Predictor
A module for making predictions on fabrication variations in photonic devices using machine learning models deployed in the cloud.
predict
predict(device, model_name, model_tags, binarize=False)
Generates a prediction for a photonic device using a specified cloud-based ML model.
The function sends an image of the device to a cloud function, which uses the specified machine learning model to generate a prediction.
Parameters
- Name
device- Type
- np.ndarray
- Description
A binary numpy matrix representing the shape of a device.
- Name
model_name- Type
- str
- Description
The name of the ML model to use for the prediction. Consult the module's documentation for available models.
- Name
model_tags- Type
- Union[str, List[str]]
- Description
The tags of the ML model. Consult the module's documentation for available tags.
- Name
binarize- Type
- bool, optional
- Description
If set to True, the prediction will be binarized (default is False).
Returns
- Name
prediction- Type
- np.ndarray
- Description
A numpy matrix representing the predicted shape of the device. Pixel values closer to 1 indicate a higher likelihood of core material, while pixel values closer to 0 suggest a higher likelihood of cladding material. Pixel values in between represent prediction uncertainty.
correct
correct(device, model_name, model_tags, binarize=False, multi_correct: bool=False)
Generates a correction for a photonic device using a specified cloud-based ML model.
The function sends an image of the device to a cloud function, which uses the specified machine learning model to generate a correction.
Parameters
- Name
device- Type
- np.ndarray
- Description
A binary numpy matrix representing the shape of a device.
- Name
model_name- Type
- str
- Description
The name of the ML model to use for the correction. Consult the module's documentation for available models.
- Name
model_tags- Type
- Union[str, List[str]]
- Description
The tags of the ML model. Consult the module's documentation for available tags.
- Name
binarize- Type
- bool, optional
- Description
If set to True, the correction will be binarized (default is False).
- Name
multi_correct- Type
- bool, optional
- Description
If set to True, the correction will be generated using a iterative approach (default is False).
Returns
- Name
correction- Type
- np.ndarray
- Description
A numpy matrix representing the corrected shape of the device. Pixel values closer to 1 indicate a higher likelihood of core material, while pixel values closer to 0 suggest a higher likelihood of cladding material. Pixel values in between represent correction uncertainty.