7])
y = np.array([196, 221, 136, 255, 244, 230, 232, 255, 267])
X = x[:, None]         # X.ndim should be 2

lr = LinearRegression()
lr.fit(X, y)

prediction = lr.predict([[2.4]])

If your input is a pandas column, then use double brackets ([[]]) get a 2D feature array.

df = pd.DataFrame({'feature': x, 'target': y})
lr = LinearRegression()
lr.fit(df['feature'], df['target'])            # <---- error
lr.fit(df[['feature']], df['target'])          # <---- OK
#        ^^         ^^                           <---- double brackets
Why should X be 2D?

If we look at the source code of fit() (of any model in scikit-learn), one of the first things done is to validate the input via the validate_data() method, which calls check_array() to validate X. check_array() checks among other things, whether X is 2D. It is essential for X to be 2D because ultimately, LinearRegression().fit() calls scipy.linalg.lstsq to solve the least squares problem and lstsq requires X to be 2D to perform matrix multiplication.

For classifiers, the second dimension is needed to get the number of features, which is essential to get the model coefficients in the correct shape.

cottontail
  • 10,268
  • 18
  • 50
  • 51