One of the main challenges in cloud computing is resource management, the ability to schedule workloads and services over the infrastructure in the most automated way. By optimizing cloud assignment and resource usage, energy can be saved, production incident can be anticipated and services QoS improved. With the recent years emergence of light virtualisation, known as containerization, the resource allocation problem was brought back, notably to support containers elasticity, hence the dynamic allocation of ressource at runtime at a single service scale. In this paper we show that using an hybrid loop system, which combines unsupervised learning and optimization techniques, our algorithm provides and iteratively improves scheduling solutions to containers resource assignment, enabling capacity planning over dynamic resource loads. Within our benchmarks, these solutions outperform state of the art algorithms, by an average of 6.3%, while providing more expressivity and control over input parameters. We describe also the implementation of this method, through an open source Python library called HOTS, which allows hybrid optimization for time series based use cases.