forked from AlexanderFengler/nn_likelihoods
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathddm_mle.py
375 lines (299 loc) · 17.8 KB
/
ddm_mle.py
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
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
import numpy as np
import scipy as scp
import os
import pandas as pd
import matplotlib as plt
# Local imports
import ddm_data_simulation
import make_data_wfpt as mdw
import dnnregressor_train_eval_keras as dnnk
class ddm_mle_estimator:
def __init__(self):
self.ddm_sim_params = dict({'v': 1,
'a': 1,
'w': 0.5,
's': 1,
'delta_t': 0.001,
'max_t': 20,
'sample_size': 1000}
)
self.parameter_bounds = dict({'v': [-2, 2],
'a': [0.3, 4],
'w': [0.001, 0.999]
}
)
# Genetic algorithm related
self.gen_alg_params = dict({'population_size': 40,
'mutation_probability': 0.1,
'print_steps': True,
'steps': 100}
)
self.gen_alg_population = []
self.gen_alg_population_record = np.zeros((self.gen_alg_params['steps'], self.gen_alg_params['population_size'], 3))
self.gen_alg_fitness = []
self.gen_alg_fitness_record = np.zeros((self.gen_alg_params['steps'], self.gen_alg_params['population_size']))
# Grid search related
self.grid_search_parameters = dict({'precision_v': 0.2,
'precision_a': 0.2,
'precision_w': 0.1})
# Data and miscellaneous
self.data = []
# self.model_directory = 'none' # directory for model that can predict likelihoods (to be passed as parameter to a function that restores the model from tensorflow for example)
# self.hyper_parameter_file = ''
self.model_data = {'model_path': '',
'ckpt_path': ''
}
# Meta
self.meta_parameters = dict({'model': 'navarro',
'datatype': 'choice_probabilities',
'optimizer': 'genetic'})
def get_keras_model(self):
model = keras.models.load_model(self.model_data['model_path'])
model.load_weights(self.model_data['ckpt_path'])
self.model = model
return model
def make_data(self):
rts, choices = ddm_data_simulation.ddm_simulate_rts(v = self.ddm_sim_params['v'],
a = self.ddm_sim_params['a'] / 2, # we divide by 2 => correct conversion when simulating with euler narayujan
w = self.ddm_sim_params['w'],
s = self.ddm_sim_params['s'],
delta_t = self.ddm_sim_params['delta_t'],
max_t = self.ddm_sim_params['max_t'],
n_samples = self.ddm_sim_params['sample_size']
)
self.p_lower_barrier = np.sum(choices > 0) / len(choices)
self.data = np.multiply(rts, choices)
# functions related to genetic algorithm setup
def make_pop_init(self):
population = pd.DataFrame(np.zeros((self.gen_alg_params['population_size'], 3)), columns = ['v', 'a', 'w'])
population.loc[:, list(self.ddm_sim_params.keys())[0]] = np.random.uniform(low = self.parameter_bounds['v'][0], high = self.parameter_bounds['v'][1], size = self.gen_alg_params['population_size'])
population.loc[:, list(self.ddm_sim_params.keys())[1]] = np.random.uniform(low = self.parameter_bounds['a'][0], high = self.parameter_bounds['a'][1], size = self.gen_alg_params['population_size'])
population.loc[:, list(self.ddm_sim_params.keys())[2]] = np.random.uniform(low = self.parameter_bounds['w'][0], high = self.parameter_bounds['w'][1], size = self.gen_alg_params['population_size'])
self.gen_alg_population = population.copy()
def make_selection_probabilities(self, n_keep = 0):
probabilities = np.zeros((n_keep,))
denom = sum(np.arange(1,n_keep + 1, 1))
for n in range(1, n_keep + 1, 1):
num = n_keep - n + 1
probabilities[n-1] = num / denom
return probabilities
def population_fitness_rt_choice_nf(self):
fitness = np.zeros((self.gen_alg_params['population_size'], ))
for i in range(0, self.gen_alg_params['population_size'], 1):
fitness[i] = self.loglik_choice_rt_nf(eps = 1e-29,
v = self.gen_alg_population.loc[i, 'v'],
a = self.gen_alg_population.loc[i, 'a'],
w = self.gen_alg_population.loc[i, 'w']
)
self.gen_alg_fitness = fitness
def population_fitness_rt_choice_dnn(self):
fitness = np.zeros((self.gen_alg_params['population_size'], ))
for i in range(0, self.gen_alg_params['population_size'], 1):
data = pd.DataFrame(np.zeros((len(self.data), 5)), columns = ['v','a', 'w', 'rt', 'nf_likelihood'])
data.loc[:, 'v'] = self.gen_alg_population.loc[i, 'v']
data.loc[:, 'a'] = self.gen_alg_population.loc[i, 'a']
data.loc[:, 'w'] = self.gen_alg_population.loc[i, 'w']
data.loc[:, 'rt'] = self.data.flatten()
data.loc[:, 'nf_likelihood'] = 0.0
features, labels, __, ___ = mdw.train_test_split_rt_choice(data = data,
p_train = 1.0,
write_to_file = False,
from_file = False,
backend = 'keras'
)
#self.features = features
#self.labels = labels
fitness[i] = np.sum(np.log(1e-29 + np.abs(self.model.predict(features))))
# fitness[i] = np.sum(np.log(1e-29 + dnn_predictor.get_predictions(regressor = self.dnn_predictor,
# features = features,
# labels = labels)
# )
# )
self.gen_alg_fitness = fitness
def population_fitness_choice_p_nf(self):
fitness = np.zeros((self.gen_alg_params['population_size'], ))
for i in range(0, self.gen_alg_params['population_size'], 1):
fitness[i] = self.loglik_choice_probability(v = self.gen_alg_population.loc[i, 'v'],
a = self.gen_alg_population.loc[i, 'a'],
w = self.gen_alg_population.loc[i, 'w']
)
self.gen_alg_fitness = fitness
def make_next_generation(self):
pop_size = self.gen_alg_params['population_size']
assert ((pop_size % 2) % 2) == 0, 'Population size divided by two should be an even integer'
n_keep = pop_size // 2
# Get indices of sorted fitness values for current generation
# Allows us to shave of the top # (n_keep)
fit_sort_id = list(reversed(np.argsort(self.gen_alg_fitness)))
# Create pool to start from for next generation (best n_keep values)
pop_pool_new_gen = self.gen_alg_population.loc[fit_sort_id[:n_keep], :].values.copy()
selection_probabilities = self.make_selection_probabilities(n_keep = n_keep)
# Get base for next gen
indices = np.random.choice(np.arange(0, n_keep, 1), p = selection_probabilities, size = n_keep)
print('indices: ', indices)
#print('pop_pool_new_gen_going_in: ', pop_pool_new_gen)
pop_pool_new_gen = pop_pool_new_gen[indices, :]
#print('pop_pool_new_gen filtered: ', pop_pool_new_gen[indices, :])
# Now creating full next generation
new_gen = np.zeros((pop_size, len(self.parameter_bounds)))
# Fill in upper half of new_gen array with new offspring
cnt = 0
while cnt < ((pop_size // 2) - 1):
new_gen[cnt, :] = pop_pool_new_gen[cnt, :]
cnt += 1
new_gen[cnt, :] = pop_pool_new_gen[cnt, :]
# randomly select indice
indice = np.random.choice(np.arange(0, len(self.parameter_bounds), 1))
weighting_beta = np.random.uniform(low = 0, high = 1)
new_gen[(cnt - 1), indice] = ((1 - weighting_beta) * pop_pool_new_gen[cnt - 1, indice]) + (weighting_beta * pop_pool_new_gen[cnt, indice])
new_gen[cnt, indice] = ((1 - weighting_beta) * pop_pool_new_gen[cnt, indice]) + (weighting_beta * pop_pool_new_gen[cnt - 1, indice])
cnt += 1
# Fill in lower half with old generations top values
new_gen[cnt:, :] = pop_pool_new_gen
#print('new gen: ', new_gen)
# Include random mutations
# Generate mutation locations
mutations = np.reshape(np.random.choice([1,0], p = [self.gen_alg_params['mutation_probability'], 1 - self.gen_alg_params['mutation_probability']],
size = pop_size * len(self.parameter_bounds)),
newshape = (pop_size, len(self.parameter_bounds)))
# Make iterable
mutation_locations = zip(*np.where(mutations == 1))
# Apply mutations
my_keys = list(self.parameter_bounds.keys())
for (i,j) in mutation_locations:
new_gen[i, j] = np.random.uniform(low = self.parameter_bounds[my_keys[j]][0],
high = self.parameter_bounds[my_keys[j]][1]
)
#print('new_gen_just_before_making_it_dataframe: ', new_gen)
#print('new_gen_data_frame : ', pd.DataFrame(new_gen, columns = ['v', 'a', 'w']))
self.gen_alg_population = pd.DataFrame(new_gen, columns = ['v', 'a', 'w'])
if self.meta_parameters['model'] == 'navarro':
self.population_fitness_rt_choice_nf()
if self.meta_parameters['model'] == 'dnn':
self.population_fitness_rt_choice_dnn()
def make_fitness_stats(self):
return np.mean(self.gen_alg_fitness), np.max(self.gen_alg_fitness)
def run_gen_alg(self):
self.make_pop_init()
if self.meta_parameters['model'] == 'navarro':
self.population_fitness_rt_choice_nf()
#self.gen_alg_fitness_max = np.zeros((self.gen_alg_steps, ))
#self.gen_alg_fitness_mean = np.zeros((self.gen_alg_steps, ))
if self.meta_parameters['model'] == 'dnn':
self.population_fitness_rt_choice_dnn()
if self.meta_parameters['model'] == 'choice_probabilities':
self.population_fitness_choice_p_nf()
for step in range(0, self.gen_alg_params['steps'], 1):
self.make_next_generation()
print('cur_population: ', self.gen_alg_population)
print('cur_fitness: ', self.gen_alg_fitness)
#self.gen_alg_fitness_mean[step], self.gen_alg_fitness_max[step] = self.make_fitness_stats()
# Record current population and fitness
self.gen_alg_fitness_record[step] = self.gen_alg_fitness
self.gen_alg_population_record[step] = self.gen_alg_population
if self.gen_alg_params['print_steps']:
print(step)
def plot_gen_alg(self):
plt.plot(arange(0, self.gen_alg_steps, 1), self.gen_alg_fitness_mean, 'g.')
plt.plot(arange(0, self.gen_alg_steps, 1), self.gen_alg_fitness_mean, 'r.')
plt.show()
# compute log likelihood one set of data
def loglik_choice_rt_nf(self,
eps = 1e-29,
v = 0,
a = 1,
w = 0.5):
tmp = 0
for t in self.data:
tmp += np.log(mdw.fptd(t, v, a, w, eps))
return tmp
def d_kl_choice_probabiliy_navarro(self,
eps = 1e-29,
v = 0,
a = 1,
w = 0.5):
p_navarro = mdw.choice_probabilities(v = v,
a = a,
w = w)
p_data = self.p_lower_barrier
d_kl = p_navarro * np.log(p_data) + (1 - p_navarro) * np.log(1 - p_data)
return d_kl
def loglik_choice_probability_nf(self,
eps = 1e-29,
v = 0,
a = 1,
w = 0.5):
log_p_navarro = np.log(mdw.choice_probabilities(v = v,
a = a,
w = w)
)
tmp = 0
for c in np.sign(self.data):
if c > 0:
tmp += log_p_navarro
else:
tmp += log_p_navarro
return tmp
# Methods related to grid search
def grid_search_make_grid(self):
v_vals = np.arange(self.parameter_bounds['v'][0], self.parameter_bounds['v'][1], self.grid_search_parameters['precision_v'][0])
a_vals = np.arange(self.parameter_bounds['a'][0], self.parameter_bounds['a'][1], self.grid_search_parameters['precision_a'][0])
w_vals = np.arange(self.parameter_bounds['w'][0], self.parameter_bounds['w'][1], self.grid_search_parameters['precision_w'][0])
grid = pd.DataFrame(np.zeros(len(v_vals)*len(a_vals)*len(w_vals), 3),
columns = ['v', 'a', 'w'])
cnt = 0
for v_tmp in v_vals:
for a_tmp in a_vals:
for w_tmp in w_vals:
grid.loc[cnt] = [v_tmp, a_tmp, w_tmp]
cnt += 1
self.grid_search_grid = grid
def run_grid_search(self):
self.grid_search_make_grid()
self.grid_search_results = self.grid_search_grid.copy()
self.grid_search_results['loglik'] = 0
self.grid_search_results['d_kl'] = 0
for i in range(0, len(self.grid_search_grid['v']), 1):
if self.meta_parameters['datatype'] == 'choice_rt':
if self.meta_parameters['model'] == 'navarro':
self.grid_search_results.loc[i, 'loglik'] = self.loglik_choice_rt_nf(v = self.grid_search_grid.loc[i, 'v'],
a = self.grid_search_grid.loc[i, 'a'],
w = self.grid_search_grid.loc[i, 'w'])
if self.meta_parameters['model'] == 'dnn':
#self.grid_search_results.loc[i, 'loglik'] =
if self.meta_parameters['datatype'] == 'choice_probabilities':
self.grid_search_results.loc[i, 'loglik'] = self.loglik_choice_probability_nf(v = self.grid_search_grid.loc[i, 'v'],
a = self.grid_search_grid.loc[i, 'a'],
w = self.grid_search_grid.loc[i, 'w'])
self.grid_search_results['d_kl'] = self.d_kl_choice_probability_navarro(v = self.grid_search_grid.loc[i, 'v'],
a = self.grid_search_grid.loc[i, 'a'],
w = self.grid_search_grid.loc[i, 'w'])
# ---------------------------------------------------------------------------------------------------------------------------------------------------
# # Methods related to importing tensorflow model for likelihood
# def tf_estimator_hyperparameters(self):
# hyper_params= pd.read_csv(self.model_directory + '/' + hyper_parameter_file,
# converters = {'hidden_units': eval,
# 'activations': eval}
# )
# model_params = hyper_params.to_dict(orient = 'list')
# for key in model_params.keys():
# model_params[key] = model_params[key][0]
#
# return model_params
# def initialize_dnn_predictor(self):
# if self.model_directory == 'none':
# print('please specify directory of your dnn_regressor model files!')
# if self.model_directory != 'none':
# # Get hyperparameters to feed into dnn_regressor model_input function
# self.dnn_model_params = self.tf_estimator_hyperparameters()
#
# # Make feature columns and add to
# features = dict({'v': [],
# 'a': [],
# 'w': [],
# 'rt': [],
# 'choice': []}
# )
# feature_columns = dnn_model_input.make_feature_columns_numeric(features = features)
# self.dnn_model_params['feature_columns'] = feature_columns
# self.dnn_predictor = dnn_predictor.get_dnnreg_predictor(model_directory = self.model_directory, params = self.dnn_model_params)