TensorFlow LSTM on Arduino and ESP32

TensorFlow LSTM on Arduino and ESP32 cover

When working with time series data (e.g. accelerometer or EMG signals), many of my readers asked me for help on how to deploy their LSTM model to their board.

Until today, I couldn't help, since TensorFlow support in the Arduino IDE seems to have stalled since a few months. Every time I started looking into syncing my EloquentTinyML library to the latest release of TF, I gave up due to the endless count of errors popping out.

But after yet another reader asked me for help, I decided to settle down and fix this. It took me 3 days of work to make it work, but I finally succeded.

The goal of this tutorial is to teach you:

  1. how to train a LSTM model in the browser, without installing anything on your PC
  2. how to run that model on your ESP32

Train & Export a LSTM model

Run a LSTM model on ESP32

Running the exported network is pretty easy thanks to the EloquentTinyML library. You can install it from the Arduino IDE Library Manager.

Then copy the sketch below.

See source

Filename: LSTMExample.ino

/**
 * Run a TensorFlow model to predict the IRIS dataset
 * For a complete guide, visit
 * https://eloquentarduino.com/tensorflow-lite-esp32
 */
// replace with your own model
// include BEFORE <eloquent_tinyml.h>!
#include <Arduino.h>
#include "model.h"
// include the runtime specific for your board
// either tflm_esp32 or tflm_cortexm
#include <tflm_esp32.h>
// now you can include the eloquent tinyml wrapper
#include <eloquent_tinyml.h>


// this is trial-and-error process
// when developing a new model, start with a high value
// (e.g. 10000), then decrease until the model stops
// working as expected
#define ARENA_SIZE 30000

Eloquent::TF::Sequential<TF_NUM_OPS, ARENA_SIZE> tf;

// sample data
float idle[450] = { -0.3334, 0.0289, -0.0031, -0.3315, 0.0305, -0.0030, -0.3325, 0.0292, -0.0023, -0.3364, 0.0265, -0.0017, -0.3391, 0.0242, -0.0012, -0.3380, 0.0229, -0.0010, -0.3367, 0.0209, -0.0012, -0.3370, 0.0206, -0.0013, -0.3369, 0.0231, -0.0012, -0.3369, 0.0261, -0.0012, -0.3359, 0.0292, -0.0018, -0.3344, 0.0300, -0.0025, -0.3338, 0.0286, -0.0026, -0.3341, 0.0278, -0.0026, -0.3351, 0.0276, -0.0026, -0.3359, 0.0279, -0.0026, -0.3367, 0.0265, -0.0026, -0.3373, 0.0248, -0.0028, -0.3369, 0.0253, -0.0026, -0.3357, 0.0266, -0.0023, -0.3344, 0.0286, -0.0023, -0.3338, 0.0281, -0.0026, -0.3341, 0.0253, -0.0028, -0.3341, 0.0253, -0.0026, -0.3349, 0.0266, -0.0022, -0.3357, 0.0279, -0.0022, -0.3354, 0.0292, -0.0022, -0.3346, 0.0300, -0.0022, -0.3339, 0.0289, -0.0022, -0.3343, 0.0266, -0.0025, -0.3341, 0.0266, -0.0026, -0.3338, 0.0283, -0.0026, -0.3349, 0.0296, -0.0026, -0.3367, 0.0300, -0.0025, -0.3373, 0.0296, -0.0022, -0.3365, 0.0261, -0.0018, -0.3354, 0.0222, -0.0015, -0.3356, 0.0201, -0.0009, -0.3364, 0.0200, -0.0005, -0.3375, 0.0204, -0.0004, -0.3385, 0.0217, -0.0000, -0.3377, 0.0245, -0.0002, -0.3367, 0.0268, -0.0007, -0.3360, 0.0283, -0.0012, -0.3357, 0.0292, -0.0020, -0.3356, 0.0297, -0.0030, -0.3356, 0.0297, -0.0038, -0.3349, 0.0296, -0.0038, -0.3344, 0.0291, -0.0035, -0.3346, 0.0284, -0.0028, -0.3349, 0.0281, -0.0025, -0.3351, 0.0278, -0.0022, -0.3343, 0.0274, -0.0020, -0.3331, 0.0281, -0.0020, -0.3336, 0.0292, -0.0018, -0.3352, 0.0279, -0.0022, -0.3370, 0.0232, -0.0023, -0.3373, 0.0204, -0.0018, -0.3369, 0.0222, -0.0017, -0.3369, 0.0245, -0.0017, -0.3367, 0.0274, -0.0018, -0.3346, 0.0294, -0.0017, -0.3326, 0.0292, -0.0017, -0.3336, 0.0279, -0.0023, -0.3339, 0.0281, -0.0030, -0.3338, 0.0292, -0.0033, -0.3347, 0.0292, -0.0031, -0.3357, 0.0289, -0.0026, -0.3360, 0.0274, -0.0025, -0.3352, 0.0245, -0.0025, -0.3341, 0.0229, -0.0026, -0.3341, 0.0227, -0.0025, -0.3360, 0.0216, -0.0018, -0.3385, 0.0200, -0.0012, -0.3391, 0.0208, -0.0012, -0.3382, 0.0234, -0.0015, -0.3377, 0.0255, -0.0020, -0.3364, 0.0274, -0.0025, -0.3341, 0.0300, -0.0028, -0.3331, 0.0310, -0.0035, -0.3341, 0.0286, -0.0038, -0.3347, 0.0263, -0.0033, -0.3354, 0.0263, -0.0030, -0.3367, 0.0266, -0.0028, -0.3380, 0.0252, -0.0025, -0.3377, 0.0240, -0.0022, -0.3352, 0.0250, -0.0022, -0.3339, 0.0261, -0.0022, -0.3341, 0.0265, -0.0022, -0.3352, 0.0247, -0.0018, -0.3367, 0.0226, -0.0013, -0.3369, 0.0229, -0.0015, -0.3351, 0.0248, -0.0023, -0.3331, 0.0271, -0.0030, -0.3326, 0.0283, -0.0033, -0.3346, 0.0273, -0.0031, -0.3362, 0.0268, -0.0025, -0.3364, 0.0276, -0.0022, -0.3362, 0.0270, -0.0023, -0.3360, 0.0232, -0.0022, -0.3370, 0.0206, -0.0020, -0.3380, 0.0214, -0.0018, -0.3383, 0.0227, -0.0022, -0.3369, 0.0242, -0.0026, -0.3356, 0.0247, -0.0026, -0.3347, 0.0253, -0.0025, -0.3343, 0.0273, -0.0023, -0.3341, 0.0283, -0.0025, -0.3344, 0.0273, -0.0026, -0.3351, 0.0253, -0.0028, -0.3359, 0.0245, -0.0028, -0.3365, 0.0255, -0.0031, -0.3362, 0.0266, -0.0035, -0.3356, 0.0258, -0.0038, -0.3352, 0.0245, -0.0036, -0.3354, 0.0255, -0.0030, -0.3360, 0.0274, -0.0026, -0.3360, 0.0261, -0.0028, -0.3360, 0.0221, -0.0028, -0.3388, 0.0224, -0.0033, -0.3380, 0.0250, -0.0031, -0.3356, 0.0263, -0.0026, -0.3351, 0.0278, -0.0031, -0.3339, 0.0283, -0.0036, -0.3333, 0.0286, -0.0038, -0.3333, 0.0289, -0.0036, -0.3343, 0.0283, -0.0033, -0.3354, 0.0245, -0.0035, -0.3359, 0.0195, -0.0040, -0.3354, 0.0206, -0.0043, -0.3352, 0.0250, -0.0048, -0.3367, 0.0270, -0.0051, -0.3367, 0.0273, -0.0051, -0.3352, 0.0266, -0.0046, -0.3341, 0.0274, -0.0043, -0.3338, 0.0276, -0.0040, -0.3346, 0.0263, -0.0035, -0.3359, 0.0255, -0.0031, -0.3369, 0.0244, -0.0030, -0.3367, 0.0231, -0.0031, -0.3354, 0.0234, -0.0036, -0.3346, 0.0253, -0.0041, -0.3349, 0.0265, -0.0046, -0.3349, 0.0255, -0.0048, -0.3343, 0.0250, -0.0046, -0.3344, 0.0253, -0.0041, -0.3352, 0.0245, -0.0035, -0.3360, 0.0237, -0.0030, -0.3369, 0.0239, -0.0028, -0.3375, 0.0232, -0.0030 };
float horizontal[450] = { -0.4638, -0.3242, 0.1993, -0.4544, -0.3369, 0.1930, -0.4449, -0.3400, 0.1868, -0.4365, -0.3304, 0.1834, -0.4184, -0.3171, 0.1814, -0.3985, -0.3026, 0.1785, -0.3805, -0.2873, 0.1752, -0.3640, -0.2648, 0.1724, -0.3491, -0.2404, 0.1692, -0.3316, -0.2199, 0.1666, -0.3177, -0.1992, 0.1651, -0.3100, -0.1804, 0.1648, -0.3048, -0.1615, 0.1638, -0.3027, -0.1390, 0.1619, -0.3009, -0.1154, 0.1606, -0.2970, -0.0912, 0.1593, -0.2918, -0.0669, 0.1570, -0.2867, -0.0453, 0.1544, -0.2838, -0.0235, 0.1519, -0.2810, -0.0008, 0.1490, -0.2768, 0.0231, 0.1451, -0.2693, 0.0532, 0.1405, -0.2597, 0.0860, 0.1357, -0.2499, 0.1197, 0.1296, -0.2412, 0.1575, 0.1230, -0.2325, 0.1974, 0.1171, -0.2221, 0.2353, 0.1113, -0.2110, 0.2726, 0.1036, -0.2003, 0.3167, 0.0933, -0.1889, 0.3660, 0.0820, -0.1743, 0.4132, 0.0701, -0.1583, 0.4562, 0.0584, -0.1469, 0.4936, 0.0473, -0.1437, 0.5221, 0.0382, -0.1388, 0.5398, 0.0282, -0.1315, 0.5452, 0.0172, -0.1306, 0.5427, 0.0094, -0.1319, 0.5364, 0.0063, -0.1298, 0.5300, 0.0076, -0.1284, 0.5229, 0.0110, -0.1318, 0.5084, 0.0141, -0.1388, 0.4863, 0.0164, -0.1494, 0.4576, 0.0188, -0.1637, 0.4208, 0.0232, -0.1804, 0.3803, 0.0297, -0.1988, 0.3447, 0.0387, -0.2182, 0.3188, 0.0496, -0.2366, 0.2981, 0.0616, -0.2517, 0.2704, 0.0754, -0.2670, 0.2327, 0.0881, -0.2820, 0.1938, 0.0997, -0.2888, 0.1619, 0.1108, -0.2838, 0.1420, 0.1220, -0.2727, 0.1396, 0.1275, -0.2856, 0.1498, 0.1311, -0.3346, 0.1396, 0.1515, -0.3696, 0.0859, 0.1685, -0.3649, 0.0104, 0.1632, -0.3456, -0.0480, 0.1549, -0.3302, -0.0713, 0.1513, -0.3160, -0.0786, 0.1497, -0.3066, -0.0956, 0.1508, -0.3136, -0.1258, 0.1554, -0.3320, -0.1602, 0.1630, -0.3587, -0.1945, 0.1724, -0.3748, -0.2282, 0.1803, -0.3749, -0.2591, 0.1861, -0.3787, -0.2829, 0.1900, -0.3878, -0.2956, 0.1952, -0.3994, -0.3018, 0.2016, -0.4086, -0.3062, 0.2066, -0.4184, -0.3125, 0.2100, -0.4256, -0.3127, 0.2107, -0.4282, -0.3039, 0.2083, -0.4291, -0.2881, 0.2045, -0.4319, -0.2676, 0.2024, -0.4357, -0.2408, 0.2035, -0.4340, -0.2051, 0.2070, -0.4311, -0.1664, 0.2125, -0.4282, -0.1276, 0.2175, -0.4231, -0.0936, 0.2201, -0.4122, -0.0642, 0.2192, -0.3948, -0.0324, 0.2138, -0.3740, 0.0058, 0.2039, -0.3517, 0.0463, 0.1915, -0.3334, 0.0877, 0.1790, -0.3160, 0.1284, 0.1674, -0.2978, 0.1625, 0.1555, -0.2791, 0.1959, 0.1427, -0.2568, 0.2338, 0.1288, -0.2324, 0.2748, 0.1150, -0.2099, 0.3170, 0.1003, -0.1883, 0.3598, 0.0820, -0.1660, 0.3972, 0.0588, -0.1437, 0.4207, 0.0338, -0.1259, 0.4353, 0.0097, -0.1179, 0.4389, -0.0113, -0.1153, 0.4332, -0.0264, -0.1126, 0.4275, -0.0362, -0.1114, 0.4215, -0.0424, -0.1126, 0.4127, -0.0455, -0.1160, 0.4011, -0.0455, -0.1243, 0.3933, -0.0414, -0.1355, 0.3909, -0.0344, -0.1404, 0.3953, -0.0272, -0.1398, 0.4049, -0.0210, -0.1409, 0.4163, -0.0147, -0.1456, 0.4221, -0.0082, -0.1494, 0.4177, -0.0026, -0.1497, 0.4081, 0.0027, -0.1495, 0.3969, 0.0089, -0.1495, 0.3829, 0.0173, -0.1497, 0.3701, 0.0291, -0.1521, 0.3578, 0.0431, -0.1640, 0.3354, 0.0582, -0.1842, 0.3028, 0.0732, -0.2057, 0.2636, 0.0870, -0.2249, 0.2184, 0.1005, -0.2389, 0.1731, 0.1143, -0.2514, 0.1341, 0.1298, -0.2659, 0.1000, 0.1466, -0.2812, 0.0694, 0.1617, -0.2949, 0.0499, 0.1739, -0.3069, 0.0370, 0.1851, -0.3276, 0.0091, 0.1951, -0.3518, -0.0334, 0.2047, -0.3743, -0.0752, 0.2135, -0.3945, -0.1122, 0.2169, -0.4134, -0.1405, 0.2187, -0.4300, -0.1608, 0.2206, -0.4391, -0.1753, 0.2232, -0.4508, -0.1911, 0.2242, -0.4663, -0.2082, 0.2247, -0.4773, -0.2233, 0.2281, -0.4820, -0.2411, 0.2299, -0.4861, -0.2600, 0.2275, -0.4900, -0.2774, 0.2226, -0.4881, -0.2923, 0.2179, -0.4786, -0.3031, 0.2130, -0.4669, -0.3033, 0.2073, -0.4536, -0.2958, 0.2011, -0.4347, -0.2819, 0.1944, -0.4116, -0.2609, 0.1879, -0.3881, -0.2359, 0.1824, -0.3642, -0.2053, 0.1785, -0.3430, -0.1729, 0.1754, -0.3240, -0.1437, 0.1720, -0.3069, -0.1166, 0.1684, -0.2932, -0.0936, 0.1646, -0.2838, -0.0734, 0.1594 };
float vertical[450] = { -0.5387, 0.1786, 0.0916, -0.6150, 0.1673, 0.1033, -0.6876, 0.1510, 0.1156, -0.7535, 0.1344, 0.1272, -0.8173, 0.1166, 0.1376, -0.8772, 0.0886, 0.1458, -0.9124, 0.0748, 0.1488, -0.9212, 0.0701, 0.1516, -0.9368, 0.0598, 0.1555, -0.9498, 0.0567, 0.1589, -0.9444, 0.0623, 0.1628, -0.9234, 0.0755, 0.1651, -0.8875, 0.0867, 0.1645, -0.8354, 0.0981, 0.1593, -0.7742, 0.1139, 0.1524, -0.7153, 0.1261, 0.1451, -0.6629, 0.1306, 0.1357, -0.6201, 0.1287, 0.1262, -0.5838, 0.1199, 0.1186, -0.5484, 0.1049, 0.1126, -0.5148, 0.0891, 0.1059, -0.4842, 0.0732, 0.0982, -0.4523, 0.0602, 0.0906, -0.4160, 0.0470, 0.0846, -0.3714, 0.0359, 0.0787, -0.3193, 0.0344, 0.0704, -0.2670, 0.0419, 0.0595, -0.2161, 0.0528, 0.0468, -0.1691, 0.0566, 0.0343, -0.1319, 0.0634, 0.0196, -0.1031, 0.0729, 0.0051, -0.0756, 0.0769, -0.0046, -0.0471, 0.0795, -0.0131, -0.0229, 0.0769, -0.0232, -0.0050, 0.0676, -0.0328, 0.0113, 0.0502, -0.0393, 0.0241, 0.0323, -0.0450, 0.0350, 0.0167, -0.0502, 0.0449, 0.0017, -0.0531, 0.0495, -0.0103, -0.0533, 0.0503, -0.0170, -0.0510, 0.0480, -0.0150, -0.0459, 0.0415, -0.0069, -0.0383, 0.0336, -0.0000, -0.0297, 0.0239, -0.0015, -0.0227, 0.0093, -0.0100, -0.0176, -0.0042, -0.0191, -0.0131, -0.0126, -0.0238, -0.0085, -0.0232, -0.0262, -0.0036, -0.0382, -0.0292, 0.0030, -0.0543, -0.0339, 0.0102, -0.0696, -0.0389, 0.0157, -0.0821, -0.0402, 0.0195, -0.0934, -0.0350, 0.0227, -0.1052, -0.0213, 0.0269, -0.1181, -0.0004, 0.0323, -0.1284, 0.0265, 0.0388, -0.1409, 0.0527, 0.0439, -0.1650, 0.0753, 0.0474, -0.2068, 0.0960, 0.0527, -0.2262, 0.0961, 0.0626, -0.2644, 0.1108, 0.0652, -0.3168, 0.1314, 0.0655, -0.3609, 0.1349, 0.0725, -0.4247, 0.1401, 0.0798, -0.4861, 0.1352, 0.0906, -0.5576, 0.1238, 0.1015, -0.6331, 0.1087, 0.1130, -0.7120, 0.0956, 0.1238, -0.7760, 0.0789, 0.1339, -0.8230, 0.0663, 0.1433, -0.8611, 0.0605, 0.1492, -0.8878, 0.0572, 0.1541, -0.8979, 0.0657, 0.1581, -0.8951, 0.0660, 0.1625, -0.9029, 0.0541, 0.1664, -0.9016, 0.0491, 0.1681, -0.8821, 0.0468, 0.1687, -0.8590, 0.0460, 0.1684, -0.8225, 0.0484, 0.1679, -0.7770, 0.0496, 0.1661, -0.7317, 0.0497, 0.1620, -0.6878, 0.0512, 0.1558, -0.6427, 0.0540, 0.1487, -0.5974, 0.0593, 0.1422, -0.5530, 0.0620, 0.1349, -0.5082, 0.0621, 0.1262, -0.4636, 0.0628, 0.1166, -0.4213, 0.0610, 0.1078, -0.3829, 0.0605, 0.0995, -0.3354, 0.0519, 0.0922, -0.2830, 0.0397, 0.0847, -0.2394, 0.0359, 0.0753, -0.2031, 0.0305, 0.0640, -0.1695, 0.0182, 0.0502, -0.1326, 0.0151, 0.0346, -0.0869, 0.0187, 0.0193, -0.0414, 0.0118, 0.0032, -0.0086, 0.0008, -0.0140, 0.0173, -0.0101, -0.0292, 0.0430, -0.0155, -0.0419, 0.0718, -0.0157, -0.0529, 0.1011, -0.0212, -0.0606, 0.1244, -0.0340, -0.0640, 0.1384, -0.0515, -0.0669, 0.1376, -0.0744, -0.0681, 0.1195, -0.0939, -0.0651, 0.0902, -0.1005, -0.0578, 0.0500, -0.0918, -0.0461, 0.0111, -0.0723, -0.0316, -0.0156, -0.0448, -0.0175, -0.0364, -0.0122, -0.0053, -0.0527, 0.0161, 0.0053, -0.0613, 0.0340, 0.0141, -0.0652, 0.0460, 0.0199, -0.0694, 0.0528, 0.0224, -0.0753, 0.0556, 0.0225, -0.0831, 0.0603, 0.0245, -0.1023, 0.0712, 0.0308, -0.1368, 0.0859, 0.0385, -0.1712, 0.0939, 0.0440, -0.1962, 0.0973, 0.0455, -0.2088, 0.0838, 0.0504, -0.2324, 0.0872, 0.0496, -0.2542, 0.0870, 0.0445, -0.2685, 0.0703, 0.0445, -0.2901, 0.0621, 0.0431, -0.2968, 0.0410, 0.0471, -0.3167, 0.0255, 0.0585, -0.3644, 0.0222, 0.0745, -0.4148, 0.0437, 0.0717, -0.4622, 0.0803, 0.0273, -0.5445, 0.0794, -0.0079, -0.6842, 0.0440, 0.0237, -0.8333, 0.0216, 0.0992, -0.9581, 0.0257, 0.1858, -1.0000, 0.0738, 0.2670, -1.0000, 0.1581, 0.3233, -1.0000, 0.2535, 0.3445, -1.0000, 0.3333, 0.3417, -1.0000, 0.3915, 0.3193, -0.9859, 0.4314, 0.2779, -0.8331, 0.4270, 0.2291, -0.6754, 0.3801, 0.1840, -0.5463, 0.3095, 0.1471, -0.4643, 0.2306, 0.1114, -0.4247, 0.1570, 0.0831, -0.4067, 0.0976, 0.0701, -0.3966, 0.0496, 0.0642, -0.3878, 0.0152, 0.0600 };

/**
 * Run prediction
 */
void predictSample(const char *classLabel, float *input, uint8_t expectedOutput) {
    // classify class 0
    if (!tf.predict(input).isOk()) {
        Serial.println(tf.exception.toString());
        return;
    }

    Serial.print("Predicting sample of ");
    Serial.print(classLabel);
    Serial.print(": expcted id=");
    Serial.print(expectedOutput);
    Serial.print(", predicted=");
    Serial.println(tf.classification);
}

/**
 *
 */
void setup() {
    Serial.begin(115200);
    delay(3000);
    Serial.println("__TENSORFLOW LSTM__");

    // configure input/output
    // (not mandatory if you generated the .h model
    // using the eloquent_tensorflow Python package)
    tf.setNumInputs(TF_NUM_INPUTS);
    tf.setNumOutputs(TF_NUM_OUTPUTS);

    registerNetworkOps(tf);

    while (!tf.begin(tfModel).isOk()) {
        Serial.println(tf.exception.toString());
        delay(1000);
    }
}


void loop() {
    predictSample("idle", idle, 0);
    predictSample("horizontal", horizontal, 1);
    predictSample("vertical", vertical, 2);

    // how long does it take to run a single prediction?
    Serial.print("It takes ");
    Serial.print(tf.benchmark.microseconds());
    Serial.println("us for a single prediction");

    delay(1000);
}
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82

Here's a brief explanation of the important parts of the sketch.

Import library and instantiate nn

// replace with your own model
// include BEFORE <eloquent_tinyml.h>!
#include "tfModel.h"
// include the runtime specific for your board
// ESP32 is supported ATM
#include <tflm_esp32.h>
#include <eloquent_tinyml.h>

#define ARENA_SIZE 30000

Eloquent::TF::Sequential<TF_NUM_OPS, ARENA_SIZE> tf;
1 2 3 4 5 6 7 8 9 10 11

We begin with importing the eloquent libraries and the exported model:

The only thing you need to customize here is ARENA_SIZE. This value defines how much memory the model will be allocated. Finding the optimal value is a trial-and-error process because there's not an exact formula to apply. Larger values will work but leave less space to your own code. Smaller values will prevent the network to run correctly. I suggest you start with a large value (e.g. 30000) and decrease it until your model start throwing errors about tensors allocation.

Model configuration and initialization

// configure input/output
// these constants are defined in tfModel.h
tf.setNumInputs(TF_NUM_INPUTS);
tf.setNumOutputs(TF_NUM_OUTPUTS);

// register the required ops
registerNetworkOps(tf);

while (!tf.begin(tfModel).isOk()) 
    Serial.println(tf.exception.toString());
1 2 3 4 5 6 7 8 9 10

These lines set the number of inputs/outputs and add the required operations to the model. The Notebook above populated these values for you. If you are doing something manually, you should customize these values and register all the required ops.

Once done, initialize the network by passing the exported model from Step 2.

Execution

Finally, you can execute the network by passing it an input vector.

// sample data
float idle[450] = { -0.3334, 0.0289, -0.0031, -0.3315, ... };

if (!tf.predict(idle).isOk())
    Serial.println(tf.exception.toString());
1 2 3 4 5

After the execution, you can access the results with

tf.output(i)
1

where i is the index of the output.

For example, the Notebook dataset is made by 3 classes (idle, horizontal, vertical) and the model outputs 3 values, representing the probability of each class.

Since this is a classification task, you have access to tf.classification, which returns the class index with the highest probability.

To iterate over the results you can use a for loop:

for (int i = 0; i < tf.numOutputs; i++) {
    Serial.print(tf.output(i));
    Serial.print(", ");
}
1 2 3 4

lstm-output.png 119.73 KB

Subscribe to my newsletter

Join 1169 businesses and hobbysts skyrocketing their Arduino + ESP32 skills twice a month