Why DBSCAN clustering returns single cluster on Movie lens data set?

Question

The Scenario:

I'm performing Clustering over Movie Lens Dataset, where I have this Dataset in 2 formats:

OLD FORMAT:

uid iid rat
941 1   5
941 7   4
941 15  4
941 117 5
941 124 5
941 147 4
941 181 5
941 222 2
941 257 4
941 258 4
941 273 3
941 294 4

NEW FORMAT:

uid 1               2               3               4
1   5               3               4               3
2   4               3.6185548023    3.646073985     3.9238342172
3   2.8978348799    2.6692556753    2.7693015618    2.8973463681
4   4.3320762062    4.3407749532    4.3111995162    4.3411425423
940 3.7996234581    3.4979386925    3.5707888503    2
941 5               NaN             NaN             NaN
942 4.5762594612    4.2752554573    4.2522440019    4.3761477591
943 3.8252406362    5               3.3748860659    3.8487417604

over which I need to perform Clustering using KMeans, DBSCAN and HDBSCAN. With KMeans I'm able to set and get clusters.

The Problem

The Problem persists only with DBSCAN & HDBSCAN that I'm unable to get enough amount of clusters (I do know we cannot set Clusters manually)

Techniques Tried:

• Tried this with IRIS data-set, where I found Species wasn't included. Clearly that is in String and besides is to be predicted, and everything just works fine with that Dataset (Snippet 1)
• Tried with Movie Lens 100K dataset in OLD FORMAT (with and without UID) since I tried an Analogy that, UID == SPECIES and hence tried without it. (Snippet 2)
• Tried same with NEW FORMAT (with and without UID) yet the results ended up in same style.

Snippet 1:

print "\n\n FOR IRIS DATA-SET:"
from sklearn.datasets import load_iris

iris = load_iris()
dbscan = DBSCAN()

d = pd.DataFrame(iris.data)
dbscan.fit(d)
print "Clusters", set(dbscan.labels_)

Snippet 1 (Output):

FOR IRIS DATA-SET:
Clusters set([0, 1, -1])
Out[30]: 
array([ 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, -1,  0,  0,  0,  0,  0,  0,  0,  0,  1,
        1,  1,  1,  1,  1,  1, -1,  1,  1, -1,  1,  1,  1,  1,  1,  1,  1,
       -1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,
        1,  1, -1,  1,  1,  1,  1,  1, -1,  1,  1,  1,  1, -1,  1,  1,  1,
        1,  1,  1, -1, -1,  1, -1, -1,  1,  1,  1,  1,  1,  1,  1, -1, -1,
        1,  1,  1, -1,  1,  1,  1,  1,  1,  1,  1,  1, -1,  1,  1, -1, -1,
        1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1])

Snippet 2:

import pandas as pd
from sklearn.cluster import DBSCAN

data_set = pd.DataFrame

ch = int(input("Extended Cluster Methods for:\n1. Main Matrix IBCF \n2. Main Matrix UBCF\nCh:"))
if ch is 1:
    data_set = pd.read_csv("MainMatrix_IBCF.csv")
    data_set = data_set.iloc[:, 1:]
    data_set = data_set.dropna()
elif ch is 2:
    data_set = pd.read_csv("MainMatrix_UBCF.csv")
    data_set = data_set.iloc[:, 1:]
    data_set = data_set.dropna()
else:
    print "Enter Proper choice!"

print "Starting with DBSCAN for Clustering on\n", data_set.info()

db_cluster = DBSCAN()
db_cluster.fit(data_set)
print "Clusters assigned are:", set(db_cluster.labels_)

Snippet 2 (Output):

Extended Cluster Methods for:
1. Main Matrix IBCF 
2. Main Matrix UBCF
Ch:>? 1
Starting with DBSCAN for Clustering on
<class 'pandas.core.frame.DataFrame'>
Int64Index: 942 entries, 0 to 942
Columns: 1682 entries, 1 to 1682
dtypes: float64(1682)
memory usage: 12.1 MB
None
Clusters assigned are: set([-1])

As seen, it returns only 1 Cluster. I'd like to hear what am I doing wrong.

Answer 1

As pointed by @faraway and @Anony-Mousse, the solution is more Mathematical on Dataset than Programming.

Could finally figure out the clusters. Here's how:

db_cluster = DBSCAN(eps=9.7, min_samples=2, algorithm='ball_tree', metric='minkowski', leaf_size=90, p=2)
arr = db_cluster.fit_predict(data_set)
print "Clusters assigned are:", set(db_cluster.labels_)

uni, counts = np.unique(arr, return_counts=True)
d = dict(zip(uni, counts))
print d

The Epsilon and Out-lier concept turned out more brightening from SO: How can I choose eps and minPts (two parameters for DBSCAN algorithm) for efficient results?.