learning-orchestra-client 0.6.0

Learning Orchestra client for Python

learningOrchestra client package

Installation

Ensure which you have the python 3 installed in your machine and run:

pip install learning_orchestra_cliet

Documentation

After downloading the package, import all classes:

from learning_orchestra_client import *

create a Context object passing a ip from your cluster in constructor parameter:

cluster_ip = "34.95.222.197"
Context(cluster_ip)

After create a Context object, you will able to usage learningOrchestra, each learningOrchestra functionality is contained in your own class, therefore, to use a specific functionality, after you instantiate and configure Context class, you need instantiate and call the method class of interest, in below, there are all class and each class methods, also have an example of workflow using this package in a python code.

DatabaseApi

read_resume_files

read_resume_files(pretty_response=True)

Read all metadata files in learningOrchestra

• pretty_response: return indented string to visualization (default True, if False, return dict)

read_file

read_file(filename_key, skip=0, limit=10, query={}, pretty_response=True)

• filename_ley : filename of file
• skip: number of rows amount to skip in pagination (default 0)
• limit: number of rows to return in pagination (default 10) (max setted in 20 rows per request)
• query: query to make in mongo (default empty query)
• pretty_response: return indented string to visualization (default True, if False, return dict)

create_file

create_file(filename, url, pretty_response=True)

• filename: filename of file to be created
• url: url to csv file
• pretty_response: return indented string to visualization (default True, if False, return dict)

delete_file

delete_file(filename, pretty_response=True)

• filename: file filename to be deleted
• pretty_response: return indented string to visualization (default True, if False, return dict)

Projection

create_projection

create_projection(filename, projection_filename, fields, pretty_response=True)

• filename: filename of file to make projection
• projection_filename: filename used to create projection
• fields: list with fields to make projection
• pretty_response: return indented string to visualization (default True, if False, return dict)

DataTypeHandler

change_file_type

change_file_type(filename, fields_dict, pretty_response=True)

• filename: filename of file
• fields_dict: dictionary with "field": "number" or field: "string" keys
• pretty_response: return indented string to visualization (default True, if False, return dict)

Histogram

create_histogram

create_histogram(filename, histogram_filename, fields, 
                 pretty_response=True)

• filename: filename of file to make histogram
• histogram_filename: filename used to create histogram
• fields: list with fields to make histogram
• pretty_response: return indented string to visualization (default True, if False, return dict)

Tsne

create_image_plot

create_image_plot(tsne_filename, parent_filename,
                  label_name=None, pretty_response=True)

• parent_filename: filename of file to make histogram
• tsne_filename: filename used to create image plot
• label_name: label name to dataset with labeled tuples (default None, to datasets without labeled tuples)
• pretty_response: return indented string to visualization (default True, if False, return dict)

read_image_plot_filenames

read_image_plot_filenames(pretty_response=True)

• pretty_response: return indented string to visualization (default True, if False, return dict)

read_image_plot

read_image_plot(tsne_filename, pretty_response=True)

• tsne_filename: filename of a created image plot
• pretty_response: return indented string to visualization (default True, if False, return dict)

delete_image_plot

delete_image_plot(tsne_filename, pretty_response=True)

• tsne_filename: filename of a created image plot
• pretty_response: return indented string to visualization (default True, if False, return dict)

Pca

create_image_plot

create_image_plot(tsne_filename, parent_filename,
                  label_name=None, pretty_response=True)

• parent_filename: filename of file to make histogram
• pca_filename: filename used to create image plot
• label_name: label name to dataset with labeled tuples (default None, to datasets without labeled tuples)
• pretty_response: return indented string to visualization (default True, if False, return dict)

read_image_plot_filenames

read_image_plot_filenames(pretty_response=True)

• pretty_response: return indented string to visualization (default True, if False, return dict)

read_image_plot

read_image_plot(pca_filename, pretty_response=True)

• pca_filename: filename of a created image plot
• pretty_response: return indented string to visualization (default True, if False, return dict)

delete_image_plot

delete_image_plot(pca_filename, pretty_response=True)

• pca_filename: filename of a created image plot
• pretty_response: return indented string to visualization (default True, if False, return dict)

ModelBuilder

create_model

create_model(training_filename, test_filename, preprocessor_code, 
             model_classificator, pretty_response=True)

• training_filename: filename to be used in training
• test_filename: filename to be used in test
• preprocessor_code: python3 code for pyspark preprocessing model
• model_classificator: list of initial from classificators to be used in model
• pretty_response: return indented string to visualization (default True, if False, return dict)

model_classificator

• "lr": LogisticRegression
• "dt": DecisionTreeClassifier
• "rf": RandomForestClassifier
• "gb": Gradient-boosted tree classifier
• "nb": NaiveBayes

to send a request with LogisticRegression and NaiveBayes classificators:

create_model(training_filename, test_filename, preprocessor_code, ["lr", "nb"])

preprocessor_code environment

The python 3 preprocessing code must use the environment instances in bellow:

• training_df (Instantiated): Spark Dataframe instance for training filename
• testing_df (Instantiated): Spark Dataframe instance for testing filename

The preprocessing code must instantiate the variables in bellow, all instances must be transformed by pyspark VectorAssembler:

• features_training (Not Instantiated): Spark Dataframe instance for train the model
• features_evaluation (Not Instantiated): Spark Dataframe instance for evaluate trained model accuracy
• features_testing (Not Instantiated): Spark Dataframe instance for test the model

Case you don't want evaluate the model prediction, define features_evaluation as None.

Handy methods

self.fields_from_dataframe(dataframe, is_string)

• dataframe: dataframe instance
• is_string: Boolean parameter, if True, the method return the string dataframe fields, otherwise, return the numbers dataframe fields.

learning_orchestra_client usage example

In below there is a python script using the package with titanic challengue datasets:

from learning_orchestra_client import *

cluster_ip = "34.95.187.26"

Context(cluster_ip)

database_api = DatabaseApi()

print(database_api.create_file(
    "titanic_training",
    "https://filebin.net/rpfdy8clm5984a4c/titanic_training.csv?t=gcnjz1yo"))
print(database_api.create_file(
    "titanic_testing",
    "https://filebin.net/mguee52ke97k0x9h/titanic_testing.csv?t=ub4nc1rc"))

print(database_api.read_resume_files())


projection = Projection()
non_required_columns = ["Name", "Ticket", "Cabin",
                        "Embarked", "Sex", "Initial"]
print(projection.create("titanic_training",
                        "titanic_training_projection",
                        non_required_columns))
print(projection.create("titanic_testing",
                        "titanic_testing_projection",
                        non_required_columns))


data_type_handler = DataTypeHandler()
type_fields = {
    "Age": "number",
    "Fare": "number",
    "Parch": "number",
    "PassengerId": "number",
    "Pclass": "number",
    "SibSp": "number"
}

print(data_type_handler.change_file_type(
    "titanic_testing_projection",
    type_fields))

type_fields["Survived"] = "number"

print(data_type_handler.change_file_type(
    "titanic_training_projection",
    type_fields))


preprocessing_code = '''
from pyspark.ml import Pipeline
from pyspark.sql.functions import (
    mean, col, split,
    regexp_extract, when, lit)

from pyspark.ml.feature import (
    VectorAssembler,
    StringIndexer
)

TRAINING_DF_INDEX = 0
TESTING_DF_INDEX = 1

training_df = training_df.withColumnRenamed('Survived', 'label')
testing_df = testing_df.withColumn('label', lit(0))
datasets_list = [training_df, testing_df]

for index, dataset in enumerate(datasets_list):
    dataset = dataset.withColumn(
        "Initial",
        regexp_extract(col("Name"), "([A-Za-z]+)\.", 1))
    datasets_list[index] = dataset


misspelled_initials = ['Mlle', 'Mme', 'Ms', 'Dr', 'Major', 'Lady', 'Countess',
                       'Jonkheer', 'Col', 'Rev', 'Capt', 'Sir', 'Don']
correct_initials = ['Miss', 'Miss', 'Miss', 'Mr', 'Mr', 'Mrs', 'Mrs',
                    'Other', 'Other', 'Other', 'Mr', 'Mr', 'Mr']
for index, dataset in enumerate(datasets_list):
    dataset = dataset.replace(misspelled_initials, correct_initials)
    datasets_list[index] = dataset


initials_age = {"Miss": 22,
                "Other": 46,
                "Master": 5,
                "Mr": 33,
                "Mrs": 36}
for index, dataset in enumerate(datasets_list):
    for initial, initial_age in initials_age.items():
        dataset = dataset.withColumn(
            "Age",
            when((dataset["Initial"] == initial) &
                 (dataset["Age"].isNull()), initial_age).otherwise(
                    dataset["Age"]))
        datasets_list[index] = dataset


for index, dataset in enumerate(datasets_list):
    dataset = dataset.na.fill({"Embarked": 'S'})
    datasets_list[index] = dataset


for index, dataset in enumerate(datasets_list):
    dataset = dataset.withColumn("Family_Size", col('SibSp')+col('Parch'))
    dataset = dataset.withColumn('Alone', lit(0))
    dataset = dataset.withColumn(
        "Alone",
        when(dataset["Family_Size"] == 0, 1).otherwise(dataset["Alone"]))
    datasets_list[index] = dataset


text_fields = ["Sex", "Embarked", "Initial"]
for column in text_fields:
    for index, dataset in enumerate(datasets_list):
        dataset = StringIndexer(
            inputCol=column, outputCol=column+"_index").\
                fit(dataset).\
                transform(dataset)
        datasets_list[index] = dataset


training_df = datasets_list[TRAINING_DF_INDEX]
testing_df = datasets_list[TESTING_DF_INDEX]

assembler = VectorAssembler(
    inputCols=training_df.columns[1:],
    outputCol="features")
assembler.setHandleInvalid('skip')

features_training = assembler.transform(training_df)
(features_training, features_evaluation) =\
    features_training.randomSplit([0.1, 0.9], seed=11)
features_testing = assembler.transform(testing_df)
'''

model_builder = Model()

print(model_builder.create_model(
    "titanic_training_projection",
    "titanic_testing_projection",
    preprocessing_code,
    ["lr", "dt", "gb", "rf", "nb"]))