Step 1: Data Preparation
The first step of data analysis is like preparing ingredients before cooking. Do this step solidly, and the modeling that follows will go much more smoothly.
💡 Lazy path: If you don't want to configure the data structure manually, you can take the AI Build route — drag in an Excel file, pick a modeling type, and leave the rest to AI. See AI Automatic Modeling for details. This page covers the complete manual configuration workflow.
📥 Drag the Data In
Open the software and you will see a clean welcome interface.
1. Prepare the file
The platform supports Excel format: .xlsx and .xls (file size limit 50MB).
🔗 Regression sample data: 点击下载 钢材_demo.xlsx
🔗 Classification sample data: 点击下载 酒品_demo.xlsx
💡 In the help panel on the right of the "Add Instance" dialog, you can also download the three sample datasets for PCA / PLS / PLS-DA directly.
2. One-click import
Drag the file straight into the blue dashed box, or click the button to select it.
3. Two tabs, two paths
The "Add Instance" dialog has two tabs at the top:
| Tab | Applicable scenario |
|---|---|
| General | Configure the role of each column manually (the focus of this page) |
| AI Build | Let AI identify the structure automatically; requires configuring an LLM in Settings first |
⚙️ Tell the Software How to Read
After the file is uploaded, we need to give it a home first, and then briefly "introduce" this dataset.
1. Save the instance and create a model
- Save instance: After a successful upload, click save, and the system will automatically set it as the current working instance.
- Create model: Click "Add Model" to enter model creation, and give your first model a name.
- Select data table: If your Excel has multiple sheets, remember to select the one that contains the data.
2. Two header rows: Header Row vs Name Row
This is the most confusing step. The platform supports two header rows, each with its own job:
| Row | Button | What to put | Example |
|---|---|---|---|
| Header Row | "Set Header Row" | Point / variable English names | x1, x2, TIC, Temp, pH |
| Name Row | "Set Name Row" | Chinese description (optional) | 温度, 压力, 良品率 |
- The header row is usually row 1 and is required — the software relies on it to distinguish each column.
- The name row is usually row 2 and is optional — with it, charts display both the English name and the Chinese name, greatly improving readability.
- How to set it: click the row number to set that row as the header row or the name row.
💡 Only one header row? Then just set the header row and leave the name row empty.
💡 Header not in the first two rows? No problem — just click the row number where it actually is; it is not limited to rows 1 and 2.
3. Data pre-cleaning
After completing the settings above, click any column to link it with the pre-analysis charts on the right.
Data overview: Intuitively view the mean, standard deviation (std), count, and null count.
Null imputation: Click the "Impute" button to open the imputation panel, which supports four methods:
Imputation method Applicable situation Mean Continuous variables with a fairly symmetric distribution (most common) Median More robust when extreme values exist or the distribution is skewed Fill 0 The missingness itself has business meaning (such as "not detected") Custom Fill with a fixed value you specify manually The panel lists the row number, current value, and null count of every missing entry, processed column by column.
Outlier removal: Combined with the trend chart and distribution chart below, use the lasso tool to box-select outliers, or right-click a row number / column header to quickly delete an entire row or column.
4. Assign roles
This is like assigning script roles to actors:
Observation column: This is each sample's "ID card", such as a timestamp or batch number. Select it!
X columns (features): These are the factors that affect the result, such as raw material ratio or reaction temperature. Check them all (select all is supported).
Y columns (target): This is the result you want to predict, such as product purity. When building a prediction model you must select it; for PCA analysis, leave it empty.
Model type: The platform automatically recommends one based on your X/Y configuration, and you can also specify it manually:
Option Description PCA Principal Component Analysis, unsupervised dimensionality reduction and exploration PLS Partial Least Squares regression, X continuous → Y continuous PLS-DA Partial Least Squares Discriminant Analysis, X → Y category OPLS Orthogonal Partial Least Squares regression, strips out orthogonal noise for stronger interpretability OPLS-DA Orthogonal Partial Least Squares Discriminant Analysis, clearer separation between groups After you select one, the interface shows a usage hint for that algorithm (for example, OPLS shows "suitable for supervised learning with orthogonal noise").
Dataset split:
- Test set ratio: Set aside part of the data (such as 20%) for an exam, to verify whether the model is accurate.
- Random or not: Whether to shuffle the data order before splitting. If it is time-series data (in chronological order), choosing not random is recommended.
⚠️ When you need Lag Influence Analysis, you must choose "not random" — that chart analyzes the lead/lag relationship of variables over time, and once the data is shuffled the temporal information is lost.
Once configured, click "Confirm Configuration" and the data is ready! ✅
⚠️ Common Errors and How to Handle Them
| Message | Cause | Handling |
|---|---|---|
Column [xxx] contains empty data, please handle it first | The column has null values | Use "Impute" to handle the nulls |
Column [xxx] contains nulls or non-numeric values, please handle it first | The column contains text/nulls | Check whether you mistakenly selected a text column |
| Please set an observation column / Please select at least one X column | Roles are not fully assigned | Go back to "Column Configuration" and complete them |
| PLS/PLS-DA models require at least one Y column | A supervised model was chosen without selecting a Y | Select a Y column, or switch to PCA |
| Only Excel files can be uploaded | A non-Excel file was uploaded | Convert it to .xlsx / .xls |
| File size cannot exceed 50MB | The file is too large | Split the data, or filter it in Excel first |
🔗 Related Reading
- AI Automatic Modeling —— Let AI complete all the configuration on this page automatically
- Step 2: Modeling Analysis —— Start fitting the model once configuration is done
- Key Terms Explained —— Concepts such as R², Q², and number of components