sentence-transformers How to use lfsolis/my-finetuned-sentence-bert with sentence-transformers:
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("lfsolis/my-finetuned-sentence-bert")
sentences = [
"Auto-WEKA: Combined Selection and Hyperparameter Optimization of\n Classification Algorithms",
" It has been a long time, since data mining technologies have made their ways\nto the field of data management. Classification is one of the most important\ndata mining tasks for label prediction, categorization of objects into groups,\nadvertisement and data management. In this paper, we focus on the standard\nclassification problem which is predicting unknown labels in Euclidean space.\nMost efforts in Machine Learning communities are devoted to methods that use\nprobabilistic algorithms which are heavy on Calculus and Linear Algebra. Most\nof these techniques have scalability issues for big data, and are hardly\nparallelizable if they are to maintain their high accuracies in their standard\nform. Sampling is a new direction for improving scalability, using many small\nparallel classifiers. In this paper, rather than conventional sampling methods,\nwe focus on a discrete classification algorithm with O(n) expected running\ntime. Our approach performs a similar task as sampling methods. However, we use\ncolumn-wise sampling of data, rather than the row-wise sampling used in the\nliterature. In either case, our algorithm is completely deterministic. Our\nalgorithm, proposes a way of combining 2D convex hulls in order to achieve high\nclassification accuracy as well as scalability in the same time. First, we\nthoroughly describe and prove our O(n) algorithm for finding the convex hull of\na point set in 2D. Then, we show with experiments our classifier model built\nbased on this idea is very competitive compared with existing sophisticated\nclassification algorithms included in commercial statistical applications such\nas MATLAB.\n",
" Many different machine learning algorithms exist; taking into account each\nalgorithm's hyperparameters, there is a staggeringly large number of possible\nalternatives overall. We consider the problem of simultaneously selecting a\nlearning algorithm and setting its hyperparameters, going beyond previous work\nthat addresses these issues in isolation. We show that this problem can be\naddressed by a fully automated approach, leveraging recent innovations in\nBayesian optimization. Specifically, we consider a wide range of feature\nselection techniques (combining 3 search and 8 evaluator methods) and all\nclassification approaches implemented in WEKA, spanning 2 ensemble methods, 10\nmeta-methods, 27 base classifiers, and hyperparameter settings for each\nclassifier. On each of 21 popular datasets from the UCI repository, the KDD Cup\n09, variants of the MNIST dataset and CIFAR-10, we show classification\nperformance often much better than using standard selection/hyperparameter\noptimization methods. We hope that our approach will help non-expert users to\nmore effectively identify machine learning algorithms and hyperparameter\nsettings appropriate to their applications, and hence to achieve improved\nperformance.\n",
" Nonnegative matrix factorization (NMF) has become a ubiquitous tool for data\nanalysis. An important variant is the sparse NMF problem which arises when we\nexplicitly require the learnt features to be sparse. A natural measure of\nsparsity is the L$_0$ norm, however its optimization is NP-hard. Mixed norms,\nsuch as L$_1$/L$_2$ measure, have been shown to model sparsity robustly, based\non intuitive attributes that such measures need to satisfy. This is in contrast\nto computationally cheaper alternatives such as the plain L$_1$ norm. However,\npresent algorithms designed for optimizing the mixed norm L$_1$/L$_2$ are slow\nand other formulations for sparse NMF have been proposed such as those based on\nL$_1$ and L$_0$ norms. Our proposed algorithm allows us to solve the mixed norm\nsparsity constraints while not sacrificing computation time. We present\nexperimental evidence on real-world datasets that shows our new algorithm\nperforms an order of magnitude faster compared to the current state-of-the-art\nsolvers optimizing the mixed norm and is suitable for large-scale datasets.\n"
]
embeddings = model.encode(sentences)
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [4, 4] 
