Table Fill Nulls

This node fills missing (null/NaN) values in a single table column using one of several strategies: constant value, forward fill, backward fill, mean, or median. It reports how many nulls were replaced and returns both a human-readable summary and a structured JSON log for auditing. Use it to clean and impute incomplete datasets before analysis, joins, or modeling.

Usage

Use Table Fill Nulls whenever you have missing values in a specific column and need a consistent way to handle them before downstream processing. Typical scenarios include preparing CSV-imported datasets, cleaning analytics tables, or making features usable for ML models.

Position this node after a table creation or import node such as SaltTableFromCSV, SaltTableFromJSON, or another transforming node like SaltTableFilter, SaltTableSort, or SaltTableGroupBy. It outputs a cleaned table that can then be passed to nodes like SaltTableJoin, SaltTableCastType, SaltChart, or SaltTableOutput.

Common patterns: - Use strategy = "constant" to use a default like "Unknown" for text columns or 0 for numeric columns. - Use "forward_fill" or "backward_fill" on time-ordered data (for example daily metrics) to propagate last known values. - Use "mean" or "median" to impute numeric columns such as prices, scores, or counts. - Chain multiple Table Fill Nulls nodes to clean several columns one by one (for example fill country, then age, then income).

Best practices: - Ensure the column name matches exactly; otherwise the node will fail with a clear error listing available columns. - Only use mean/median on numeric columns; non-numeric columns will throw an error. - When using constant, choose a value that makes sense analytically (for example "N/A", 0, or a clear sentinel) and be aware that it will be auto-cast for numeric columns when possible.

Inputs

Field	Required	Type	Description	Example
table	True	TABLE	Input tabular data containing the column with nulls to fill. Accepts TABLE or DATAFRAME types and converts internally as needed. The table must already have clearly labeled columns; rows may contain null/NaN/None values.	A TABLE representing customer records with columns: ["customer_id", "country", "age", "lifetime_value"], where some rows have null in "country" and "age".
column	True	STRING	Name of the column whose null values will be filled. Must exactly match one of the table's column names; otherwise the node raises an error listing all available columns.	country
strategy	True	STRING	Strategy used to replace nulls in the selected column. Allowed values: "constant" (use a specific value), "forward_fill" (copy down the last non-null value), "backward_fill" (copy up the next non-null value), "mean" (fill with column mean, numeric only), "median" (fill with column median, numeric only).	forward_fill
fill_value	False	STRING	Value used when strategy is "constant". Provided as a string, but will be auto-cast to the column's data type when possible (for example "0" becomes numeric 0.0 for numeric columns). Ignored for other strategies.	Unknown

Outputs

Field	Type	Description	Example
table	TABLE	The output table with the specified column's null values filled according to the chosen strategy. All original columns and rows are preserved; only that column's values are modified and metadata is carried through.	A TABLE where the "country" column has no null values; previously null entries are now "Unknown" or have been forward-filled from the last known value, depending on the strategy used.
filled_count	INT	Number of null values that were actually replaced in the target column. This is the difference between the null count before and after the operation and may be 0 if the column had no nulls.	42
log	STRING	Human-readable summary of the operation, including column name, strategy, and value (for constant). Useful for quick inspection in logs or debug panels.	Filled 42 nulls in [country] using constant (value: Unknown)
log_json	STRING	Structured JSON log string describing the operation, including node name, input and output row counts, parameters (column, strategy, fill_value), and filled_count. Intended for programmatic auditing and monitoring.	{"node":"SaltTableFillNull","input_rows":1000,"output_rows":1000,"parameters":{"column":"country","strategy":"constant","fill_value":"Unknown"},"filled_count":42}

Important Notes

• Performance: The node operates on a single column and processes in-memory; performance is generally good even for large tables, but extremely large datasets may be constrained by available memory.
• Limitations: The "mean" and "median" strategies require the target column to be numeric; using them on non-numeric columns will raise an explicit error instead of silently failing.
• Behavior: For the "constant" strategy, the fill_value is provided as a string but is auto-cast for numeric columns when possible; invalid casts fall back to the original string, which may cause type mixing in some workflows.
• Behavior: Forward and backward fill rely on the existing row order; ensure your table is sorted appropriately (for example by timestamp) before using these strategies to avoid misleading fills.
• Behavior: Only the specified column is modified; row counts do not change and no rows are dropped, making this safe to use before joins or aggregations.

Troubleshooting

• Column 'X' not found. Available: ...: This occurs when the column input does not match any column in the table. Double-check spelling, case, and whitespace, and choose one of the listed available columns.
• 'mean' strategy requires a numeric column. 'X' is not numeric.: You selected strategy = "mean" (or "median") on a non-numeric column. Switch to a different strategy (for example "constant") or first convert the column using SaltTableCastType to a numeric type.
• Unexpected mixed types after constant fill: When using strategy = "constant" on a numeric column with a non-numeric fill_value (for example "N/A"), the node attempts to cast but may fall back to the string. Use a numeric-compatible fill_value (for example "0") or cast the column to string first.
• Filled count is 0 when I expect changes: This indicates there were no null values detected in the target column. Confirm that the missing entries are actually null/NaN/None and not placeholder strings like "N/A" or an empty string; you may need a preprocessing step to convert placeholder values to true nulls before filling.