Importing
Import Predictions
Importing
Import Predictions
Encord Active does not only provide a streamlined method to curate your image data, Active also provides metrics and analytics to optimize your model’s performance. Simply upload your model’s predictions in to Active to start.
Your predictions must be imported to Active, before you can use the Predictions feature on the Explorer page and the Model Evaluation page.
STEP 1: Prepare Your Predictions for Import
Disclaimer: We strongly recommend that you are knowledgeable about the Encord SDK. If you are unfamiliar with the SDK or if you do not understand the following boilerplate code, refer to this topic in the SDK documentation. The script below is incomplete and requires you to add the
label_rows the model predictions you are importing. Always ensure you are using the latest version of the Encord SDK. Within Encord Active, predictions use the same format as labels. For the most part, creating predictions programmatically is the same as creating labels.
If you want to learn more about working with labels, the content is available here.
DO NOT SAVE the Label Row or your labels are overwritten by your predictions.
Store Predictions Boilerplate
# Import dependencies
import os
import json
from encord import EncordUserClient, Project
from encord.objects import LabelRowV2
# Authenticate client and identify project
ssh_private_key_path = os.getenv("ENCORD_CLIENT_SSH_PATH")
project_hash = os.getenv("ENCORD_PROJECT_HASH")
assert ssh_private_key_path is not None
assert project_hash is not None
client = EncordUserClient.create_with_ssh_private_key(ssh_private_key_path)
# Gets Project to add labels. This Project already exists in Encord.
project: Project = client.get_project(project_hash)
BATCH_SIZE = 100 # Batch size to split
label_rows = project.list_label_rows_v2()
# splits the label_rows into batches of size BATCH_SIZE
label_row_batches = [label_rows[i:i+BATCH_SIZE] for i in range(0, len(label_rows), BATCH_SIZE)]
serialized_output: list[dict] = []
for labels_batch in label_row_batches:
bundle_init = project.create_bundle()
for label_row in labels_batch:
label_row.initialise_labels(bundle=bundle_init,# ignore existing labels
include_object_feature_hashes=set(),
include_classification_feature_hashes=set(),)
# Execute bundle, initialising all the labels at once
bundle_init.execute()
# Create label_row from model predictions. For example, import predictions.
for label_row in labels_batch:
pass
'''
Load in model predictions for each label_row here, and replace "pass".
'''
# Add the label_row containing predictions as serialized json
serialized_output.append(label_row.to_encord_dict()) # Serialize
with open("predictions.json", "w") as f:
json.dump(serialized_output, f)
Before importing your
predictions.json file, we recommend that you compress your JSON file (basically removing the whitespace from the file). Compressing the file significantly reduces the file size by a factor of 10. You can use a tool like https://jqlang.github.io/jq/manual/ with the command jq -c '.' large_input.json > compressed_input.json where large_input.json is the name of your input file and compressed_input.json is the file you can use to import your predictions into Active.STEP 2: Import Predictions Set
Once you have the predictions.json file from STEP 1, Prediction Sets can be imported from both the Model Evaluation page and the Upload predictions button ( + ) on the Overview tab of the Predictions page in the Explorer page.
Encord Active supports importing Prediction Sets files of up to 100MB. If your Prediction Sets file exceeds 100MB, split the file into smaller files and then perform the import.
If you have any issues importing your predictions contact your CSM or contact us at support@encord.com.
Next Steps
Model and Prediction Validation
Was this page helpful?