Inference

Run predictions on new data using a trained model.

Once a model is ready, run inference to get predictions, clusters, anomaly scores, embeddings, or search results depending on the model type.

Infer (File Upload)

Upload a file and get results in the response. Best for small files.

Async Infer (File Upload)

Upload a file and poll for results. Best for larger files.

Async Infer (Dataset)

Run inference on an existing dataset. Returns a job ID for polling.

---

Output Formats

All inference endpoints support three output formats via the output_type parameter:

Format	Content-Type	Description
json	application/json	Column-oriented JSON dict. Each key maps to a list of values.
csv	text/csv	Standard CSV file with one row per input instance.
parquet	application/octet-stream	Apache Parquet file. Same columns as CSV. Includes model_id, model_name, model_version_id, and model_type in file-level metadata.

The specific columns returned depend on the model type:

---

Prediction

For binary classification, multiclass classification, and regression.

Column	Type	Present	Description
id	integer	Always	Row index from the input data.
prediction	string or float	Always	Predicted label (classification) or value (regression).
prediction_prob	float	Classification only	Confidence in the predicted class. Ranges from 0.5 to 1.0 for binary; 0 to 1.0 for multiclass. Not present for regression.

Binary classification:

JSON

{
  "id": [0, 1, 2],
  "prediction": ["yes", "no", "yes"],
  "prediction_prob": [0.88, 0.95, 0.7]
}

CSV

id,prediction,prediction_prob
0,yes,0.88
1,no,0.95
2,yes,0.7

Multiclass classification:

JSON

{
  "id": [0, 1, 2],
  "prediction": ["cat_A", "cat_B", "cat_C"],
  "prediction_prob": [0.82, 0.91, 0.65]
}

CSV

id,prediction,prediction_prob
0,cat_A,0.82
1,cat_B,0.91
2,cat_C,0.65

Regression:

JSON

{
  "id": [0, 1, 2],
  "prediction": [42.5, 17.3, 89.1]
}

CSV

id,prediction
0,42.5
1,17.3
2,89.1

---

Clustering

Column	Type	Description
id	integer	Row index from the input data.
cluster_label	integer	Cluster assignment (0-indexed).
cluster_description	string	Human-readable description of the cluster.

JSON

{
  "id": [0, 1, 2, 3],
  "cluster_label": [0, 1, 0, 2],
  "cluster_description": [
    "High-income urban professionals with frequent purchases",
    "Budget-conscious suburban families",
    "High-income urban professionals with frequent purchases",
    "Young students with occasional spending"
  ]
}

CSV

id,cluster_label,cluster_description
0,0,High-income urban professionals with frequent purchases
1,1,Budget-conscious suburban families
2,0,High-income urban professionals with frequent purchases
3,2,Young students with occasional spending

---

Anomaly Detection

Output depends on the anomaly_format parameter.

anomaly_format=ids_only (default)

Column	Type	Description
anomalous_ids	list of integers	Row indices flagged as anomalous.

JSON

{
  "anomalous_ids": [3, 17, 42]
}

CSV

anomalous_ids
3
17
42

anomaly_format=per_row

Column	Type	Description
id	integer	Row index from the input data.
is_anomaly	boolean	Whether this row is flagged as anomalous.
anomaly_score	float	Anomaly score (higher values indicate greater anomalousness).

JSON

{
  "id": [0, 1, 2, 3],
  "is_anomaly": [false, false, false, true],
  "anomaly_score": [0.021, 0.045, 0.033, 0.892]
}

CSV

id,is_anomaly,anomaly_score
0,false,0.021
1,false,0.045
2,false,0.033
3,true,0.892

---

Embeddings

Column	Type	Description
id	integer	Row index from the input data.
dim_0 … dim_N	float	Embedding vector dimensions. The number of dimensions depends on the model.

JSON

{
  "id": [0, 1, 2],
  "dim_0": [0.123, 0.456, -0.789],
  "dim_1": [-0.321, 0.654, 0.012],
  "dim_2": [0.111, -0.222, 0.333]
}

CSV

id,dim_0,dim_1,dim_2
0,0.123,-0.321,0.111
1,0.456,0.654,-0.222
2,-0.789,0.012,0.333

---

Search

Column	Type	Description
Input row ID (key)	string	Row index from the inference data.
Matched row ID (value)	string	Row index of the closest match in the training dataset.

JSON

{
  "search_results": {
    "0": "42",
    "1": "17",
    "2": "103"
  }
}

CSV

query_id,match_id
0,42
1,17
2,103

---

Factor Analysis

Output has one row per discovered factor, not per input instance.

Column	Type	Description
factor_id	integer	Factor index (0-indexed).
factor_description	string	Human-readable description of what this factor represents.
captured_variance	float	Proportion of variance in the data explained by this factor.

JSON

{
  "factor_id": [0, 1, 2],
  "factor_description": [
    "Income and spending power",
    "Geographic and urban/rural divide",
    "Age and career stage"
  ],
  "captured_variance": [0.35, 0.22, 0.15]
}

CSV

factor_id,factor_description,captured_variance
0,Income and spending power,0.35
1,Geographic and urban/rural divide,0.22
2,Age and career stage,0.15