Superconductivity#
Data on superconductors, to predict the critical temperature of a superconductor from 81 numeric features. It is a large regression dataset: 21263 samples and 81 features, all of them numbers, without missing values. The features are statistics of properties of the superconductor, so the columns are strongly related to each other.
Overview#
Loader |
|
Task |
Regression |
Samples |
21263 |
Features |
81 |
Targets |
1 continuous target, |
Missing values |
None |
License |
CC BY 4.0 |
Features#
All 81 features are numbers, stored as float64, and none of them is categorical. The UCI
page gives no description of the features, so this page only tells how their names are built.
number_of_elementsis the first column. It holds whole numbers from 1 to 9.The other 80 columns are ten statistics of each of eight properties, written as
<statistic>_<property>, for examplemean_atomic_massorwtd_std_Valence. The columns of one property follow each other.
The ten statistics, in the order of the columns of a property, are mean, wtd_mean,
gmean, wtd_gmean, entropy, wtd_entropy, range, wtd_range, std and
wtd_std. The eight properties, in the order of the columns, are atomic_mass, fie,
atomic_radius, Density, ElectronAffinity, FusionHeat,
ThermalConductivity and Valence. The names are kept as they are in the source, so
some of them have capital letters.
Two more columns, range_atomic_radius and range_Valence, hold whole numbers only.
The dataset has no missing values.
Repeated rows#
66 rows of the source file repeat another row in all columns. They are kept, so the dataset
has all 21263 rows of the source. The file unique_m.csv of the source, which has the
chemical formula of every row, has no repeated row, so these rows appear to be different
compounds that have the same values. When you split the data into a training and a test part,
equal rows can end up in both parts and make the test result look better than it is.
Target#
y is the critical temperature, the column critical_temp of the source. It goes from
0.00021 to 185 and its mean is 34.4212.
The source has a second file, unique_m.csv, with the chemical formula of every
superconductor broken up into the number of atoms of every element. It is not part of this
dataset.
Source and citation#
Source: UCI Machine Learning Repository
Obtained from: UCI Machine Learning Repository, dataset 464, downloaded with the ucimlrepo package: all 21263 examples with 81 inputs and the target. The ucimlrepo version is recorded in the file.
Creators: Kam Hamidieh
Please cite the dataset as:
Hamidieh, K. (2018). Superconductivty Data [Dataset]. UCI Machine Learning Repository. https://doi.org/10.24432/C53P47.
See also:
Hamidieh, K. (2018). A data-driven statistical model for predicting the critical temperature of a superconductor. Computational Materials Science.
License#
CC BY 4.0, as stated by the source. The license was checked at the source on 2026-11-14. See License and citation.
Changes made#
Converted to the NumPy NPZ format. The features and the target are stored as float64.
The data comes from the file train.csv of the source, which ucimlrepo gives. The file unique_m.csv, which has the chemical formula of every row, is not stored.
The names of the features and of the target are the column names of the source.
Note
The 81 features are number_of_elements and ten statistics (mean, wtd_mean, gmean, wtd_gmean, entropy, wtd_entropy, range, wtd_range, std and wtd_std) of eight properties (atomic_mass, fie, atomic_radius, Density, ElectronAffinity, FusionHeat, ThermalConductivity and Valence). The UCI page gives no description of the features, so only their names are documented here.
Note
The source file train.csv has 66 rows that repeat another row in all 82 columns. The file unique_m.csv of the source, with the chemical formula of every row, has no repeated row, so these rows appear to be different compounds that have the same values. All 21263 rows are kept. Keep this in mind when you split the data: equal rows in the training and in the test part make the test result look better.
Note
number_of_elements, range_atomic_radius and range_Valence hold whole numbers only. There are no missing values.
Note
The target critical_temp goes from 0.00021 to 185 in the source file.
Note
According to the UCI page the original data comes from http://supercon.nims.go.jp/index_en.html, which it calls public. The license above is the one that the UCI page states for this dataset.
Note
The UCI page spells the name of the dataset Superconductivty. The name here is corrected, while the URL of the source and the citation keep the spelling of UCI.
Example#
>>> from pyml_datasets import load_superconductivity
>>> data = load_superconductivity()
>>> data.X.shape, data.y.shape
((21263, 81), (21263,))
>>> data.has_missing
False
>>> data.feature_names[:2]
('number_of_elements', 'mean_atomic_mass')
>>> data.target_names
('critical_temp',)
>>> float(data.y.max())
185.0