Skip to content

Commit ff176c9

Browse files
augeorgeaugeorge
committed
updates
1 parent 68ce0aa commit ff176c9

8 files changed

Lines changed: 204 additions & 987 deletions

File tree

‎example/.png‎

-11.3 KB
Binary file not shown.

‎example/2d_example.ipynb‎

Lines changed: 168 additions & 896 deletions
Large diffs are not rendered by default.

‎example/__pycache__/toy_model.cpython-310.pyc‎

-658 Bytes
Binary file not shown.

‎example/plot_animation.py‎

Lines changed: 0 additions & 28 deletions
This file was deleted.

‎pyGridSampler/__pycache__/__init__.cpython-310.pyc‎

-475 Bytes
Binary file not shown.

‎pyGridSampler/__pycache__/grid_sampler.cpython-310.pyc‎

-7.09 KB
Binary file not shown.

‎pyGridSampler/__pycache__/grid_tools.cpython-310.pyc‎

-13 KB
Binary file not shown.

‎pyGridSampler/grid_sampler.py‎

Lines changed: 36 additions & 63 deletions
Original file line numberDiff line numberDiff line change
@@ -22,7 +22,12 @@ def __init__(self, func, args_list, dataset, x_bounds, x_shifts):
2222
self.x_shifts = x_shifts
2323
self.n_dim = len(x_bounds)
2424
self.n_data_points = len(dataset)
25-
self.output = []
25+
self.grid_results = []
26+
self.x_spacing_results = []
27+
self.log_like_results = []
28+
self.ess_results = []
29+
self.func_evals = []
30+
self.grid_size = []
2631

2732

2833
def calc_naive_grid(self, grid_resolution, data_tempering_index, n_processes=4):
@@ -40,7 +45,7 @@ def calc_naive_grid(self, grid_resolution, data_tempering_index, n_processes=4):
4045
log_likelihoods, rel_prob, weights, ess = gt.eval_grid_points(grid, self.func, self.args_list[data_tempering_index], n_processes)
4146
return grid, x_spacing, log_likelihoods, rel_prob, weights, ess
4247

43-
def initialize(self, init_grid_resolution, init_data_size, ess_min, n_processes=4, max_iter=100):
48+
def initialize(self, init_grid_resolution, init_data_size, ess_min, n_processes=4, max_iter=100, store_results=False):
4449
"""Iteratively update the initial grid until the ESS is greater than the specified minimum ESS.
4550
4651
Args:
@@ -49,6 +54,8 @@ def initialize(self, init_grid_resolution, init_data_size, ess_min, n_processes=
4954
ess_min (float): The minimum effective sample size (ESS) to be used for initialization.
5055
n_processes (int): The number of parallel processes to use when evaluating the grid. Defaults to 4.
5156
max_iter (int): The maximum number of iterations to use when updating the grid. Defaults to 100.
57+
store_results (bool, optional): Stores results at each tempering stage (may use a lot of memory for large grids / datasets)
58+
Defaults to False.
5259
5360
Returns:
5461
Tuple[int, int, np.ndarray, np.ndarray, np.ndarray, np.ndarray, np.ndarray, float]: A tuple containing the following elements:
@@ -67,16 +74,25 @@ def initialize(self, init_grid_resolution, init_data_size, ess_min, n_processes=
6774

6875
grid_resolution = init_grid_resolution
6976
iter = 0
77+
f_evals = np.size(log_likelihoods)
7078
with tqdm(total=None, desc="Initializing...") as pbar:
7179
while ess < ess_min and iter < max_iter:
7280
grid_resolution = grid_resolution + 1
7381
grid, x_spacing, log_likelihoods, rel_prob, weights, ess = self.calc_naive_grid(grid_resolution, data_tempering_index, n_processes)
7482
iter = iter + 1
83+
f_evals = f_evals + np.size(log_likelihoods)
7584
pbar.update()
76-
pbar.set_description(f"Intialization: data_size={data_tempering_index+1}, grid_resolution={grid_resolution}, n_grid_points={np.shape(grid)[0]}, ESS={ess}")
85+
pbar.set_description(f"Intialization: data_size={data_tempering_index+1}, grid_resolution={grid_resolution}, n_grid_points={np.shape(grid)[0]}, ESS={ess}, init_func_evals:{f_evals}")
86+
if store_results:
87+
self.grid_results.append(grid)
88+
self.x_spacing_results.append(x_spacing)
89+
self.log_like_results.append(log_likelihoods)
90+
self.ess_results.append(ess)
91+
self.func_evals.append(f_evals)
92+
self.grid_size.append(np.shape(grid)[0])
7793
return grid_resolution, data_tempering_index+1, grid, x_spacing, log_likelihoods, rel_prob, weights, ess
7894

79-
def initialize_and_sample(self, init_grid_resolution, init_data_size, ess_min, delta, n_processes=4, max_iter=100, store_grid=False):
95+
def initialize_and_sample(self, init_grid_resolution, init_data_size, ess_min, delta, n_processes=4, max_iter=100, store_results=False):
8096
""" Initializes the grid and performs the data tempering (iterative batching) to obtain posterior samples.
8197
8298
Args:
@@ -88,7 +104,7 @@ def initialize_and_sample(self, init_grid_resolution, init_data_size, ess_min, d
88104
Defaults to 4.
89105
max_iter (int, optional): The maximum number of iterations for updating the grid.
90106
Defaults to 100.
91-
store_grid (bool, optional): Stores reduced grid at each tempering stage (may use a lot of memory for large grids / datasets)
107+
store_results (bool, optional): Stores results at each tempering stage (may use a lot of memory for large grids / datasets)
92108
Defaults to False.
93109
94110
Returns:
@@ -104,47 +120,18 @@ def initialize_and_sample(self, init_grid_resolution, init_data_size, ess_min, d
104120
"""
105121

106122
# generate initial (uniform) grid samples
107-
grid_resolution, data_size, grid, x_spacing, log_likelihoods, rel_prob, weights, ess = self.initialize(init_grid_resolution, init_data_size, ess_min, n_processes, max_iter)
108-
# gt.plot_2d_scatter(grid,
109-
# f"initial grid: {data_size} data points, {np.shape(grid)[0]} grid point",
110-
# r'$x_1$',
111-
# r'$x_2$',
112-
# self.x_bounds[0],
113-
# self.x_bounds[1],
114-
# 5,
115-
# 0.1,
116-
# "initial_grid.png"
117-
# )
118-
# # remove low probability samples
119-
# gt.plot_2d_scatter(grid,
120-
# f"initial grid reduced: {data_size} data points, {np.shape(grid)[0]} grid point",
121-
# r'$x_1$',
122-
# r'$x_2$',
123-
# self.x_bounds[0],
124-
# self.x_bounds[1],
125-
# 5,
126-
# 0.1,
127-
# "initial_grid_reduced.png"
128-
# )
123+
grid_resolution, data_size, grid, x_spacing, log_likelihoods, rel_prob, weights, ess = self.initialize(init_grid_resolution, init_data_size, ess_min, n_processes, max_iter, store_results)
129124

130125
# iterate through remaining data (i.e. data tempering)
131126
data_tempering_index = data_size-1
132127
pbar = tqdm(range(data_tempering_index, self.n_data_points-1),desc='Processing:')
133128
for i in pbar:
129+
f_evals = 0
134130
data_tempering_index = i+1
135131
args = self.args_list[data_tempering_index]
136132
grid = gt.add_grid_points(grid, self.x_bounds, self.x_shifts, x_spacing) # expand
137133
log_likelihoods, rel_prob, weights, ess = gt.eval_grid_points(grid, self.func, args, n_processes)
138-
# gt.plot_2d_scatter(grid,
139-
# f"expanded grid: {data_tempering_index+1} data points, {np.shape(grid)[0]} grid points",
140-
# r'$x_1$',
141-
# r'$x_2$',
142-
# self.x_bounds[0],
143-
# self.x_bounds[1],
144-
# 5,
145-
# 0.1,
146-
# f"expanded_grid_{i}.png"
147-
# )
134+
f_evals = f_evals + np.size(log_likelihoods)
148135

149136
# ensure ess is high enough for added datapoint, if not, make grid finer
150137
iter = 0
@@ -153,34 +140,20 @@ def initialize_and_sample(self, init_grid_resolution, init_data_size, ess_min, d
153140
x_spacing = gt.update_x_spacing(x_spacing,2) # make grid spacing finer by 2x
154141
grid = gt.add_grid_points(grid, self.x_bounds, self.x_shifts, x_spacing, prev_x_spacing) # expand and pack
155142
log_likelihoods, rel_prob, weights, ess = gt.eval_grid_points(grid, self.func, args, n_processes)
156-
iter = iter + 1
157-
# gt.plot_2d_scatter(grid,
158-
# f"packed grid: {data_tempering_index+1} data points, {np.shape(grid)[0]} grid points",
159-
# r'$x_1$',
160-
# r'$x_2$',
161-
# self.x_bounds[0],
162-
# self.x_bounds[1],
163-
# 5,
164-
# 0.1,
165-
# f"packed_grid_{i}_{iter}.png"
166-
# )
143+
iter = iter + 1
144+
f_evals = f_evals + np.size(log_likelihoods)
167145

168146
grid = gt.reduce_grid_points(grid,weights,delta)
169-
# gt.plot_2d_scatter(grid,
170-
# f"reduced grid: {data_tempering_index+1} data points, {np.shape(grid)[0]} grid points",
171-
# r'$x_1$',
172-
# r'$x_2$',
173-
# self.x_bounds[0],
174-
# self.x_bounds[1],
175-
# 5,
176-
# 0.1,
177-
# f"reduced_grid_{i}.png"
178-
# )
179-
# self.output.append(grid)
180-
if store_grid:
181-
self.output.append(grid)
182-
183-
pbar.set_description(f"Processing: data_size={data_tempering_index+1}, n_grid_points={np.shape(grid)[0]}, ESS={ess}")
147+
148+
if store_results:
149+
self.grid_results.append(grid)
150+
self.x_spacing_results.append(x_spacing)
151+
self.log_like_results.append(log_likelihoods)
152+
self.ess_results.append(ess)
153+
self.func_evals.append(f_evals)
154+
self.grid_size.append(np.shape(grid)[0])
155+
156+
pbar.set_description(f"Processing: data_size={data_tempering_index+1}, n_grid_points={np.shape(grid)[0]}, ESS={ess}, func_evals:{f_evals}")
184157
return grid_resolution, data_size, grid, x_spacing, log_likelihoods, rel_prob, weights, ess
185158

186159

0 commit comments

Comments
 (0)