python-mdp 3.3-1 / usr / share / pyshared / mdp / test / test_RBM.py

This file is indexed.

/usr/share/pyshared/mdp/test/test_RBM.py is in python-mdp 3.3-1.

This file is owned by root:root, with mode 0o644.

The actual contents of the file can be viewed below. 

  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
import mdp
from _tools import *

def test_RBM_sample_h():
    # number of visible and hidden units
    I, J = 2, 4

    # create RBM node
    bm = mdp.nodes.RBMNode(J, I)
    # fake training to initialize internals
    bm.train(numx.zeros((1,I)))
    # init to deterministic model
    bm.w[0,:] = [1,0,1,0]
    bm.w[1,:] = [0,1,0,1]
    bm.w *= 2e4
    bm.bv *= 0.
    bm.bh *= 0.

    # ### test 1
    v = numx.array([[0,0],[1,0],[0,1],[1,1.]])
    h = []
    for n in xrange(1000):
        prob, sample = bm.sample_h(v)
        h.append(sample)

    # check inferred probabilities
    expected_probs = numx.array([[0.5, 0.5, 0.5, 0.5],
                                  [1.0, 0.5, 1.0, 0.5],
                                  [0.5, 1.0, 0.5, 1.0],
                                  [1.0, 1.0, 1.0, 1.0]])
    assert_array_almost_equal(prob, expected_probs, 8)

    # check sampled units
    h = numx.array(h)
    for n in xrange(4):
        distr = h[:,n,:].mean(axis=0)
        assert_array_almost_equal(distr, expected_probs[n,:], 1)

    # ### test 2, with bias
    bm.bh -= 1e2
    h = []
    for n in xrange(100):
        prob, sample = bm.sample_h(v)
        h.append(sample)

    # check inferred probabilities
    expected_probs = numx.array([[0., 0., 0., 0.],
                                  [1.0, 0., 1.0, 0.],
                                  [0., 1.0, 0., 1.0],
                                  [1.0, 1.0, 1.0, 1.0]])
    assert_array_almost_equal(prob, expected_probs, 8)

    # check sampled units
    h = numx.array(h)
    for n in xrange(4):
        distr = h[:,n,:].mean(axis=0)
        assert_array_almost_equal(distr, expected_probs[n,:], 1)

def test_RBM_sample_v():
    # number of visible and hidden units
    I, J = 4, 2

    # create RBM node
    bm = mdp.nodes.RBMNode(J, I)
    # fake training to initialize internals
    bm.train(numx.zeros((1,I)))
    # init to deterministic model
    bm.w[:,0] = [1,0,1,0]
    bm.w[:,1] = [0,1,0,1]
    bm.w *= 2e4
    bm.bv *= 0
    bm.bh *= 0

    # test 1
    h = numx.array([[0,0],[1,0],[0,1],[1,1.]])
    v = []
    for n in xrange(1000):
        prob, sample = bm.sample_v(h)
        v.append(sample)

    # check inferred probabilities
    expected_probs = numx.array([[0.5, 0.5, 0.5, 0.5],
                                  [1.0, 0.5, 1.0, 0.5],
                                  [0.5, 1.0, 0.5, 1.0],
                                  [1.0, 1.0, 1.0, 1.0]])
    assert_array_almost_equal(prob, expected_probs, 8)

    # check sampled units
    v = numx.array(v)
    for n in xrange(4):
        distr = v[:,n,:].mean(axis=0)
        assert_array_almost_equal(distr, expected_probs[n,:], 1)

    # test 2, with bias
    bm.bv -= 1e2
    v = []
    for n in xrange(1000):
        prob, sample = bm.sample_v(h)
        v.append(sample)

    # check inferred probabilities
    expected_probs = numx.array([[0., 0., 0., 0.],
                                  [1.0, 0., 1.0, 0.],
                                  [0., 1.0, 0., 1.0],
                                  [1.0, 1.0, 1.0, 1.0]])
    assert_array_almost_equal(prob, expected_probs, 8)

    # check sampled units
    v = numx.array(v)
    for n in xrange(4):
        distr = v[:,n,:].mean(axis=0)
        assert_array_almost_equal(distr, expected_probs[n,:], 1)

def test_RBM_stability():
    # number of visible and hidden units
    I, J = 8, 2

    # create RBM node
    bm = mdp.nodes.RBMNode(J, I)
    bm._init_weights()
    # init to random model
    bm.w = mdp.utils.random_rot(max(I,J), dtype='d')[:I, :J]
    bm.bv = numx_rand.randn(I)
    bm.bh = numx_rand.randn(J)

    # save original weights
    real_w = bm.w.copy()
    real_bv = bm.bv.copy()
    real_bh = bm.bh.copy()

    # Gibbs sample to reach the equilibrium distribution
    N = 1e4
    v = numx_rand.randint(0,2,(N,I)).astype('d')
    for k in xrange(100):
        if k%5==0: spinner()
        p, h = bm._sample_h(v)
        p, v = bm._sample_v(h)

    # see that w remains stable after learning
    for k in xrange(100):
        if k%5==0: spinner()
        bm.train(v)
    bm.stop_training()

    assert_array_almost_equal(real_w, bm.w, 1)
    assert_array_almost_equal(real_bv, bm.bv, 1)
    assert_array_almost_equal(real_bh, bm.bh, 1)

def test_RBM_learning():
    # number of visible and hidden units
    I, J = 4, 2

    bm = mdp.nodes.RBMNode(J, I)
    bm.w = mdp.utils.random_rot(max(I,J), dtype='d')[:I, :J]

    # the observations consist of two disjunct patterns that
    # never appear together
    N = 1e4
    v = numx.zeros((N,I))
    for n in xrange(int(N)):
        r = numx_rand.random()
        if r>0.666: v[n,:] = [0,1,0,1]
        elif r>0.333: v[n,:] = [1,0,1,0]

    for k in xrange(1500):
        if k%5==0: spinner()

        if k>5:
            mom = 0.9
        else:
            mom = 0.5
        bm.train(v, epsilon=0.3, momentum=mom)
        if bm._train_err/N<0.1: break
        #print '-------', bm._train_err

    assert bm._train_err / N < 0.1

def _generate_data(bm, I, N):
    data = []
    h = numx.ones(I, dtype='d')
    for t in range(N):
        prob, v = bm._sample_v(h)
        prob, h = bm._sample_h(v)
        if (t > 500):
            data.append(v)

    return numx.asarray(data, dtype='d')

def test_RBM_bv_learning():
    # number of visible and hidden units
    I, J = 4, 4

    bm = mdp.nodes.RBMNode(J, I)
    bm._init_weights()
    # init to random biases, unit generation matrix
    bm.w = numx.eye(I, dtype='d')
    bm.bh *= 0.0
    bm.bv = numx.linspace(0.1, 0.9, I) * 5

    #### generate training data
    data = _generate_data(bm, I, 5000)

    #### learn from generated data
    train_bm = mdp.nodes.RBMNode(J, I)

    train_bm.train(data)
    train_bm.w = numx.eye(I, dtype='d')

    N = data.shape[0]
    for k in xrange(5000):
        if k%5==0: spinner()

        train_bm.train(data, epsilon=0.6, momentum=0.7)
        if abs(train_bm.bv - bm.bv).max() < 0.5: break

        # bv, bh, and w are dependent, so we need to keep one of them clamped
        train_bm.w = numx.eye(I, dtype='d')

    assert abs(train_bm.bv - bm.bv).max() < 0.5

def _test_RBM_bh_learning():
    # This one is tricky, as hidden biases are a very indirect parameter
    # of the input. We need to keep the rest of the weights clamped or there
    # would be alternative ways to explain the data

    # number of visible and hidden units
    I, J = 4, 4

    bm = mdp.nodes.RBMNode(J, I)
    bm._init_weights()
    # init to random biases, unit generation matrix
    bm.w = numx.eye(I, dtype='d')
    bm.bv *= 0.0
    bm.bh = numx.linspace(0.1, 0.9, I) * 5

    #### generate training data
    data = _generate_data(bm, I, 10000)

    #### learn from generated data
    train_bm = mdp.nodes.RBMNode(J, I)

    train_bm.train(data)
    train_bm.w = bm.w.copy()
    train_bm.bv *= 0.0

    N = data.shape[0]
    for k in xrange(5000):
        if k%5==0: spinner()

        train_bm.train(data, epsilon=3.0, momentum=0.8, update_with_ph=False)
        if abs(train_bm.bh - bm.bh).max() < 0.75: break

        # keep other weights clamped
        train_bm.w = bm.w.copy()
        train_bm.bv *= 0.0

    assert abs(train_bm.bh - bm.bh).max() < 0.75


def test_RBMWithLabelsNode():
    I, J, L = 4, 4, 2
    bm = mdp.nodes.RBMWithLabelsNode(J,L,I)
    assert bm.input_dim == I+L

    # generate input data
    N = 2500
    v = numx.zeros((2*N,I))
    l = numx.zeros((2*N,L))
    for n in xrange(N):
        r = numx_rand.random()
        if r>0.1:
            v[n,:] = [1,0,1,0]
            l[n,:] = [1,0]
    for n in xrange(N):
        r = numx_rand.random()
        if r>0.1:
            v[n,:] = [0,1,0,1]
            l[n,:] = [1,0]

    x = numx.concatenate((v, l), axis=1)
    for k in xrange(2500):
        if k%5==0: spinner()

        if k>200:
            mom = 0.9
            eps = 0.7
        else:
            mom = 0.5
            eps = 0.2
        bm.train(v, l, epsilon=eps, momentum=mom)

        ph, sh = bm._sample_h(x)
        pv, pl, sv, sl = bm._sample_v(sh, concatenate=False)

        v_train_err = float(((v-sv)**2.).sum())
        #print '-------', k, v_train_err/(2*N)
        if v_train_err / (2*N) < 0.1:
            break

    # visible units are reconstructed
    assert v_train_err / (2*N) < 0.1

    # units with 0 input have 50/50 labels
    idxzeros = v.sum(axis=1)==0
    nzeros = idxzeros.sum()
    point5 = numx.zeros((nzeros, L)) + 0.5
    assert_array_almost_equal(pl[idxzeros], point5, 2)

APT Browse - Built by Thomas Orozco.