How a project works

A project starts with one or two data files:

• Train file — your real, collected survey responses.
• Predict file (optional) — the population you want to generate synthetic answers for. If you don't have a separate file, n-Infinite can automatically split a single uploaded file into train and predict portions based on where the questionnaire begins (Swift mode), using the same respondent structure throughout.

n-Infinite accepts common survey and data-export formats: CSV, Excel (XLSX/XLS), JSON, Parquet, Feather and SPSS (.sav) — including SPSS variable labels.

Once uploaded, a project moves through a clear pipeline: initial setup and validation, training, prediction, and completion. Along the way it can be paused, resumed or restarted, and the platform tracks status at every step.

Project types

	Use case	Train file	Predict file	Behavior	Notes
Single File	Boost field projects, API-based projects	Field data (completed surveys) + profiling data	—	Automatically splits the file into train and predict sets	If the last question is not mandatory, add a Completed = Y column to identify complete responses
Split File	Boost field projects: field data + stock profiles	Field data (completed surveys)	Profiling data	Predicts all target columns. The predict file must not contain the names of the columns to be predicted	The last column is not required, even if optional — automatic detection is supported
Complete-the-Incomplete	Convert dropouts into complete surveys	Field data (completed surveys) + field data (dropouts)	—	Reads rows right to left to predict missing data; existing data is used for training, even in incomplete rows	If the last question is not mandatory, add a Completed = Y column
Boost — Swift	Boost projects without profile availability. Ideal for B2B	Field data (completed surveys)	—	Adds an identical number of synthetic prediction rows using existing profiling. Requires post-weighting	Select the initial question to predict and the desired weighting targets
Boost — Express	Boost projects without profile availability. Ideal for big boosts and B2C	Field data (completed surveys)	—	Adds 1× to n× the number of prediction rows; imputes real profiles from national census data (pre-loaded country options). Requires post-weighting	Select the initial question to predict and weighting targets. Census column mapping is required

Typical file structure

Horizontally, a typical n-Infinite file is organized as:

1. Profiling data
   • Respondent identification (UserId, Id, …)
   • Demographics: gender, age, residence, ethnicity, …
   • Extra profiling: psychographics (B2C) or firmographics (B2B)
   • Screening questions (if you prefer to keep them outside training data)
2. Training data — the field answers.

Vertically:

• Single file: train and predict rows are mixed — complete vs. profile-only rows, or complete vs. incomplete rows.
• Split files: training data lives in the train file; in the predict file, delete the column names of the fields to be predicted.

Tips for optimizing files

1. Admitted formats: Excel (.xlsx, .xls), comma-separated (.csv) and SPSS (.sav).
2. Never include duplicated field names.
3. Remove non-profiling information apart from identification: checksums, timestamps, personal information.
4. Avoid low-density fields (more than 90% empty cells).
5. If possible, include a field called Start with a single value (e.g. Y) right before the training data begins.
6. If columns finish unevenly (the last column may be empty), include an End field with a single value (e.g. Y) at the end of the file.

See Preparing your data file for the coding conventions the platform relies on to read your questionnaire's structure correctly on the first try.