-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcond_compare_iterations_dim_grid.py
141 lines (112 loc) · 6.35 KB
/
cond_compare_iterations_dim_grid.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
import sys
sys.path.append('/home/hsw967/Programming/Hannahs-CEBRAs')
sys.path.append('/home/hsw967/Programming/Hannahs-CEBRAs/scripts')
sys.path.append('/Users/Hannah/Programming/Hannahs-CEBRAs')
from itertools import product
import numpy as np
import pandas as pd
import torch
import random
from sklearn.neighbors import KNeighborsRegressor
from sklearn.metrics import r2_score
from scipy import stats
from datetime import datetime
from cebra import CEBRA
from hold_out import hold_out
from CSUS_score import CSUS_score
import gc
#decodes conditioning in envB using envA.
#Outputs percent correct in envA after being trained in env A(based on a 75/25 split)
#Outputs percent correct in envB using the model trained in envA
#does not make figures
#use to run a bunch of times
# Example usage:
# traceA1An_An, traceAnB1_An, traceA1An_A1, traceAnB1_B1, CSUSAn, CSUSA1, CSUSB1 are to be defined or loaded before calling this function.
# results = pos_compare_iterations(traceA1An_An, traceAnB1_An, traceA1An_A1, traceAnB1_B1, CSUSAn, CSUSA1, CSUSB1)
# print(results)
# results = pos_compare_iterations(traceA1An_An, traceAnB1_An, traceA1An_A1, traceAnB1_B1, CSUSAn, CSUSA1, CSUSB1, args.iterations, args.parameter_set_name)
def train_and_evaluate(cebra_model, trace_train, trace_test, test_trace, pos_train, pos_test, test_pos):
cebra_model.fit(trace_train, pos_train)
train_transformed = cebra_model.transform(trace_train)
test_transformed = cebra_model.transform(trace_test)
test_external_transformed = cebra_model.transform(test_trace)
return CSUS_score(train_transformed, test_transformed, pos_train, pos_test), CSUS_score(train_transformed, test_external_transformed, pos_train, test_pos)
def generate_headers():
prefixes = ["A1An_held_out", "A1", "B1An_held_out", "B1", "SHUFF_A1An_held_out", "SHUFF_A1", "SHUFF_B1An_held_out", "SHUFF_B1", "output_dimension"]
metrics = ["% correct"]
headers = []
for prefix in prefixes:
for metric in metrics:
headers.append(f"{prefix}_{metric}")
return ','.join(headers)
def cond_compare_iterations_dim_grid(traceA1An_An, traceAnB1_An, traceA1An_A1, traceAnB1_B1, CSUSAn, CSUSA1, CSUSB1, iterations, parameter_set, dimensions):
all_results = []
learning_rate = parameter_set["learning_rate"]
min_temperature = parameter_set["min_temperature"]
max_iterations = parameter_set["max_iterations"]
distance = parameter_set["distance"]
temp_mode = parameter_set["temp_mode"]
for dm in dimensions:
if temp_mode == 'auto':
cebra_model = CEBRA(
learning_rate=learning_rate,
max_iterations=max_iterations,
model_architecture='offset10-model',
batch_size=512,
temperature_mode='auto',
min_temperature=min_temperature,
output_dimension=dm,
distance=distance,
conditional='time_delta',
device='cuda_if_available',
num_hidden_units=32,
time_offsets=1,
verbose=False)
if temp_mode == 'constant':
cebra_model = CEBRA(
learning_rate=learning_rate,
max_iterations=max_iterations,
model_architecture='offset10-model',
batch_size=512,
temperature_mode='constant',
temperature=min_temperature,
output_dimension=dm,
distance=distance,
conditional='time_delta',
device='cuda_if_available',
num_hidden_units=32,
time_offsets=1,
verbose=False)
results = np.zeros((iterations, 9)) # Each iteration results in 8 outputs
min_length = (len(CSUSAn))
if min_length % 10 == 9:
CSUSAn = CSUSAn[9:]
traceA1An_An = traceA1An_An[9:]
traceAnB1_An = traceAnB1_An[9:]
min_length = (len(CSUSB1))
if min_length % 10 == 9:
CSUSB1 = CSUSB1[9:]
traceAnB1_B1 = traceAnB1_B1[9:]
for i in range(iterations):
traceA1An_An_train, traceA1An_An_test = hold_out(traceA1An_An, 75)
CSUSAn_train, CSUSAn_test = hold_out(CSUSAn, 75)
traceAnB1_An_train, traceAnB1_An_test = hold_out(traceAnB1_An, 75)
CSUSAnB1_train, CSUSAnB1_test = hold_out(CSUSAn, 75)
indices = np.random.permutation(CSUSAn.shape[0])
CSUSAn_shuffled = CSUSAn[indices]
CSUSAn_train_shuffled, CSUSAn_test_shuffled = hold_out(CSUSAn_shuffled, 75)
regular_A1 = train_and_evaluate(cebra_model, traceA1An_An_train, traceA1An_An_test, traceA1An_A1, CSUSAn_train, CSUSAn_test, CSUSA1)
regular_B1 = train_and_evaluate(cebra_model, traceAnB1_An_train, traceAnB1_An_test, traceAnB1_B1, CSUSAn_train, CSUSAn_test, CSUSB1)
shuffled_A1 = train_and_evaluate(cebra_model, traceA1An_An_train, traceA1An_An_test, traceA1An_A1, CSUSAn_train_shuffled, CSUSAn_test_shuffled, CSUSA1)
shuffled_B1 = train_and_evaluate(cebra_model, traceAnB1_An_train, traceAnB1_An_test, traceAnB1_B1, CSUSAn_train_shuffled, CSUSAn_test_shuffled, CSUSB1)
# Combine results into a single row
result_row = np.concatenate((np.ravel(regular_A1), np.ravel(regular_B1), np.ravel(shuffled_A1), np.ravel(shuffled_B1), [dm]))
all_results.append(result_row) # Append the row to the all_results list
# Outside the loop: save all results to a CSV file
all_results_array = np.array(all_results)
current_time = datetime.now().strftime('%Y-%m-%d_%H-%M-%S')
filename = f"cond_lr{learning_rate}_mt{min_temperature}_mi{max_iterations}_d{distance}_mode{temp_mode}_{current_time}_DIM_GRID.csv"
header = generate_headers()
np.savetxt(filename, all_results_array, delimiter=',', header=header, comments='', fmt='%.3f') # Adjust precision as needed
print(f"Results saved to {filename}")
return all_results_array