How can I upload my pipeline model to edgeimpulse project?

Help
#1

Question/Issue:
Hello, I’m trying to upload simple model to edge impulse directly. But It can not be uploaded by direct model upload menu from my project. How could I upload my model?
The code below is what I am trying.

Project ID:
(onnx_upload_test - Pretrained model - Edge Impulse

Context/Use case:

import numpy as np
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC
from skl2onnx import convert_sklearn
from skl2onnx.common.data_types import FloatTensorType

Sample data for training

X_train = np.array([[1, 2, 3, 4, 5], [2, 3, 4, 5, 6], [3, 4, 5, 1, 2], [4, 5, 6, 7, 8], [5, 6, 7, 8, 9]])
y_train = np.array([0, 1, 2, 0, 1])

Create the pipeline

pipeline = Pipeline([
(‘scaler’, StandardScaler()), # StandardScaler
(‘svc’, SVC()) # SVC with probability
])

Fit the model

pipeline.fit(X_train, y_train)

evaluation

pipeline.fit(X_train, y_train)
X_test = np.array([[1, 2, 3, 4, 5], [3, 4, 5, 1, 2], [5, 6, 7, 8, 9], [4, 5, 6, 7, 8]])
y_test = np.array([0, 2, 1, 0])
print((X_test))
print((pipeline.named_steps[‘scaler’].transform(X_test)))
print(pipeline.predict(X_test))
score = pipeline.score(X_test, y_test)
print(“pipeline score:”, score)

Convert the pipeline to ONNX

Set the input shape to a fixed size (e.g., [1, 5] for batch size of 1)

initial_type = [(‘float_input’, FloatTensorType([1, 5]))]
final_types = [(‘output_label’, FloatTensorType([1, 1])), (‘output_probability’, FloatTensorType([1, 1]))]

onnx_model = convert_sklearn(pipeline, initial_types=initial_type, final_types= final_types)

Save the ONNX model

with open(‘svc_pipeline.onnx’, ‘wb’) as f:
f.write(onnx_model.SerializeToString())

Steps Taken:

  1. making pipeline model
  2. making onnx file from the code
  3. upload onnx file to the project

Expected Outcome:
simply upload

Actual Outcome:
Creating job… OK (ID: 25546624)

Scheduling job in cluster…
Container image pulled!
Job started
Converting ONNX model…
Scheduling job in cluster…
Job started
INFO: No representative features passed in, won’t quantize this model

Trying conversion using onnx2tf…
Grabbing input information…
No input shape conversion needed
ERROR: Scaler OP is not yet implemented.
Grabbing input information OK

Conversion using onnx2tf failed
Trying conversion using onnx-tf…
Traceback (most recent call last):
File “convert-via-tf-onnx.py”, line 68, in
ei_tensorflow.model_conversion.onnx.onnx_to_tflite(args.onnx_file, file_float32, file_int8,
File “/app/./resources/libraries/ei_tensorflow/model_conversion/onnx.py”, line 77, in onnx_to_tflite
onnx.checker.check_model(onnx_graph)
File “/app/.venv-onnx-tf/lib/python3.8/site-packages/onnx/checker.py”, line 106, in check_model
C.check_model(protobuf_string)
onnx.onnx_cpp2py_export.checker.ValidationError: Your model ir_version is higher than the checker’s.

Conversion using onnx-tf also failed, cannot use this ONNX file - contact your solutions engineer, or post the logs on the forum.
Application exited with code 1

Converting ONNX model failed, see above
Job failed (see above)

Reproducibility:

  • [O] Always
  • [ ] Sometimes
  • [ ] Rarely

Environment:

  • Just website

Additional Information:
Nothing

#2

Hi @ecekim - the short answer is that arbitrary Scikit-learn pipelines are not currently supported.

There is a specific subset of classical ML models that are supported, which can be found in the Supported classical ML algorithms section of our Classical ML doc.

1 Like