Understanding inputs for google ai platform custom prediction routines

Question

I am following this documentation on custom prediction routines and I am trying to understand how the inputs for custom prediction routine looks like. The code to send the input looks like this:

instances = [
        [6.7, 3.1, 4.7, 1.5],
        [4.6, 3.1, 1.5, 0.2],
    ]
service = discovery.build('ml', 'v1')
name = 'projects/{}/models/{}'.format(project, model)

if version is not None:
    name += '/versions/{}'.format(version)

response = service.projects().predict(
    name=name,
    body={'instances': instances}
).execute()

and the Predictor.py at the moment is very simple. I am just trying to understand how the input looks like...

class Predictor(object):
    """An example Predictor for an AI Platform custom prediction routine."""

    def __init__(self, model):
        self._model = model

    def predict(self, instances, **kwargs):

        inputs = np.asarray(instances)
        if kwargs.get('max'):
            return np.argmax(inputs, axis=1)

        return np.sum(inputs)


    @classmethod
    def from_path(cls, model_dir):
        return cls(None)

But when I try to get the response i get the following error:

{
  "error": "Prediction failed: unknown error."
}

Furthermore it is extremely difficult to debug the code, because there is no way to step into the code or print logs... I have no idea what's going on... How the input looks like? how should i access them? This is just a simple test, but eventually I want to send images, it will be even more difficult to debug then. How will I receive them? How will I preprocess them in the preprocessor? Let's assume that the proporcessing i have done at training time looks like this

data = cv2.imread(str(img_path))
data = cv2.resize(data, (224, 224))
data = cv2.cvtColor(data, cv2.COLOR_BGR2RGB)
x = data.astype(np.float32) / 255.
return np.expand_dims(x, axis=0)

How the instances object looks like so i can construct the preprocessor accordingly? thank you in advance.

Answer 1

I'm builiding a new sample for Custom Prediction which may be useful for your to debug: First I write file locally via a Notebook (Colab)

%%writefile model_prediction.py

import numpy as np
import os
import pickle
import pandas as pd
import importlib

class CustomModelPrediction(object):
    _UNUSED_COLUMNS = ['fnlwgt', 'education', 'gender']
    _CSV_COLUMNS = [
        'age', 'workclass', 'fnlwgt', 'education', 'education_num',
        'marital_status', 'occupation', 'relationship', 'race', 'gender',
        'capital_gain', 'capital_loss', 'hours_per_week', 'native_country',
        'income_bracket'
    ]
    _CATEGORICAL_TYPES = {
        'workclass': pd.api.types.CategoricalDtype(categories=[
            'Federal-gov', 'Local-gov', 'Never-worked', 'Private',
            'Self-emp-inc',
            'Self-emp-not-inc', 'State-gov', 'Without-pay'
        ]),
        'marital_status': pd.api.types.CategoricalDtype(categories=[
            'Divorced', 'Married-AF-spouse', 'Married-civ-spouse',
            'Married-spouse-absent', 'Never-married', 'Separated', 'Widowed'
        ]),
        'occupation': pd.api.types.CategoricalDtype([
            'Adm-clerical', 'Armed-Forces', 'Craft-repair',
            'Exec-managerial',
            'Farming-fishing', 'Handlers-cleaners', 'Machine-op-inspct',
            'Other-service', 'Priv-house-serv', 'Prof-specialty',
            'Protective-serv',
            'Sales', 'Tech-support', 'Transport-moving'
        ]),
        'relationship': pd.api.types.CategoricalDtype(categories=[
            'Husband', 'Not-in-family', 'Other-relative', 'Own-child',
            'Unmarried',
            'Wife'
        ]),
        'race': pd.api.types.CategoricalDtype(categories=[
            'Amer-Indian-Eskimo', 'Asian-Pac-Islander', 'Black', 'Other',
            'White'
        ]),
        'native_country': pd.api.types.CategoricalDtype(categories=[
            'Cambodia', 'Canada', 'China', 'Columbia', 'Cuba',
            'Dominican-Republic',
            'Ecuador', 'El-Salvador', 'England', 'France', 'Germany',
            'Greece',
            'Guatemala', 'Haiti', 'Holand-Netherlands', 'Honduras', 'Hong',
            'Hungary',
            'India', 'Iran', 'Ireland', 'Italy', 'Jamaica', 'Japan', 'Laos',
            'Mexico',
            'Nicaragua', 'Outlying-US(Guam-USVI-etc)', 'Peru',
            'Philippines', 'Poland',
            'Portugal', 'Puerto-Rico', 'Scotland', 'South', 'Taiwan',
            'Thailand',
            'Trinadad&Tobago', 'United-States', 'Vietnam', 'Yugoslavia'
        ])
    }

    def __init__(self, model, processor):
        self._model = model
        self._processor = processor
        self._class_names = ['<=50K', '>50K']

    def _preprocess(self, instances):
        """Dataframe contains both numeric and categorical features, convert
        categorical features to numeric.

        Args:
          dataframe: A `Pandas.Dataframe` to process.
        """
        dataframe = pd.DataFrame(data=[instances], columns=self._CSV_COLUMNS[:-1])
        dataframe = dataframe.drop(columns=self._UNUSED_COLUMNS)
        # Convert integer valued (numeric) columns to floating point
        numeric_columns = dataframe.select_dtypes(['int64']).columns
        dataframe[numeric_columns] = dataframe[numeric_columns].astype(
            'float32')

        # Convert categorical columns to numeric
        cat_columns = dataframe.select_dtypes(['object']).columns
        # Keep categorical columns always using same values based on dict.
        dataframe[cat_columns] = dataframe[cat_columns].apply(
            lambda x: x.astype(self._CATEGORICAL_TYPES[x.name]))
        dataframe[cat_columns] = dataframe[cat_columns].apply(
            lambda x: x.cat.codes)
        return dataframe

    def predict(self, instances, **kwargs):
        preprocessed_data = self._preprocess(instances)
        preprocessed_inputs = self._processor.preprocess(preprocessed_data)
        outputs = self._model.predict_classes(preprocessed_inputs)
        if kwargs.get('probabilities'):
            return outputs.tolist()
        else:
            return [self._class_names[index] for index in
                    np.argmax(outputs, axis=1)]

    @classmethod
    def from_path(cls, model_dir):
        import tensorflow as tf
        model_path = os.path.join(model_dir, 'model.h5')
        model = tf.keras.models.load_model(model_path)

        preprocessor_path = os.path.join(model_dir, 'preprocessor.pkl')
        with open(preprocessor_path, 'rb') as f:
            preprocessor = pickle.load(f)

        return cls(model, preprocessor)

Once file is written I can test it like this locally before deploying the model:

from model_prediction import CustomModelPrediction
model = CustomModelPrediction.from_path('.')
instance = [25, 'Private', 226802, '11th', 7, 'Never-married', 'Machine-op-inspct', 'Own-child', 'Black', 'Male', 0, 0, 40, 'United-States']
model.predict(instance)

Other option is once you build the setup package you can also test installation locally where my_custom_code-0.1.tar.gz is the file intended to be deployed in AI Platform:

 pip install --target=/tmp/custom_lib --no-cache-dir -b /tmp/pip_builds my_custom_code-0.1.tar.gz

Also take a look at this section:

You can use the --enable-console-logging to get logs exported to your project. You may need to create a new Model.

Answer 2

It looks like that using debug code (at the time of this post) without a model do not work. I used the following code to have everything worked for my image prediction use case:

image_filename = 'your image path'
PROJECT_ID = ''
MODEL_NAME = ''
VERSION_NAME = ''

img = base64.b64encode(open(image_filename, "rb").read()).decode()
image_bite_dict = {"key": "0", "image_bytes": {"b64": img}}

instances = [
            image_bite_dict
        ]


service = googleapiclient.discovery.build('ml', 'v1')
    name = 'projects/{}/models/{}/versions/{}'.format(PROJECT_ID, MODEL_NAME, VERSION_NAME)
response = service.projects().predict(
        name=name,
        body={'instances': instances}
    ).execute()