Table EDA

See also

UI Tutorial: Set Default Cleaning Operations | Concepts: Table EDA | API Tutorial: Credit Default — Step 6

Prerequisites

This page uses the client and wait_for_task helpers defined in API Overview.

After registering a table via the SDK, you can run exploratory data analysis through the API to discover data quality issues at the table and column level.

Run Table EDA

client = fb.Configurations().get_client()

response = client.post(
    "/table_eda",
    json={
        "table_id": table_id,
        "analysis_size": 10000,
        "seed": 42,
    },
)
task_id = response.json()["id"]
wait_for_task(client, task_id)

Parameters:

Parameter	Type	Required	Description
table_id	string	Yes	ID of the registered table
analysis_size	integer	Yes	Number of rows to sample for analysis
seed	integer	No	Random seed for reproducibility
from_timestamp	string	No	Start timestamp filter (ISO format)
to_timestamp	string	No	End timestamp filter (ISO format)
timestamp_column	string	No	Column to use for timestamp filtering
overwrite	boolean	No	Overwrite existing EDA results (default: false)

List Table EDA Runs

response = client.get(
    "/table_eda",
    params={"table_id": table_id, "page": 1, "page_size": 20},
)
eda_runs = response.json()["data"]
table_analysis_id = eda_runs[0]["_id"]

Response fields (each item in data):

Field	Type	Description
id	string	Table EDA ID
table_id	string	Table that was analyzed
analysis_size	integer	Number of rows sampled
seed	integer	Random seed used
last_seen	datetime	When the analysis was last run
from_timestamp	datetime	Start of analysis time range (if filtered)
to_timestamp	datetime	End of analysis time range (if filtered)

View Column-Level Analysis

List Column Analyses

response = client.patch(
    "/column_analysis",
    json={
        "table_analysis_id": table_analysis_id,
    },
)
columns = response.json()["data"]

Response fields (each item in data):

Field	Type	Description
id	string	Column analysis ID
table_analysis_id	string	Parent table EDA ID
table_id	string	Table ID
column_name	string	Column being analyzed
column_dtype	string	Data type (e.g., "INT", "FLOAT", "VARCHAR")
column_description	string	Column description
issues	array	Detected data quality issues (missing values, outliers, etc.)
cleaning_reco	array	Recommended cleaning operations
semantic_level_1	string	Detected semantic type
summary	object	Summary statistics (count, unique, min, max, mean, etc.)
outdated	boolean	Whether the analysis needs to be refreshed

Filter to only show columns with issues:

response = client.patch(
    "/column_analysis",
    json={
        "table_analysis_id": table_analysis_id,
        "issues": True,
    },
)

Get Detailed Column Analysis

column_analysis_id = columns[0]["_id"]

response = client.patch(f"/column_analysis/{column_analysis_id}")
analysis = response.json()

The response contains the same fields as each item in the column analysis list above.

Column EDA Plots

Generate interactive Bokeh plots for a column's data distribution.

Get Plot Options

Check what categories are available for the column (e.g., dictionary keys or embedding dimensions):

response = client.request(
    "OPTIONS",
    f"/column_eda/{column_analysis_id}",
    params={"cleaned": False},
)
options = response.json()

Parameters:

Parameter	Type	Required	Description
cleaned	boolean	Yes	false for raw data categories, true for categories after cleaning operations

Response fields:

Field	Type	Description
column_categories	array	Available categories for the column (dictionary keys or embedding dimensions to filter by)

Get Rendered Plots

response = client.patch(
    f"/column_eda/{column_analysis_id}",
    params={
        "cleaned": False,
        "height": 500,
        "width": 1000,
        "font_size": 16,
        "output_format": "html",
    },
)
plots = response.json()

Parameters:

Parameter	Type	Required	Description
cleaned	boolean	Yes	false for raw data, true for data after cleaning operations are applied
column_category	any	No	Filter by dictionary key or embedding dimension (from OPTIONS response)
height	integer	No	Plot height in pixels (default: 500)
width	integer	No	Plot width in pixels (default: 1000)
font_size	integer	No	Font size in pixels (default: 16)
output_format	string	No	Output format: "html" (default) or "json"

Response fields (each item in the response array):

Field	Type	Description
plot_type	string	Type of plot (e.g., "distribution", "histogram")
plots	array	List of rendered plot objects, each with a content field containing self-contained Bokeh HTML

The response is a plot list. See Displaying Plots for how to extract and render plots in Jupyter, save as HTML, or embed in a web application.

Compare Raw vs Cleaned

Generate plots for both raw and cleaned data to see the effect of cleaning operations:

# Raw data
raw_plots = client.patch(
    f"/column_eda/{column_analysis_id}",
    params={"cleaned": False},
).json()

# After cleaning operations
cleaned_plots = client.patch(
    f"/column_eda/{column_analysis_id}",
    params={"cleaned": True},
).json()