def simulate_switch_states(num_points,
p_switch_to_switch,
p_noswitch_to_switch,
p_init_switch=0.5,
init_state=None):
"""
Simulate switch states.
"""
if init_state is None:
init_state = prob_utils.sample_binary_state(p_init_switch)
data = [init_state]
if num_points == 1:
return data
points = range(num_points)
for n in points[1:]:
if data[n-1] == 0:
# probability of transitioning from switch to switch state
next_state = prob_utils.sample_binary_state(p_switch_to_switch)
else:
next_state = prob_utils.sample_binary_state(p_noswitch_to_switch)
data.append(next_state)
return data
def ssm_nutrient_simulator(time_obj,
out_trans_mat1=None,
out_trans_mat2=None,
p_switch_to_switch=0.8,
p_noswitch_to_switch=0.2,
p_init_switch=0.5,
p_init_output=0.5):
"""
Nutrient simulator where the generative process is
the 2-hidden state switch SSM.
This uses the two given output transition matrices
(out_trans_mat1, out_trans_mat2) to do the sampling.
Note that this only supports 2 hidden states.
"""
# two transition matrices to switch between
####
#### TODO: move these transition matrices to be
#### parameters
####
if (out_trans_mat1 is None) or (out_trans_mat2 is None):
raise Exception, "Expected an output transition matrix."
out_trans_mat1 = np.array(out_trans_mat1)
# constant transition matrix
out_trans_mat2 = np.array(out_trans_mat2)
trans_mats = [out_trans_mat1, out_trans_mat2]
num_points = len(time_obj.t)
# draw switch states
switch_states = \
simulate_switch_states(num_points - 1,
p_switch_to_switch=p_switch_to_switch,
p_noswitch_to_switch=p_noswitch_to_switch,
p_init_switch=p_init_switch)
data = [prob_utils.sample_binary_state(p_init_output)]
prev_output = data[0]
for n in xrange(1, num_points, 1):
# choose which transition matrix to draw from
trans_mat = trans_mats[switch_states[n - 1]]
# draw data point
data_point = \
np.random.multinomial(1, trans_mat[prev_output, :]).argmax()
data.append(data_point)
prev_output = data_point
data = np.array(data)
return data
def plastic_growth_policy(time_obj, env_obj, params):
"""
The plastic growth policy from Jablonka (1995). Match whatever
nutrient environment we saw in previous time step (based
on what the lag in the environment is.)
"""
####
#### TODO: incorporate the lag here
####
result = {"t": time_obj.t}
data = env_obj.hist
# choose initial state randomly
prev_data = prob_utils.sample_binary_state(0.5)
result["nutrient_states"] = []
num_nutrients = len(params["nutr_labels"])
num_timepoints = len(time_obj.t)
result["prob_nutrient_states"] = np.zeros((num_timepoints,
num_nutrients))
for n in xrange(num_timepoints):
result["nutrient_states"].append(prev_data)
result["prob_nutrient_states"][n, prev_data] = 1
prev_data = data[n]
return result
