Chapter 2 - Compression Techniques “I have made this longer than usual because I have not had time to make it shorter.” Blaise Pascal In the last chapter, we discussed a few ideas to improve the deep deep learning efficiency. Now, we will elaborate on one of those ideas, the compression techniques. Compression techniques aim to reduce the model footprint (size, latency, memory etc.). We can reduce the chapter, we introduce Quantization, a model compression technique that addresses both these issues. We’ll start with a gentle introduction to the idea of compression. Details of quantization and its applications

Advanced Compression Techniques “The problem is that we attempt to solve the simplest questions cleverly, thereby rendering them unusually complex. One should seek the simple solution.” — Anton Pavlovich Pavlovich Chekhov In this chapter, we will discuss two advanced compression techniques. By ‘advanced’ we mean that these techniques are slightly more involved than quantization (as discussed in the second of our models. Did we get you excited yet? Let’s learn about these techniques together! Model Compression Using Sparsity Sparsity or Pruning refers to the technique of removing (pruning) weights during

Minimum Version Notes PyTables 3.6.1 HDF5-based reading / writing blosc 1.20.1 Compression for HDF5 zlib Compression for HDF5 fastparquet 0.4.0 Parquet reading / writing pyarrow 1.0.1 Parquet, ORC Clipboard I/O on linux Compression Dependency Minimum Version Notes brotli 0.7.0 Brotli compression python-snappy 0.6.0 Snappy compression Zstandard 0.15.2 Zstandard compression 1.4.2 Package overview TextFileReader object for iteration. See iterating and chunking below. Quoting, compression, and file format compression [{'infer', 'gzip', 'bz2', 'zip', 'xz', 'zstd', None, dict}, default 'infer'] For

Minimum Version Notes PyTables 3.6.1 HDF5-based reading / writing blosc 1.20.1 Compression for HDF5 zlib Compression for HDF5 fastparquet 0.4.0 Parquet reading / writing pyarrow 1.0.1 Parquet, ORC Clipboard I/O on linux Compression Dependency Minimum Version Notes brotli 0.7.0 Brotli compression python-snappy 0.6.0 Snappy compression Zstandard 0.15.2 Zstandard compression 1.4.2 Package overview TextFileReader object for iteration. See iterating and chunking below. Quoting, compression, and file format compression [{'infer', 'gzip', 'bz2', 'zip', 'xz', 'zstd', None, dict}, default 'infer'] For

Minimum Version Notes PyTables 3.6.1 HDF5-based reading / writing blosc 1.21.0 Compression for HDF5 zlib Compression for HDF5 fastparquet 0.4.0 Parquet reading / writing pyarrow 1.0.1 Parquet, ORC Clipboard I/O on linux Compression Dependency Minimum Version Notes brotli 0.7.0 Brotli compression python-snappy 0.6.0 Snappy compression Zstandard 0.15.2 Zstandard compression 1.4.2 Package overview TextFileReader object for iteration. See iterating and chunking below. Quoting, compression, and file format compression [{'infer', 'gzip', 'bz2', 'zip', 'xz', 'zstd', None, dict}, default 'infer'] For

efficiency in deep learning models. We will also introduce core areas of efficiency techniques (compression techniques, learning techniques, automation, efficient models & layers, infrastructure). Our hope where there might not be a single algorithm that works perfectly, and there is a large amount of unseen data that the algorithm needs to process. Unlike traditional algorithm problems where we expect exact leeway in model quality, we can trade off some of it for a smaller footprint by using lossy model compression techniques7. For example, when compressing a model naively we might reduce the model size, RAM

ValueError (GH10384). • Enable reading gzip compressed files via URL, either by explicitly setting the compression parameter or by inferring from the presence of the HTTP Content-Encoding header in the response required for compressed files in Python 2.) (GH11070, GH11073) • pd.read_csv is now able to infer compression type for files read from AWS S3 storage (GH11070, GH11074). 1.1. v0.17.0 (October 9, 2015) 15 (GH9777) • By default, read_csv and read_table will now try to infer the compression type based on the file exten- sion. Set compression=None to restore the previous behavior (no decompression). (GH9770) Deprecations

order as Series.order; add na_position arg to conform to Series.order (GH6847) • default sorting algorithm for Series.order is now quicksort, to conform with Series.sort (and numpy defaults) • add inplace is a way of visualizing multi-variate data. It is based on a simple spring tension minimization algorithm. Basically you set up a bunch of points in a plane. In our case they are equally spaced on a unit Parse whitespace-delimited (spaces or tabs) file (much faster than using a regular expression) • compression: decompress ’gzip’ and ’bz2’ formats on the fly. • dialect: string or csv.Dialect instance to

is a way of visualizing multi-variate data. It is based on a simple spring tension minimization algorithm. Basically you set up a bunch of points in a plane. In our case they are equally spaced on a unit Parse whitespace-delimited (spaces or tabs) file (much faster than using a regular expression) • compression: decompress ’gzip’ and ’bz2’ formats on the fly. • dialect: string or csv.Dialect instance to (see below). 19.8.14 Compression PyTables allows the stored data to be compressed. Tthis applies to all kinds of stores, not just tables. • Pass complevel=int for a compression level (1-9, with 0 being

is a way of visualizing multi-variate data. It is based on a simple spring tension minimization algorithm. Basically you set up a bunch of points in a plane. In our case they are equally spaced on a unit Parse whitespace-delimited (spaces or tabs) file (much faster than using a regular expression) • compression: decompress ’gzip’ and ’bz2’ formats on the fly. • dialect: string or csv.Dialect instance to (see below). 18.7.14 Compression PyTables allows the stored data to be compressed. Tthis applies to all kinds of stores, not just tables. • Pass complevel=int for a compression level (1-9, with 0 being