I looked at your model / data and found that it’s possible to predict the type of movement with 100% accuracy by just looking at log_cantidad. If this variable is greater than ~5, then it will always give a value of 1.
Try running the following so you can see for yourself:
pd.plotting.scatter_matrix(data[['tipo_movimiento','log_cantidad','dia_normalized']], figsize=(10,10));
When there is virtually zero stochastic noise, the sampler is going to favor making the distinction between movement / no movement as sharp as possible which means very large values for the relevant coefficients.
Also, as a minor piece of advice, you can use pm.Bernoulli instead of pm.Binomial with n=1 for binary data.