First Adapt embedding net, then fine-tune the last layer.

Used for test only

split sensor data into small interval, for LSTM unit training sensor_data: numpy.array with shape [seq_len, 8] interval_len: int, the len

add the fourth dim for the original :param data: numpy.array ax, ay, az, mx, my, mz data with shape [interval_len, 6] :return: numpy.arra

Args: in_dir: folder where the original txt data stores (format of the txt file can be found in readme) out_dir: output .pickle f

Feature extraction for each data sample, FFT, frequency and magnitude

:param rounds: global rounds for federated learning: type: float :param in_dir: input data dir: type: string :param out_dir: output re

:param rounds: global rounds for federated learning: type: float :param out_dir: output result dir: type: string :param lr: initial le

model_1 and model_2 with same structure return weights dictionary with values w_model_1 - w_model_2 sigma = 1 : federated learning

model 1 and model 2 with same structure return weights dict with values w_model1 - w_model2 sigma = 1 : federated learning

Args: index (int): Index Returns: tuple: (image, target, target_t) target is index of the g

graph can be norm_cce, merge_cce

Args: filename: a list of pickle files. transform: transform applied to the feature data.

graph can be norm_embed

First Adapt embedding net, then fine-tune the last layer.

First Adapt embedding net, then fine-tune the last layer.

First Adapt embedding net, then fine-tune the last layer.

Args: filename: a list of pickle files that stores the processed HAR data (after FFT) batch_size: batch size

input x with shape b, c = 1, seq_len, sensor_dim = 16, interval_len = (7 /12 ?)

x with shape [batch, channel, seq_len, H, W]

filename example: /path/checkpoint/model-100.t7 State dict contains: "model" key at least.

filename example: /path/checkpoint/model-100.t7 State dict contains: "model" key at least.

abandon: more complicated way to update the model on the server side.