I'm trying to cluster a group of points in a probabilistic manner. Using below, I have a single set of xy points, which are recorded in X and Y. I want to cluster into groups using a reference point, which is displayed in X2 and Y2.
With the help of an answer the current approach is to measure the distance from the reference point and group using k-means. Although, it provides a method to cluster using the reference point, the hard cutoff and adherence to k clusters makes it somewhat unsuitable when dealing with numerous datasets. For instance, the number of clusters needed for this example is probably 3. But a separate example may different. I'd have to manually go through and alter k every time.
Given the non-probabilistic nature of k-means a separate option could be GMM. Is it possible to account for the reference point when modelling? If I attach the output below the underlying model isn't clustering as I'm hoping for.
If I look at the probability each point is within a group it's not clustered as I'd hoped. With this I run into the same problem with manually altering the amount of components. Because the points are distributed randomly, using “AIC” or “BIC” to select the appropriate number of clusters doesn't work. There is no optimal number.
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.cluster import KMeans
df = pd.DataFrame({
'X' : [-1.0,-1.0,0.5,0.0,0.0,2.0,3.0,5.0,0.0,-2.5,2.0,8.0,-10.5,15.0,-20.0,-32.0,-20.0,-20.0,-10.0,20.5,0.0,20.0,-30.0,-15.0,20.0,-15.0,-10.0],
'Y' : [0.0,1.0,-0.5,0.5,-0.5,0.0,1.0,4.0,5.0,-3.5,-2.0,-8.0,-0.5,-10.5,-20.5,0.0,16.0,-15.0,5.0,13.5,20.0,-20.0,2.0,-17.5,-15,19.0,20.0],
'X2' : [0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0],
'Y2' : [0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0],
})
k-means:
df['distance'] = np.sqrt(df['X']**2 + df['Y']**2)
df['distance'] = np.sqrt((df['X2'] - df['Y2'])**2 + (df['BallY'] - df['y_post'])**2)
model = KMeans(n_clusters = 2)
model_data = np.array([df['distance'].values, np.zeros(df.shape[0])])
model.fit(model_data.T)
df['group'] = model.labels_
plt.scatter(df['X'], df['Y'], c = model.labels_, cmap = 'bwr', marker = 'o', s = 5)
plt.scatter(df['X2'], df['Y2'], c ='k', marker = 'o', s = 5)
GMM:
Y_sklearn = df[['X','Y']].values
gmm = mixture.GaussianMixture(n_components=3, covariance_type='diag', random_state=42)
gmm.fit(Y_sklearn)
labels = gmm.predict(Y_sklearn)
df['group'] = labels
plt.scatter(Y_sklearn[:, 0], Y_sklearn[:, 1], c=labels, s=5, cmap='viridis');
plt.scatter(df['X2'], df['Y2'], c='red', marker = 'x', edgecolor = 'k', s = 5, zorder = 10)
proba = pd.DataFrame(gmm.predict_proba(Y_sklearn).round(2)).reset_index(drop = True)
df_pred = pd.concat([df, proba], axis = 1)
In my opinion, if you want to define clusters as "regions where points are close to each other", you should use DBSCAN.
This clustering algorithm finds clusters by looking at regions where points are close to each other (i.e. dense regions), and are separated from other clusters by regions where points are less dense.
This algorithm can categorize points as noise (outliers). Outliers are labelled -1.
They are points that do not belong to any cluster.
Here is some code to perform DBSCAN clustering, and to insert the cluster labels as a new categorical column in the original Y_sklearn DataFrame. It also prints how many clusters and how many outliers are found.
import numpy as np
import pandas as pd
from sklearn.cluster import DBSCAN
Y_sklearn = df.loc[:, ["X", "Y"]].copy()
n_points = Y_sklearn.shape[0]
dbs = DBSCAN()
labels_clusters = dbs.fit_predict(Y_sklearn)
#Number of found clusters (outliers are not considered a cluster).
n_clusters = labels_clusters.max() + 1
print(f"DBSCAN found {n_clusters} clusters in dataset with {n_points} points.")
#Number of found outliers (possibly no outliers found).
n_outliers = np.count_nonzero((labels_clusters == -1))
if n_outliers:
print(f"{n_outliers} outliers were found.\n")
else:
print(f"No outliers were found.\n")
#Add cluster labels as a new column to original DataFrame.
Y_sklearn["cluster"] = labels_clusters
#Setting `cluster` column to Categorical dtype makes seaborn function properly treat
#cluster labels as categorical, and not numerical.
Y_sklearn["cluster"] = Y_sklearn["cluster"].astype("category")
If you want to plot the results, I suggest you use Seaborn. Here is some code to plot the points of Y_sklearn DataFrame, and color them by the cluster they belong to. I also define a new color palette, which is just the default Seaborn color palette, but where outliers (with label -1) will be in black.
import matplotlib.pyplot as plt
import seaborn as sns
name_palette = "tab10"
palette = sns.color_palette(name_palette)
if n_outliers:
color_outliers = "black"
palette.insert(0, color_outliers)
else:
pass
sns.set_palette(palette)
fig, ax = plt.subplots()
sns.scatterplot(data=Y_sklearn,
x="X",
y="Y",
hue="cluster",
ax=ax,
)
Using default hyperparameters, the DBSCAN algorithm finds no cluster in the data you provided: all points are considered outliers, because there is no region where points are significantly more dense. Is that your whole dataset, or is it just a sample? If it is a sample, the whole dataset will have much more points, and DBSCAN will certainly find some high density regions.
Or you can try tweaking the hyperparameters, min_samples and eps in particular. If you want to "force" the algorithm to find more clusters, you can decrease min_samples (default is 5), or increase eps (default is 0.5). Of course, the optimal hyperparamete values depends on the specific dataset, but default values are considered quite good for DBSCAN. So, if the algorithm considers all points in your dataset to be outliers, it means that there are no "natural" clusters!
Do you mean density estimation? You can model your data as a Gaussian Mixture and then get a probability of a point to belong to the mixture. You can use sklearn.mixture.GaussianMixture for that. By changing number of components you can control how many clusters you will have. The metric to cluster on is Euclidian distance from the reference point. So the GMM model will provide you with prediction of which cluster the data point should be classified to.
Since your metric is 1d, you will get a set of Gaussian distributions, i.e. a set of means and variances. So you can easily calculate the probability of any point to be in certain cluster, just by calculating how far it is from the reference point and put the value in the normal distribution pdf formula.
To make image more clear, I'm changing the reference point to (-5, 5) and select number of clusters = 4. In order to get the best number of clusters, use some metric that minimizes total variance and penalizes growth of number of mixtures. For example argmin(model.covariances_.sum()*num_clusters)
import pandas as pd
from sklearn.mixture import GaussianMixture
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm
df = pd.DataFrame({
'X' : [-1.0,-1.0,0.5,0.0,0.0,2.0,3.0,5.0,0.0,-2.5,2.0,8.0,-10.5,15.0,-20.0,-32.0,-20.0,-20.0,-10.0,20.5,0.0,20.0,-30.0,-15.0,20.0,-15.0,-10.0],
'Y' : [0.0,1.0,-0.5,0.5,-0.5,0.0,1.0,4.0,5.0,-3.5,-2.0,-8.0,-0.5,-10.5,-20.5,0.0,16.0,-15.0,5.0,13.5,20.0,-20.0,2.0,-17.5,-15,19.0,20.0],
})
ref_X, ref_Y = -5, 5
dist = np.sqrt((df.X-ref_X)**2 + (df.Y-ref_Y)**2)
n_mix = 4
gmm = GaussianMixture(n_mix)
model = gmm.fit(dist.values.reshape(-1,1))
x = np.linspace(-35., 35.)
y = np.linspace(-30., 30.)
X, Y = np.meshgrid(x, y)
XX = np.sqrt((X.ravel() - ref_X)**2 + (Y.ravel() - ref_Y)**2)
Z = model.score_samples(XX.reshape(-1,1))
Z = Z.reshape(X.shape)
# plot grid points probabilities
plt.set_cmap('plasma')
plt.contourf(X, Y, Z, 40)
plt.scatter(df.X, df.Y, c=model.predict(dist.values.reshape(-1,1)), edgecolor='black')
You can read more here and here
P.S. score_samples() returns log likelihoods, use exp() to convert to probability
Taking your centre point of 0,0 we can calculate the Euclidean distance from this point to all points in your df.
df['distance'] = np.sqrt(df['X']**2 + df['Y']**2)
If you have a centre point other than zero it would be:
df['distance'] = np.sqrt((centre_point_x - df['X'])**2 + (centre_point_y - df['Y'])**2)
Using your data and chart as before, we can plot this and see the distance metric increasing as we move away from the centre.
fig, ax = plt.subplots(figsize = (6,6))
ax.scatter(df['X'], df['Y'], c = df['distance'], cmap = 'viridis', marker = 'o', s = 30)
ax.set_xlim([-35, 35])
ax.set_ylim([-35, 35])
plt.show()
K-means
We can now use this distance data and use it to calculate K-means clusters as you did before, but this time using the distance data and an array of zeros (zeros because this k-means requires a 2d-array but we only want to split the 1d aray of dimensional data. So the zeros act as 'filler'
model = KMeans(n_clusters = 2) #choose how many clusters
# create this 2d array for the KMeans model
model_data = np.array([df['distance'].values, np.zeros(df.shape[0])])
model.fit(model_data.T) # transformed array because the above code produces
# data with 27 columns and 2 rows but we want it the other way round
df['group'] = model.labels_ # put the labels into the dataframe
Then we can plot the results
fig, ax = plt.subplots(figsize = (6,6))
ax.scatter(df['X'], df['Y'], c = df['group'], cmap = 'viridis', marker = 'o', s = 30)
ax.set_xlim([-35, 35])
ax.set_ylim([-35, 35])
plt.show()
With three clusters we get the following result:
Other clustering methods
Check out SKlearn's clustering page for more options. I experimented with DBSCAN with some good results but it depends on what you are trying to achieve exactly. Check out the table underneath their example charts to see how they each compare.
In the new implementation of cox ph survival model in xgboost 0.81 how does one specify start and end time of an event?
Thanks
The R equivalent function would be for example :
cph_mod = coxph(Surv(Start, Stop, Status) ~ Age + Sex + SBP, data=data)
XGBoost do not allow for start (i.e. delayed entry). If it makes sense for the application, you can always change the underlying time scale so all subjects start at time=0. However, XGBoost does allow for right censored data. It seems impossible to find any documentation/example for how to implement a Cox model, but from the source code you can read "Cox regression for censored survival data (negative labels are considered censored)."
Here is a short example for anyone who want to try XGBoost with obj="survival:cox". We can compare the results to to the scikit-learn survival package sksurv. To make XGBoost more similar to that framework we use a linear booster instead of a tree booster.
import pandas as pd
import xgboost as xgb
from sksurv.datasets import load_aids
from sksurv.linear_model import CoxPHSurvivalAnalysis
# load and inspect the data
data_x, data_y = load_aids()
data_y[10:15]
Out[586]:
array([(False, 334.), (False, 285.), (False, 265.), ( True, 206.),
(False, 305.)], dtype=[('censor', '?'), ('time', '<f8')])
# Since XGBoost only allow one column for y, the censoring information
# is coded as negative values:
data_y_xgb = [x[1] if x[0] else -x[1] for x in data_y]
data_y_xgb[10:15]
Out[3]: [-334.0, -285.0, -265.0, 206.0, -305.0]
data_x = data_x[['age', 'cd4']]
data_x.head()
Out[4]:
age cd4
0 34.0 169.0
1 34.0 149.5
2 20.0 23.5
3 48.0 46.0
4 46.0 10.0
# Since sksurv output log hazard ratios (here relative to 0 on predictors)
# we must use 'output_margin=True' for comparability.
estimator = CoxPHSurvivalAnalysis().fit(data_x, data_y)
gbm = xgb.XGBRegressor(objective='survival:cox',
booster='gblinear',
base_score=1,
n_estimators=1000).fit(data_x, data_y_xgb)
prediction_sksurv = estimator.predict(data_x)
predictions_xgb = gbm.predict(data_x, output_margin=True)
d = pd.DataFrame({'xgb': predictions_xgb,
'sksurv': prediction_sksurv})
d.head()
Out[13]:
sksurv xgb
0 -1.892490 -1.843828
1 -1.569389 -1.524385
2 0.144572 0.207866
3 0.519293 0.502953
4 1.062392 1.045287
d.plot.scatter('xgb', 'sksurv')
Note that these are predictions on the same data that was use to fit the model. It seems that XGBoost get the values right but sometimes with a linear transformation. I do not know why. Play around with base_score and n_estimators. Perhaps someone can add to this answer.
I have some categorical features in my data along with continuous ones. Is it a good or absolutely bad idea to hot encode category features to find correlation of it to labels along with other continuous creatures?
There is a way to calculate the correlation coefficient without one-hot encoding the category variable. Cramers V statistic is one method for calculating the correlation of categorical variables. It can be calculated as follows. The following link is helpful. Using pandas, calculate Cramér's coefficient matrix For variables with other continuous values, you can categorize by using cut of pandas.
import numpy as np
import pandas as pd
import scipy.stats as ss
import seaborn as sns
print('Pandas version:', pd.__version__)
# Pandas version: 1.3.0
tips = sns.load_dataset("tips")
tips["total_bill_cut"] = pd.cut(tips["total_bill"],
np.arange(0, 55, 5),
include_lowest=True,
right=False)
def cramers_v(confusion_matrix):
""" calculate Cramers V statistic for categorial-categorial association.
uses correction from Bergsma and Wicher,
Journal of the Korean Statistical Society 42 (2013): 323-328
"""
chi2 = ss.chi2_contingency(confusion_matrix)[0]
n = confusion_matrix.sum()
phi2 = chi2 / n
r, k = confusion_matrix.shape
phi2corr = max(0, phi2 - ((k-1)*(r-1))/(n-1))
rcorr = r - ((r-1)**2)/(n-1)
kcorr = k - ((k-1)**2)/(n-1)
return np.sqrt(phi2corr / min((kcorr-1), (rcorr-1)))
confusion_matrix = pd.crosstab(tips["day"], tips["time"])
cramers_v(confusion_matrix.values)
# Out[2]: 0.9386619340722221
confusion_matrix = pd.crosstab(tips["total_bill_cut"], tips["time"])
cramers_v(confusion_matrix.values)
# Out[3]: 0.1649870749498837
please note the .as_matrix() is deprecated in pandas since verison 0.23.0 . use .values instead
I found phik library quite useful in calculating correlation between categorical and interval features. This is also useful for binning numerical features. Try this once: phik documentation
I was looking to do same thing in BigQuery.
For numeric features you can use built in CORR(x,y) function.
For categorical features, you can calculate it as:
cardinality (cat1 x cat2) / max (cardinality(cat1), cardinality(cat2).
Which translates to following SQL:
SELECT
COUNT(DISTINCT(CONCAT(cat1, cat2))) / GREATEST (COUNT(DISTINCT(cat1)), COUNT(DISTINCT(cat2))) as cat1_2,
COUNT(DISTINCT(CONCAT(cat1, cat3))) / GREATEST (COUNT(DISTINCT(cat1)), COUNT(DISTINCT(cat3))) as cat1_3,
....
FROM ...
Higher number means lower correlation.
I used following python script to generate SQL:
import itertools
arr = range(1,10)
query = ',\n'.join(list('COUNT(DISTINCT(CONCAT({a}, {b}))) / GREATEST (COUNT(DISTINCT({a})), COUNT(DISTINCT({b}))) as cat{a}_{b}'.format(a=a,b=b)
for (a,b) in itertools.combinations(arr,2)))
query = 'SELECT \n ' + query + '\n FROM `...`;'
print (query)
It should be straightforward to do same thing in numpy.
I'm pursuing course 2 on this coursera course on linear regression (https://www.coursera.org/specializations/machine-learning)
I've solved the training using graphlab but wanted to try out sklearn for the experience and learning. I'm using sklearn and pandas for this.
The model overfits on the data. How can I fix this? This is the code.
These are the coefficients i'm getting.
[ -3.33628603e-13 1.00000000e+00]
poly1_data = polynomial_dataframe(sales["sqft_living"], 1)
poly1_data["price"] = sales["price"]
model1 = LinearRegression()
model1.fit(poly1_data, sales["price"])
print(model1.coef_)
plt.plot(poly1_data['power_1'], poly1_data['price'], '.',poly1_data['power_1'], model1.predict(poly1_data),'-')
plt.show()
The plotted line is like this. As you see it connects every data point.
and this is the plot of the input data
I wouldn't even call this overfit. I'd say you aren't doing what you think you should be doing. In particular, you forgot to add a column of 1's to your design matrix, X. For example:
# generate some univariate data
x = np.arange(100)
y = 2*x + x*np.random.normal(0,1,100)
df = pd.DataFrame([x,y]).T
df.columns = ['x','y']
You're doing the following:
model1 = LinearRegression()
X = df["x"].values.reshape(1,-1)[0] # reshaping data
y = df["y"].values.reshape(1,-1)[0]
model1.fit(X,y)
Which leads to:
plt.plot(df['x'].values, df['y'].values,'.')
plt.plot(X[0], model1.predict(X)[0],'-')
plt.show()
Instead, you want to add a column of 1's to your design matrix (X):
X = np.column_stack([np.ones(len(df['x'])),df["x"].values.reshape(1,-1)[0]])
y = df["y"].values.reshape(1,-1)
model1.fit(X,y)
And (after some reshaping) you get:
plt.plot(df['x'].values, df['y'].values,'.')
plt.plot(df['x'].values, model1.predict(X),'-')
plt.show()