In [1]:
Copied!
from google.colab import drive
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.decomposition import PCA
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler
import warnings
drive.mount('/content/gdrive')
df = pd.read_csv('/content/gdrive/My Drive/datasets/penguins.csv')
df = df.dropna()
df = df.drop([9, 14])
# Inspect the results
df.head()
from google.colab import drive
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.decomposition import PCA
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler
import warnings
drive.mount('/content/gdrive')
df = pd.read_csv('/content/gdrive/My Drive/datasets/penguins.csv')
df = df.dropna()
df = df.drop([9, 14])
# Inspect the results
df.head()
Mounted at /content/gdrive
Out[1]:
| culmen_length_mm | culmen_depth_mm | flipper_length_mm | body_mass_g | sex | |
|---|---|---|---|---|---|
| 0 | 39.1 | 18.7 | 181.0 | 3750.0 | MALE |
| 1 | 39.5 | 17.4 | 186.0 | 3800.0 | FEMALE |
| 2 | 40.3 | 18.0 | 195.0 | 3250.0 | FEMALE |
| 4 | 36.7 | 19.3 | 193.0 | 3450.0 | FEMALE |
| 5 | 39.3 | 20.6 | 190.0 | 3650.0 | MALE |
2. One-Hot Encode Categorical Variables¶
We think the penguin sex might be useful for our clustering. Let's one-hot encode it.
df = pd.get_dummies(df).drop("sex_.", axis=1)
Inspect the dataframe after you've done this.
In [2]:
Copied!
df = pd.get_dummies(df).drop("sex_.", axis=1)
df.head()
df = pd.get_dummies(df).drop("sex_.", axis=1)
df.head()
Out[2]:
| culmen_length_mm | culmen_depth_mm | flipper_length_mm | body_mass_g | sex_FEMALE | sex_MALE | |
|---|---|---|---|---|---|---|
| 0 | 39.1 | 18.7 | 181.0 | 3750.0 | False | True |
| 1 | 39.5 | 17.4 | 186.0 | 3800.0 | True | False |
| 2 | 40.3 | 18.0 | 195.0 | 3250.0 | True | False |
| 4 | 36.7 | 19.3 | 193.0 | 3450.0 | True | False |
| 5 | 39.3 | 20.6 | 190.0 | 3650.0 | False | True |
3. Scale the features¶
We're going to do some scaling of the features to transform them all into a range of [0, 1]
Note: Because this is an unsupervised algorithm, the whole dataframe is out X feature matrix!
scaler = StandardScaler()
X = scaler.fit_transform(df)
df = pd.DataFrame(data=X, columns=df.columns)
Inspect the dataframe after you've done this.
In [3]:
Copied!
scaler = StandardScaler()
X = scaler.fit_transform(df)
df = pd.DataFrame(data=X, columns=df.columns)
df
scaler = StandardScaler()
X = scaler.fit_transform(df)
df = pd.DataFrame(data=X, columns=df.columns)
df
Out[3]:
| culmen_length_mm | culmen_depth_mm | flipper_length_mm | body_mass_g | sex_FEMALE | sex_MALE | |
|---|---|---|---|---|---|---|
| 0 | -0.905520 | 0.793126 | -1.428125 | -0.569709 | -0.991031 | 0.997001 |
| 1 | -0.831938 | 0.128503 | -1.071522 | -0.507579 | 1.009050 | -1.003008 |
| 2 | -0.684775 | 0.435252 | -0.429637 | -1.191006 | 1.009050 | -1.003008 |
| 3 | -1.347011 | 1.099875 | -0.572278 | -0.942487 | 1.009050 | -1.003008 |
| 4 | -0.868729 | 1.764498 | -0.786240 | -0.693968 | -0.991031 | 0.997001 |
| ... | ... | ... | ... | ... | ... | ... |
| 328 | 0.584511 | -1.763116 | 0.925454 | 0.890339 | 1.009050 | -1.003008 |
| 329 | 0.510929 | -1.456367 | 0.996775 | 0.797145 | 1.009050 | -1.003008 |
| 330 | 1.173166 | -0.740619 | 1.496019 | 1.915480 | -0.991031 | 0.997001 |
| 331 | 0.216602 | -1.200743 | 0.782813 | 1.232053 | 1.009050 | -1.003008 |
| 332 | 1.081188 | -0.536120 | 0.854134 | 1.480572 | -0.991031 | 0.997001 |
333 rows × 6 columns
4. PCA¶
In addition to scaling as feature engineering, we're going to reduce the number of features using Principal Component Analysis (PCA).
pca = PCA(n_components=None)
dfx_pca = pca.fit(df)
dfx_pca.explained_variance_ratio_
n_components = sum(dfx_pca.explained_variance_ratio_ > 0.1)
pca = PCA(n_components=n_components)
df = pd.DataFrame(pca.fit_transform(df))
Inspect both the dataframe and n_components after you've done this.
In [4]:
Copied!
pca = PCA(n_components=None)
dfx_pca = pca.fit(df)
dfx_pca.explained_variance_ratio_
n_components = sum(dfx_pca.explained_variance_ratio_ > 0.1)
pca = PCA(n_components=n_components)
df = pd.DataFrame(pca.fit_transform(df))
pca = PCA(n_components=None)
dfx_pca = pca.fit(df)
dfx_pca.explained_variance_ratio_
n_components = sum(dfx_pca.explained_variance_ratio_ > 0.1)
pca = PCA(n_components=n_components)
df = pd.DataFrame(pca.fit_transform(df))
Determine the Number of Clusters¶
You have two options for determining how many cluster, you can either just guess or you can use the elbow method to plot the inertia (sum of squared distances to the nearest cluster center) for different values of k.
Here's the code for that.
inertia = []
for k in range(1, 10):
kmeans = KMeans(n_clusters=k, random_state=42).fit(df)
inertia.append(kmeans.inertia_)
plt.plot(range(1, 10), inertia, marker="o")
plt.xlabel("Number of clusters")
plt.ylabel("Inertia")
plt.title("Elbow Method")
plt.show()
With this plot, you should be able to see when a larger number of clusters starts to have diminishing returns.
Pick a number that feels right to you. Assign that number to the variable n_clusters before moving on.
n_clusters = ...
In [5]:
Copied!
inertia = []
for k in range(1, 10):
kmeans = KMeans(n_clusters=k, random_state=42).fit(df)
inertia.append(kmeans.inertia_)
plt.plot(range(1, 10), inertia, marker="o")
plt.xlabel("Number of clusters")
plt.ylabel("Inertia")
plt.title("Elbow Method")
plt.show()
inertia = []
for k in range(1, 10):
kmeans = KMeans(n_clusters=k, random_state=42).fit(df)
inertia.append(kmeans.inertia_)
plt.plot(range(1, 10), inertia, marker="o")
plt.xlabel("Number of clusters")
plt.ylabel("Inertia")
plt.title("Elbow Method")
plt.show()
/usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn( /usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn( /usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn( /usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn( /usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn( /usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn( /usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn( /usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn( /usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn(
6. K-means Clustering¶
Apply K-means clustering with the optimal number of clusters determined from the previous step.
kmeans = KMeans(n_clusters=n_clusters, random_state=42).fit(df)
Visualize the clusters on the first two principal components (it's easier to visualize things in only two dimensions).
plt.scatter(df.loc[:, 0], df.loc[:, 1], c=kmeans.labels_, cmap="viridis") plt.xlabel("First Principal Component") plt.ylabel("Second Principal Component") plt.title(f"K-means Clustering (K={n_clusters})") plt.show()
In [7]:
Copied!
n_clusters = 4
kmeans = KMeans(n_clusters=n_clusters, random_state=42).fit(df)
plt.scatter(df.loc[:,0], df.loc[:, 1], c=kmeans.labels_, cmap="viridis")
plt.xlabel("PCA Feature 0")
plt.ylabel("PCA Feature 1")
plt.title(f"K-means Clustering (K={n_clusters})")
plt.show()
n_clusters = 4
kmeans = KMeans(n_clusters=n_clusters, random_state=42).fit(df)
plt.scatter(df.loc[:,0], df.loc[:, 1], c=kmeans.labels_, cmap="viridis")
plt.xlabel("PCA Feature 0")
plt.ylabel("PCA Feature 1")
plt.title(f"K-means Clustering (K={n_clusters})")
plt.show()
/usr/local/lib/python3.10/dist-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning warnings.warn(
