[Feature] Add predict console command

[nxexox]

Problem

I don't always need to run a web application. Sometimes I want to use a model one-time in my pipelines. For example, using a bash command, to which you can give data for prediction.

Solution

Add a bash command predict, which will load the model into memory, and use it to make predictions on data from arguments.

I propose to support two use cases:

• Predict from a file - parse the file and forward it to the model line by line.
• Predict from command arguments - forward one or more lines of data directly to the arguments.

# Examples from file with data
mlup predict -m my_model.onnx -f [--file] path_to_file_with_data.csv -o [--output] path_to_result_file.csv

mlup predict -c path_to_mlup_conf.yaml -f [--file] path_to_file_with_data.csv -o [--output] path_to_result_file.csv

# Examples from bash command arguments with data
mlup predict -m my_model.onnx -r [--raw-string] "1,2,3,4,5,6,7,2,1" -o [--output] path_to_result_file.csv

mlup predict -c path_to_mlup_conf.yaml -r [--raw-string] "1,2,3,4,5,6,7,2,1" -o [--output] path_to_result_file.csv

# Examples with output to console
mlup predict -c path_to_mlup_conf.yaml -r [--raw-string] "1,2,3,4,5,6,7,2,1"
# Print predict result to stdout

Due to the fact that the prediction result can be displayed on the console, you need to learn how to enable/disable logging of loading and model operation. Because the result of the prediction from stdout can be used in the bash command chain. By default, output=stdout disables logging.

Use the -v flag:

• -v - only the result of the command.
• -vv - INFO logging of loading and operation.
• -vvv - DEBUG logging of loading and operation.

# Print only predict result
mlup predict -m my_model.onnx -f [--file] path_to_file_with_data.csv
# predict result

# Print INFO log
mlup predict -m my_model.onnx -f [--file] path_to_file_with_data.csv -vv

# Print DEBUG log
mlup predict -m my_model.onnx -f [--file] path_to_file_with_data.csv -vvv

It might be worth adding a 3rd block to the configuration file, which will contain:

• Settings for batching lines from a file for prediction. How many objects from the file can be sent to the model.
• File parser settings. Support csv, tsv, etc. Parsing settings for these formats.
• Settings for saving the model result.

version: 1
ml:
    ...
web:
    ...

console:
    parser: "mlup.console.parsers.csv.CSVParser"
    parser_kwargs: 
        separator: ","
    batching: 10
    ...