Skip to content

Commit f72d822

Browse files
author
Siyuan Cheng
committed
example: add har example to mli module
add har example to mli module Signed-off-by: Siyuan Cheng <[email protected]>
1 parent 8700dd8 commit f72d822

22 files changed

+22056
-0
lines changed

‎samples/modules/embarc_mli/example_har_smartphone/CMakeLists.txt

Lines changed: 41 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,41 @@
1+
# Copyright (c) 2022 Synopsys
2+
# SPDX-License-Identifier: Apache-2.0
3+
4+
cmake_minimum_required(VERSION 3.13.1)
5+
6+
find_package(Zephyr REQUIRED HINTS $ENV{ZEPHYR_BASE})
7+
project(hello_world)
8+
9+
zephyr_compile_definitions(
10+
MODEL_BIT_DEPTH=16
11+
V2DSP_XY
12+
_C_ARRAY_
13+
)
14+
15+
zephyr_compile_options("-Wno-ignored-qualifiers")
16+
17+
zephyr_library_include_directories(
18+
../auxiliary/
19+
src/
20+
)
21+
22+
target_sources(app PRIVATE
23+
src/ml_api_har_smartphone_main.c
24+
src/har_smartphone_coefficients_fx.c
25+
src/har_smartphone_coefficients_sa.c
26+
src/har_smartphone_constants.h
27+
src/har_smartphone_model_sa.c
28+
src/har_smartphone_model_fx.c
29+
src/har_smartphone_model.h
30+
src/har_smartphone_ref_inout.h
31+
../auxiliary/examples_aux.c
32+
../auxiliary/examples_aux.h
33+
../auxiliary/idx_file.c
34+
../auxiliary/idx_file.h
35+
../auxiliary/tensor_transform.c
36+
../auxiliary/tensor_transform.h
37+
../auxiliary/tests_aux.c
38+
../auxiliary/tests_aux.h
39+
src/small_test_base/labels.c
40+
src/small_test_base/tests.c
41+
)

‎samples/modules/embarc_mli/example_har_smartphone/README.rst

Lines changed: 113 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,113 @@
1+
.. _embarc_mli_example_har_smartphone:
2+
3+
STM Based Human Activity Recognition (HAR) Example
4+
##################################################
5+
6+
Overview
7+
********
8+
Example shows how to work with recurrent primitives (LSTM and basic RNN) implemented in embARC MLI Library. It is based on open source [GitHub project](https://github.com/guillaume-chevalier/LSTM-Human-Activity-Recognition) by Guillaume Chevalie. Chosen approach, complexity of the model and [dataset](https://archive.ics.uci.edu/ml/datasets/Human+Activity+Recognition+Using+Smartphones) are relevant to IoT domain. The model is intended to differentiate human activity between 6 classes based on inputs from embedded inertial sensors from waist-mounted smartphone. Classes:
9+
* 0: WALKING
10+
* 1: WALKING_UPSTAIRS
11+
* 2: WALKING_DOWNSTAIRS
12+
* 3: SITTING
13+
* 4: STANDING
14+
* 5: LAYING
15+
16+
Quick Start
17+
--------------
18+
19+
Example supports building with [Zephyr Software Development Kit (SDK)](https://docs.zephyrproject.org/latest/getting_started/installation_linux.html#zephyr-sdk) and running with MetaWare Debuger on [nSim simulator](https://www.synopsys.com/dw/ipdir.php?ds=sim_nSIM).
20+
21+
Add embarc_mli module to Zephyr instruction
22+
-------------------------------------------
23+
24+
1. Open command line and change working directory to './zephyrproject/zephyr'
25+
26+
2. Download embarc_mli version 2.0
27+
28+
west update
29+
30+
Build with Zephyr SDK toolchain
31+
-------------------------------
32+
33+
Build requirements:
34+
- Zephyr SDK toolchain version 0.13.2 or higher
35+
- gmake
36+
37+
1. Open command line and change working directory to './zephyrproject/zephyr/samples/modules/embarc_mli/example_har_smartphone'
38+
39+
2. Build example
40+
41+
west build -b nsim_em samples/modules/embarc_mli/example_har_smartphone
42+
43+
Run example
44+
--------------
45+
46+
1. Run example
47+
48+
west flash
49+
50+
Result Quality shall be "S/N=3579.6 (71.1 db)"
51+
52+
More options
53+
--------------
54+
55+
You can change mode in ml_api_har_smartphone_main.c to 1,2,3:
56+
57+
* mode=1:
58+
59+
Built-in input processing. Uses only hard-coded vector for the single input model inference.
60+
61+
* mode=2:
62+
63+
Unavailable right now due to hostlink error. External test-set processing. Reads vectors from input IDX file, passes it to the model, and writes it's output to the other IDX file (if input is *tests.idx* then output will be *tests.idx_out*).
64+
65+
* mode=3:
66+
67+
Accuracy measurement for testset. Reads vectors from input IDX file, passes it to the model, and accumulates number of successive classifications according to labels IDX file. If hostlink is unavailble, please add _C_ARRAY_ definition.
68+
69+
You can add different definitions to zephyr_compile_definitions() in 'zephyr/samples/modules/embarc_mli/example_har_smartphone/CMakeLists.txt' to implement numerous model:
70+
71+
* 16 bit depth of coefficients and data (default):
72+
73+
MODEL_BIT_DEPTH=16
74+
75+
* 8 bit depth of coefficients and data:
76+
77+
MODEL_BIT_DEPTH=8
78+
79+
* 8x16: 8 bit depth of coefficients and 16 bit depth of data:
80+
81+
MODEL_BIT_DEPTH=816
82+
83+
* If hostlink is not available, please reads vectors from input Array file, passes it to the model, and accumulates number of successive classifications according to labels array file:
84+
85+
_C_ARRAY_
86+
87+
Example Structure
88+
--------------------
89+
Structure of example application may be divided logically on three parts:
90+
91+
* **Application.** Implements Input/output data flow and it's processing by the other modules. Application includes:
92+
* ml_api_har_smartphone_main.c
93+
* ../auxiliary/examples_aux.h(.c)
94+
* **Inference Module.** Uses embARC MLI Library to process input according to pre-defined graph. All model related constants are pre-defined and model coefficients is declared in the separate compile unit
95+
* har_smartphone_model.h
96+
* har_smartphone_model.c
97+
* har_smartphone_constants.h
98+
* har_smartphone_coefficients.c
99+
* **Auxiliary code.** Various helper functions for measurements, IDX file IO, etc.
100+
* ../auxiliary/tensor_transform.h(.c)
101+
* ../auxiliary/tests_aux.h(.c)
102+
* ../auxiliary/idx_file.h(.c)
103+
104+
References
105+
----------------------------
106+
GitHub project served as starting point for this example:
107+
> Guillaume Chevalier, *LSTMs for Human Activity Recognition*, 2016,[https://github.com/guillaume-chevalier/LSTM-Human-Activity-Recognition](https://github.com/guillaume-chevalier/LSTM-Human-Activity-Recognition)
108+
109+
Human Activity Recognition Using Smartphones [Dataset](https://archive.ics.uci.edu/ml/datasets/Human+Activity+Recognition+Using+Smartphones):
110+
> Davide Anguita, Alessandro Ghio, Luca Oneto, Xavier Parra and Jorge L. Reyes-Ortiz. *"A Public Domain Dataset for Human Activity Recognition Using Smartphones."* 21th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, ESANN 2013. Bruges, Belgium 24-26 April 2013:
111+
112+
IDX file format originally was used for [MNIST database](http://yann.lecun.com/exdb/mnist/). There is a python [package](https://pypi.org/project/idx2numpy/) for working with it through transformation to/from numpy array. *auxiliary/idx_file.c(.h)* is used by the test app for working with IDX files:
113+
> Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. *"Gradient-based learning applied to document recognition."* Proceedings of the IEEE, 86(11):2278-2324, November 1998. [on-line version]

‎samples/modules/embarc_mli/example_har_smartphone/prj.conf

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
# Copyright (c) 2022 Synopsys
2+
# SPDX-License-Identifier: Apache-2.0
3+
CONFIG_EMBARC_MLI=y
4+
CONFIG_ARCMWDT_LIBC=y
5+
CONFIG_CPLUSPLUS=y

‎samples/modules/embarc_mli/example_har_smartphone/sample.yaml

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,7 @@
1+
sample:
2+
description: har_smartphone embarc_mli sample
3+
name: example_har_smartphone
4+
tests:
5+
sample.embarc_mli.example_har_smartphone:
6+
platform_allow: nsim_em
7+
tags: embarc_mli

0 commit comments

Comments
 (0)