changing_synapses/tau_simulations.py at master · CompNeuroTchuGroup/changing_synapses · GitHub

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
import matplotlib.pyplot as plt
import numpy as np
from qtconsole.mainwindow import background

from Utils.models import *
from Utils.simulation import *
from Utils.plot_utils import *

import time

params_dict = {
    "dataset_parameters": {
        "n_samples": 200
    },
    "network_parameters": {
        "input_size": 64,
        "hidden_size": 128,
        "output_size": 1,
        "bias": 1,
    },
    "training_parameters": {
        "num_epochs": 100,
        "learning_rate": 0.01
    },
    "simulation_parameters": {
        "mu": 1,
        "sigma": 0.1,
        "theta": 0.02,
        "dt": 0.001,
        "tau": 0.005
    },
    "seed": 42
}

rng = random.key(params_dict["seed"])

mu_LN = mu_LN_from_params(**params_dict["simulation_parameters"])
sigma_LN = sigma_LN_from_params(**params_dict["simulation_parameters"])

rng, data_key = random.split(rng)

X_train, y_train = create_binary_dataset(data_key, n_samples=params_dict["dataset_parameters"]["n_samples"],
                                         input_dim=params_dict["network_parameters"]["input_size"])

tau_list = jnp.arange(0.00, 0.05, 0.005)
loss_tau = []
acc_tau = []

training_parameters = params_dict["training_parameters"]
num_epochs = training_parameters["num_epochs"]
learning_rate = training_parameters["learning_rate"]
simulation_parameters = params_dict["simulation_parameters"]

print('tau_list', tau_list)


for tau in tau_list:
    acc_exp = []
    loss_exp = []
    print("Tau: ", tau)
    for n in range(10):
        print("Experiment: ", n)
        simulation_parameters["tau"] = tau
        rng, net_key = random.split(rng)
        params = init_elm(net_key, mu_LN, sigma_LN, **params_dict["network_parameters"])
        # params = simulate_training(gou_key, params, tau, num_epochs, X_train, y_train, learning_rate, simulation_parameters)
        for epoch in range(num_epochs):
            start_time = time.time()
            for x, y in zip(X_train, y_train):
                rng, gou_key = random.split(rng)
                # perturb the weights of W_i
                params['W_i'] = time_evolution_GOU(gou_key, params['W_i'], **simulation_parameters)
                # params['W_i'] += perturb_GOU(gou_key, params['W_i'], simulation_parameters['mu'], simulation_parameters['theta'], simulation_parameters['sigma'], simulation_parameters['dt'])

                grads = grad(loss_elm)(params, x, y)
                params['W_i'] -= learning_rate * grads['W_i']
                params['W_o'] -= learning_rate * grads['W_o']
                params['b_i'] -= learning_rate * grads['b_i']
                params['b_o'] -= learning_rate * grads['b_o']

            acc_exp.append(accuracy_elm(params, X_train, y_train))
            loss_exp.append(loss_elm(params, X_train, y_train))

            if epoch % 10 == 0:
                epoch_time = time.time() - start_time
                train_loss = loss_elm(params, X_train, y_train)
                train_acc = accuracy_elm(params, X_train, y_train)
                print("Epoch {} in {:0.2f} sec".format(epoch, epoch_time))
                print("Training set loss {}".format(train_loss))
                print("Training set accuracy {}".format(train_acc))

    acc_tau.append(acc_exp)
    loss_tau.append(loss_exp)
    print("mean acc: ", np.mean(acc_exp))
    print("mean loss: ", np.mean(loss_exp))

#convert to numpy arrays
acc_tau = np.array(acc_tau)
loss_tau = np.array(loss_tau)


#save the results
np.save('old_results/acc_tau_tot.npy', acc_tau)
np.save('old_results/loss_tau_tot.npy', loss_tau)

# Plotting
#create figure
fig, ax = plt.subplots(1, 2, figsize=(10, 5))
ax[0].plot(tau_list, np.mean(acc_tau, axis=1), label='Accuracy')
ax[0].fill_between(tau_list, np.mean(acc_tau, axis=1) - np.std(acc_tau, axis=1),
                   np.mean(acc_tau, axis=1) + np.std(acc_tau, axis=1), alpha=0.3)
ax[0].set_xlabel('Tau')
ax[0].set_ylabel('Accuracy')

ax[1].plot(tau_list, np.mean(loss_tau, axis=1), label='Loss')
ax[1].fill_between(tau_list, np.mean(loss_tau, axis=1) - np.std(loss_tau, axis=1),
                   np.mean(loss_tau, axis=1) + np.std(loss_tau, axis=1), alpha=0.3)
ax[1].set_xlabel('Tau')
ax[1].set_ylabel('Loss')
plt.savefig('tau_simulations.png', dpi=300, bbox_inches='tight', transparent=True)
plt.show()