如何将具有不同参数的多个sklearn算法应用于多个数据帧？

Question

I am looking for an efficient way to apply multiple sklearn clustering algorithm to multiple dataframes without too much repetition.

import pandas as pd
import numpy as np
from sklearn.datasets import make_moons,make_blobs
from sklearn.cluster import KMeans, DBSCAN
from matplotlib import pyplot

X1, y1 = make_moons(n_samples=100, noise=0.1)
X2, y2 = make_blobs(n_samples=100, centers=3, n_features=2)

And I want to apply both kmeans and dbscan on these datasets, but each datasets requires different parameters, how can I use a loop to apply multiple models to multiple data and eventually plot them out in a grid? Thanks.

Answer 1

You have create few dict for defining the hyperparams for each dataset|clustering_algo combination.

May be following approach could work for you! [Developed from sklearn clustering's documentation]

import pandas as pd
import numpy as np
from sklearn.datasets import make_moons,make_blobs
from sklearn.cluster import KMeans, DBSCAN
from matplotlib import pyplot as plt

noisy_moons = make_moons(n_samples=100, noise=0.1)
blobs = make_blobs(n_samples=100, centers=3 , center_box = (-1,1),cluster_std=0.1)

colors = np.array(['#377eb8', '#ff7f00', '#4daf4a',
                   '#f781bf', '#a65628', '#984ea3',
                   '#999999', '#e41a1c', '#dede00'])

#defining the clustering algo which we want to try
clustering_models = [KMeans,DBSCAN]

from collections import namedtuple
Model = namedtuple('Model', ['name', 'model'])
models = [Model(model.__module__.split('.')[-1][:-1], model) 
          for model in clustering_models]

#defn of params for each dataset|clustering_algo
datasets_w_hyperparams = [(noisy_moons[0], 
                           {models[0][0]: {'n_clusters': 2}, models[1][0]: {'eps': .3, }}),
                          (blobs[0], 
                           {models[0][0]: {'n_clusters': 2}, models[1][0]: {'eps': .1, }})]

f,axes=plt.subplots(len(datasets_w_hyperparams),len(models),figsize = (15,10))
for data_id,(dataset,params) in enumerate(datasets_w_hyperparams):
    for model_id,model in enumerate(models):
        ax = axes[data_id][model_id]
        name, clus_model = model
        pred = clus_model(**params[name]).fit_predict(dataset)
        ax.scatter(dataset[:,0],dataset[:,1], s=20, color= colors[pred])
        ax.set_title(name)
plt.show()