cloud start: machine learning
we help you build a minimal viable machine learning model of a specific business case
During the machine learning Proof of Concept, we run through the necessary steps to build an initial machine learning model, as well as demonstrate the feasibility of the solution. This process involves identifying a hypothesis, analyzing a data set, determining the appropriate algorithm, segmenting and qualifying the data, and training the algorithm against the data.
Analyze available data sources to assess state of data and potential usefulness when applied in a machine learning model.
Analyze data characteristics
Assess data quality, cleanliness, potential correlation, and patterns
Check for class imbalance
Validate hypothesis relative to data
Research modeling strategies to determine appropriate machine learning algorithm to address business problem.
Research existing strategies and white papers
Select an algorithm based on hypothesis, type of features, patterns in data
Document decisions related to algorithms
data pipeline and feature engineering
Create machine learning model features based on raw data analysis and tests.
Use domain knowledge to identify features
Transform raw data into features
Define new features as needed, and remove redundant/duplicate features, as well as highly correlated features
develop initial ML model
Develop an initial machine learning model that uses the data to solve the business problem. In this phase we will use both the selected machine learning algorithms, and the features created in the prior phase.
Define the right modeling strategy and choose the right ML algorithms
Select data set for training, test set, and validation
Develop initial model
Determine duration, and amount of data for initial experiment
iterate to improve model performance
Refine initial machine learning model to improve performance. The final model should meet performance targets agreed upon by Google and the customer.
Evaluate and visualize model result
Determine corrective actions to improve model
Iterate and improve model results
Basic knowledge of data analytics in Google Cloud Platform
Data to analyze
High involvement of customer's business experts
Machine learning pipeline
The infrastructure that surrounds a machine learning algorithm. Includes gathering the data from the front end, putting it into training data files, training one or more models
Initial machine learning model(s)
A trained algorithm that addresses the selected business problem
A plan for the customer to use, on how to deploy the machine learning models in production
As machine learning enables any organization to create deeper insights from data, and even predict future events, the PoC accommodates a diversity of use cases. Some examples:
An e-commerce marketeer or sales manager wants to predict upsells, a government needs to improve water management, whereas a farmer wants to analyze and improve the quality of his food, and a marketing manager expects to maximize the return from advertisement spending.
5 to 10 days over the period of two months.
The amount of data engineering required
Iterations over the model
scope and pricing
Pricing will be agreed upon by customer and g-company