import numpy as np
import pandas as pd
import matplotlib.pylab as plt
df = pd.read_csv("drawndata1.csv")
X = df[['x', 'y']].values
y = df['z'] == "a"
plt.scatter(X[:, 0], X[:, 1], c=y);
To see the effect from the standard scaler you need to run this;
from sklearn.preprocessing import QuantileTransformer
X_new = QuantileTransformer(n_quantiles=100).fit_transform(X)
plt.scatter(X_new[:, 0], X_new[:, 1], c=y);