FastAPI: How to return a dictionary that includes numpy arrays?

Question

I get the following error when I try to access the 'data' variable from the endpoint '/'.

ValueError: [ValueError('dictionary update sequence element #0 has length 1; 2 is required'), TypeError('vars() argument must have __dict__ attribute')]

This is how FastAPI backend looks like:

from fastapi import FastAPI
app = FastAPI()
data = {}
@app.on_event("startup")
def startup_event():
    data[1]  =  [...] ...(numpy array)
    data[2]  = [...] ...(numpy array)
    return data


@app.get("/")
def home():
    return {'Data': data}

When I launch the endpoint I see 'Internal Server Error'. Nothing would display at the endpoint '/'. However, if I add this line -> 'print(data)' just above the return in home function for endpoint '/', it does print the values stored in the data dictionary, as specified in the startup function. How can I fix the issue, so that the data variable becomes visible when accessing the '/' endpoint?

I added three more lines of code to the above code snippet. Here, app = FastAPI() data = {} ... startup: ... function definition — catBuddy, Feb 25 '22 at 09:18
I just copy pasted your code and it seems to run fine for me, maybe there is some other part in your code that causes the problem? — Sandil Ranasinghe, Feb 25 '22 at 09:23
Really? Could you see the results when you visit endpoint '/'? There is only import statements apart from the code I wrote. — catBuddy, Feb 25 '22 at 09:34
Yeah. I get this `{"Data":{"1":1,"2":11}}` at the endpoint '/' . Do you have any more information in your error log? — Sandil Ranasinghe, Feb 25 '22 at 09:40
Yes, I figured. When I change the values to be numpy arrays then, they do not work. I get internal server error. — catBuddy, Feb 25 '22 at 11:00
FastAPI has no idea how it should decode numpy arrays as JSON. Return a native Python datatype or JSON yourself. Also, include the actual error message you get in your question as that makes it easier for other people to debug your problem. — MatsLindh, Feb 25 '22 at 11:03

Chris · Answer 1 · 2023-03-02T17:32:35.257

Option 1

You should convert (serialise) any numpy arrays into JSON before returning the data. Example:

data = {}
data[1] = json.dumps(np.array([1, 2, 3, 4]).tolist())
data[2] = json.dumps(np.array([5, 6, 7, 8]).tolist())
return data

Option 2

Alternatively, you could serialise the whole dictionary object and return it in a custom Response, such as below:

from fastapi import Response

json_data = json.dumps({k: v.tolist() for k, v in data.items()})
return Response(content=json_data, media_type="application/json")

Option 3

Another option would be to use jsonable_encoder—which is used internally by FastAPI when returning a value, in order to convert objects that are not serializable into str—and then, return a JSONResponse, or a custom Response directly (as demonstrated in the previous option), which will return an application/json encoded response to the client, as described here.

from fastapi.responses import JSONResponse
from fastapi.encoders import jsonable_encoder

json_data = jsonable_encoder({k: v.tolist() for k, v in data.items()})
return JSONResponse(content=json_data)

score 0 · Answer 2 · answered Feb 25 '22 at 11:46

FastAPI has no idea how it should decode numpy arrays to JSON; instead, do it explicitly yourself and then return the structure as native Python datatypes or as a raw JSON string. You can do this by using the custom Response objects in FastAPI.

return Response(content=numpy_as_json, media_type="application/json")

Or if you have it formatted as native datatypes:

return JSONResponse(content=numpy_as_native_datatypes)

FastAPI: How to return a dictionary that includes numpy arrays?

2 Answers2

Option 1

Option 2

Option 3

Linked