What's In Your Prediction.txt

#     Welcome to...
#
#      _______ _______ _______ _______ ______ _______ ___ ___
#     |    ___|    ___|   _   |_     _|   __ \_     _|   |   |
#     |    ___|    ___|       | |   | |      <_|   |_|-     -|
#     |___|   |_______|___|___| |___| |___|__|_______|___|___|
#
#                                 What's in *your* prediction?
#
# Every Featrix prediction returns is a JSON structure that includes:
# 
#   1. An ID to update Featrix on the ground truth if you discover the truth later.
#   2. An echo of your query to verify the query is what you expect.
#   3. Details on what Featrix actually processed and what, if anything, was ignored.
#   4. The actual prediction results (classification with probabilities or regression).
#   5. Safety information related to your query, based on the model's knowledge.
#   6. Metadata for fully reproducing the prediction and linking it to specific 
#      artifacts, whether they are Featrix artifacts or third party embedding models 
#      mixed with Featrix embeddings.

{
    prediction_featrix_id: "....",                  # call a feedback API later with
                                                    # the ground truth if you'd like
    prediction_time: "Thu Sep 26 18:17:52 2024",    # the time the query ran.
    
    original_query: {                               # your original input query
        numerical_input_1: 2,                       # dictionary of column: <value> pairs
    },

    actual_query: {                                 # if you specified a column not
        ...                                         # present in the model, your query may
    },                                              # be reduced before execution.
                                                    
    ignored_query_columns: [                        # for emphasis, we let you know
        ...                                         # which parts of the original query
    ],                                              # were ignored.
                                                    
    results: {                                      # your actual prediction: classifier
        cat_value_1: "0.25",                        # with probability or a regression.
        ...
    },

    column_guardrails: [                            # notes if any column in the query was
        { numerical_input_1: "within 1 stdev" },    # pushing the limit of the training data
        { numerical_input_2: ">3 stdevs" }, 
        ...        
    ], 
   
    cross_column_guardrails_score: [                # a cross-column score for how far out
        0 to 1                                      # of training the query was, when considered
    ],                                              # against all the columns queried.
   
    nearest_neighbors_in_training_data: [           # if security policy allows, we return
        { ...row 234 },                             # nearest neighbors in the training space
        { ...row 645 }, ...                         # for human evaluation of the result.
    ],
    
    upstream_embedding_models: {                    # metadata of upstream models used.
        col1:  "sentence-all-MiniLM-L12-v1",        # these might be other models you made 
        col6:  "featrix-upstream-52b43b29",         # with Featrix or embedding models
        col12: "openai-text-embedding-3-large"      # from third parties.
    }
    
    featrix_model: {                                # the neural function (model)
        model_featrix_id: "...",                    # the Featrix id for this model
        model_hash: "...",                          # the hash to identify a specific artifact
                                                    # in the Featrix system that ran this prediction
        model_train_time: "...",                    # when this model was last trained
        model_train_loops: 500,                     # how many passes were made over the data
        model_train_featrix_version: "1.2",         # the Featrix version that trained this model.
        model_predict_featrix_version: "1.2",       # the Featrix version that ran this model.
        model_metrics: {                            # metrics from the validation set on this model
            f1: 0.75                                # MSE, loss, accuracy, precision, recall, et al.
        }
    },
    
    featrix_embedding_space: {                      # the Featrix embedding space information
        embedding_space_featrix_id: "...",          # the Featrix id for this embedding space
        embedding_space_hash: "...",                # the hash to identify a specific artifact
        embedding_space_train_time: "...",          # when this model was last trained
        embedding_space_train_loops: 500,           # how many passes were made over the data
        embedding_space_featrix_version: "1.2",     # the Featrix version that trained this model.
        embedding_space_run_featrix_version: "1.2", # the Featrix version that ran this model.        
    }
}

# Learn more at www.featrix.ai