Character Trajectories

The characters here were used for a PhD study on primitive extraction using HMM based models. The data consists of 2858 character samples, contained in the cell array 'mixout'. The struct variable 'consts' contains a field consts.charlabels which provides ennummerated labels for the characters. consts.key provides the key for each label. The data was captured using a WACOM tablet. 3 Dimensions were kept - x, y, and pen tip force. The data has been numerically differentiated and Gaussian smoothed, with a sigma value of 2. Data was captured at 200Hz. The data was normalised with consts.datanorm. Only characters with a single 'PEN-DOWN' segment were considered. Character segmentation was performed using a pen tip force cut-off point. The characters have also been shifted so that their velocity profiles best match the mean of the set.

This is a pre-processed version of the dataset saved in numpy format. The original dataset is obtained from UCI.

The data are 3-dimensional arrays of shape [n_samples, time_steps, n_variables]. The data can be loaded as follows:

loaded_data = np.load("CHAR.npz")
Xtr = loaded_data['Xtr'] # Training data of shape (300, 205, 3)
Ytr = loaded_data['Ytr'] # Training labels of shape (300, 1)
Xte = loaded_data['Xte'] # Test data of shape (2558, 205, 3)
Yte = loaded_data['Yte'] # Test labels of shape (2558, 1)