Getting Started
Nexosis Concepts
Forecasting
Anomaly Detection
Security
Troubleshooting
Before the Nexosis API can do anything really useful, you’re going to need to send it some data that it can use for Forecasts or Impact Analysis.
Watch the video (19:25)
Datasets
You can send data to the Nexosis API through the Data endpoint.
Data can be sent to the Nexosis API as either JSON or CSV, with optional metadata
Examples
In this example we’ll create a Named DataSet called sales by sending some data to it.
`timestamp` is the date and time being observed and `values` are a dictionary of values observed at that time.
curl -v -X PUT "https://ml.nexosis.com/v1/data/sales" \
-H "Content-Type: application/json" \
-H "api-key: {subscription key}" \
--data-binary "@/path/to/file/data.json"
{
"data": [
{
"timestamp": "2017-05-25T00:00:00+00:00",
"sales": 1000.0,
"orders": 100
},
{
"timestamp": "2017-05-26T00:00:00+00:00",
"sales": 980.0,
"orders": 98
},
{
"timestamp": "2017-05-27T00:00:00+00:00",
"sales": 1100.0,
"orders": 110
},
{
"timestamp": "2017-05-28T00:00:00+00:00",
"sales": 1080.0,
"orders": 108
},
{
"timestamp": "2017-05-29T00:00:00+00:00",
"sales": 1110.0,
"orders": 111
}
]
}
TimeStamps
The API will assume a timestamp role column if your data contains a single date type column with unique values and you have not otherwise specified a key or timestamp role column in metadata.
If your csv data has a timestamp with a different column name, you can provide that name with a parameter in the query string named timestampColumn. For JSON data, simply specify the correct role in the column metadata.
curl -v -X PUT "https://ml.nexosis.com/v1/data/sales?timestampColumn=date" \
-H "Content-Type: text/csv" \
-H "api-key: {subscription key}" \
--data-binary "@/path/to/file/data.csv"
date,sales,orders
2017-05-25,1000,100
2017-05-26,980,98
2017-05-27,1100,110
2017-05-28,1080,108
2017-05-29,1110,111
Response
The response to the PUT will be an HTTP 200 with a response body that is a summary of the DataSet.
{
"dataSetName": "sales"
}
Updating Data
You can modify a keyed DataSet or a time-series DataSet with additional and/or updated data by issuing a PUT to the same DataSet name. If your data has no user provided key, nor a timestamp role column then new data will simply be appended to the DataSet.
So, issuing a PUT to the same sales DataSet above like so
curl -v -X PUT "https://ml.nexosis.com/v1/data/sales?timestampColumn=date" \
-H "Content-Type: text/csv" \
-H "api-key: {subscription key}" \
--data-binary "@/path/to/file/data.csv"
date,sales,orders
2017-05-29,1120,112
2017-05-30,1234,123
2017-05-31,1235,123
Will result in the sales DataSet containing the following records:
2017-05-25,1000,100
2017-05-26,980,98
2017-05-27,1100,110
2017-05-28,1080,108
2017-05-29,1120,112
2017-05-30,1234,123
2017-05-31,1235,123
Where the timeStamp that overlapped between the first PUT and the second PUT overwrote the value from the first, and the non-overlapping timeStamps were appended.
