Explore dataset statistics

Explore statistics over split data

Datasets Server provides a /statistics endpoint for fetching some basic statistics precomputed for a requested dataset. This will get you a quick insight on how the data is distributed.

Currently, statistics are computed only for datasets with Parquet exports.

The /statistics endpoint requires three query parameters:

• dataset: the dataset name, for example glue

• config: the configuration name, for example cola

• split: the split name, for example train

Let’s get some stats for glue dataset, cola config, train split:

PythonJavaScriptcURLCopied

import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://datasets-server.boincai.com/statistics?dataset=glue&config=cola&split=train"
def query():
    response = requests.get(API_URL, headers=headers)
    return response.json()
data = query()

The response JSON contains two keys:

• num_examples - number of samples in a split

• statistics - list of dictionaries of statistics per each column, each dictionary has three keys: column_name, column_type, and column_statistics. Content of column_statistics depends on a column type, see Response structure by data types for more details

Copied

{
  "num_examples": 8551,
  "statistics": [
    {
      "column_name": "idx",
      "column_type": "int",
      "column_statistics": {
        "nan_count": 0,
        "nan_proportion": 0,
        "min": 0,
        "max": 8550,
        "mean": 4275,
        "median": 4275,
        "std": 2468.60541,
        "histogram": {
          "hist": [
            856,
            856,
            856,
            856,
            856,
            856,
            856,
            856,
            856,
            847
          ],
          "bin_edges": [
            0,
            856,
            1712,
            2568,
            3424,
            4280,
            5136,
            5992,
            6848,
            7704,
            8550
          ]
        }
      }
    },
    {
      "column_name": "label",
      "column_type": "class_label",
      "column_statistics": {
        "nan_count": 0,
        "nan_proportion": 0,
        "no_label_count": 0,
        "no_label_proportion": 0,
        "n_unique": 2,
        "frequencies": {
          "unacceptable": 2528,
          "acceptable": 6023
        }
      }
    },
    {
      "column_name": "sentence",
      "column_type": "string_text",
      "column_statistics": {
        "nan_count": 0,
        "nan_proportion": 0,
        "min": 6,
        "max": 231,
        "mean": 40.70074,
        "median": 37,
        "std": 19.14431,
        "histogram": {
          "hist": [
            2260,
            4512,
            1262,
            380,
            102,
            26,
            6,
            1,
            1,
            1
          ],
          "bin_edges": [
            6,
            29,
            52,
            75,
            98,
            121,
            144,
            167,
            190,
            213,
            231
          ]
        }
      }
    }
  ]
}

Response structure by data type

Currently, statistics are supported for strings, float and integer numbers, and the special datasets.ClassLabel feature type of the datasets library.

column_type in response can be one of the following values:

• class_label - for datasets.ClassLabel feature
• float - for float dtypes

• int - for integer dtypes

• string_label - for string dtypes, if there are less than or equal to 30 unique values in a string column in a given split

• string_text - for string dtypes, if there are more than 30 unique values in a string column in a given split

class_label

This type represents categorical data encoded as ClassLabel feature. The following measures are computed:

• number and proportion of null values

• number and proportion of values with no label

• number of unique values (excluding null and no label)

• value counts for each label (excluding null and no label)

Example

float

The following measures are returned for float data types:

• minimum, maximum, mean, and standard deviation values

• number and proportion of null values

• histogram with 10 bins

Example

int

The following measures are returned for integer data types:

• minimum, maximum, mean, and standard deviation values

• number and proportion of null values

• histogram with less than or equal to 10 bins

Example

string_label

If string column has less than or equal to 30 unique values within the requested split, it is considered to be a category. The following measures are returned:

• number and proportion of null values

• number of unique values (excluding null)

• value counts for each label (excluding null)

Example

string_text

If string column has more than 30 unique values within the requested split, it is considered to be a column containing texts and response contains statistics over text lengths. The following measures are computed:

• minimum, maximum, mean, and standard deviation of text lengths

• number and proportion of null values

• histogram of text lengths with 10 bins

Example

←Get the number of rows and the bytes sizeOverview→Explore statistics over split dataResponse structure by data typeclass_labelfloatintstring_labelstring_text