python类argmin()的实例源码

def multilabel_image_to_class(label_image: tf.Tensor, classes_file: str) -> tf.Tensor:
    classes_color_values, colors_labels = get_classes_color_from_file_multilabel(classes_file)
    # Convert label_image [H,W,3] to the classes [H,W,C],int32 according to the classes [C,3]
    with tf.name_scope('LabelAssign'):
        if len(label_image.get_shape()) == 3:
            diff = tf.cast(label_image[:, :, None, :], tf.float32) - tf.constant(classes_color_values[None, None, :, :])  # [H,W,C,3]
        elif len(label_image.get_shape()) == 4:
            diff = tf.cast(label_image[:, :, :, None, :], tf.float32) - tf.constant(
                classes_color_values[None, None, None, :, :])  # [B,H,W,C,3]
        else:
            raise NotImplementedError('Length is : {}'.format(len(label_image.get_shape())))

        pixel_class_diff = tf.reduce_sum(tf.square(diff), axis=-1)  # [H,W,C] or [B,H,W,C]
        class_label = tf.argmin(pixel_class_diff, axis=-1)  # [H,W] or [B,H,W]

        return tf.gather(colors_labels, class_label) > 0

def build_model_graph(self): 
        with tf.variable_scope(self.name) as self.scope:
            self.input_place_holder = tf.placeholder(tf.float32, shape=(None, self.params.window, self.params.ob_size * 4 + 2), name='input')
            self.forward_cell_layers = tf.contrib.rnn.MultiRNNCell([tf.contrib.rnn.LSTMCell(self.params.hidden_size) for i in range(self.params.hidden_depth)])
            self.rnn_output, self.final_rnn_state = tf.nn.dynamic_rnn(self.forward_cell_layers, self.input_place_holder, dtype=tf.float32)
            self.outs = tf.squeeze(tf.slice(self.rnn_output, [0, self.params.window - 1, 0], [-1, 1, self.params.hidden_size]), axis=1)

            if not self.advantage:
                self.U = tf.get_variable('U', shape=[self.params.hidden_size, self.params.actions])
                self.b_2 = tf.get_variable('b2', shape=[self.params.actions])
                self.predictions = tf.cast((tf.matmul(self.outs, self.U) + self.b_2), 'float32') 
            else:
                self.advantage_stream, self.value_stream = tf.split(self.outs, 2, 1)
                self.U_a = tf.get_variable('U_a', shape=[self.params.hidden_size//2, self.params.actions])
                self.U_v = tf.get_variable('U_v', shape=[self.params.hidden_size//2, 1])
                self.A =  tf.cast(tf.matmul(self.advantage_stream, self.U_a), 'float32') 
                self.V = tf.cast(tf.matmul(self.value_stream, self.U_v), 'float32') 
                self.predictions = self.V + tf.subtract(self.A, tf.reduce_mean(self.A, axis=1, keep_dims=True))
            self.min_score = tf.reduce_min(self.predictions, reduction_indices=[1])
            self.min_action = tf.argmin(tf.squeeze(self.predictions), axis=0, name="arg_min")

def _setup_enqueuing(self, queues, **loom_kwargs):
    """Sets up enqueuing to the approx. smallest (least full) of `queues`."""
    self.compiler.init_loom(loom_input_tensor=None, **loom_kwargs)
    input_tensor = self.compiler.loom_input_tensor
    fns = [lambda r=q: r.enqueue_many([input_tensor]) for q in queues]
    self.train_op = _tf_nth(fns, tf.argmin(_noised_q_sizes(queues), axis=0))
    self.losses.clear()
    self.losses['dummy'] = tf.constant(0.0)
    self.save_summaries_secs = 0
    self.dev_examples = None
    self.train_feeds.clear()
    self.save_model_secs = 0
    self.exact_batch_sizes = True

def argmin(x, axis=-1):
    '''Returns the index of the minimum value
    along a tensor axis.
    '''
    if axis < 0:
        axis = axis % len(x.get_shape())
    return tf.argmin(x, axis)

def label_image_to_class(label_image: tf.Tensor, classes_file: str) -> tf.Tensor:
    classes_color_values = get_classes_color_from_file(classes_file)
    # Convert label_image [H,W,3] to the classes [H,W],int32 according to the classes [C,3]
    with tf.name_scope('LabelAssign'):
        if len(label_image.get_shape()) == 3:
            diff = tf.cast(label_image[:, :, None, :], tf.float32) - tf.constant(classes_color_values[None, None, :, :])  # [H,W,C,3]
        elif len(label_image.get_shape()) == 4:
            diff = tf.cast(label_image[:, :, :, None, :], tf.float32) - tf.constant(
                classes_color_values[None, None, None, :, :])  # [B,H,W,C,3]
        else:
            raise NotImplementedError('Length is : {}'.format(len(label_image.get_shape())))

        pixel_class_diff = tf.reduce_sum(tf.square(diff), axis=-1)  # [H,W,C] or [B,H,W,C]
        class_label = tf.argmin(pixel_class_diff, axis=-1)  # [H,W] or [B,H,W]
        return class_label

def argmin(self, x, axis=-1):
        '''Returns the index of the minimum value
        along a tensor axis.
        '''
        if axis < 0:
            axis = axis % len(x.get_shape())
        return tf.argmin(x, axis)

def argmin(x, axis=-1):
    """Returns the index of the minimum value along an axis.

    # Arguments
        x: input tensor.
        axis: axis along which to perform the reduction.
        keepdims: whether the drop or broadcast the reduction axes.

    # Returns
        A tensor.
    """
    if axis < 0:
        axis = axis % len(x.get_shape())
    return tf.argmin(x, axis)

def evaluation(loss, batch_size=100):
  """Evaluate the quality of the logits at predicting the label.
  """
  ruler = tf.constant(0.5, shape=[batch_size, 1])
  loss_l = tf.reshape(loss, [batch_size, 1])
  comp = tf.concat(1, [ruler, loss_l])
  correct = tf.argmin(comp, 1)
  # Return the number of entries less than 0.5
  return tf.reduce_sum(correct)

def argmin(x, axis=-1):
    '''Returns the index of the minimum value
    along a tensor axis.
    '''
    if axis < 0:
        axis = axis % len(x.get_shape())
    return tf.argmin(x, axis)

def argmin(self, axis=None, name='argmin'):
        return as_varnode(tf.argmin(self, axis=axis, name=name))

def argmin(x, axis=-1):
    """Returns the index of the minimum value along an axis.

    # Arguments
        x: input tensor.
        axis: axis along which to perform the reduction.
        keepdims: whether the drop or broadcast the reduction axes.

    # Returns
        A tensor.
    """
    if axis < 0:
        axis = axis % len(x.get_shape())
    return tf.argmin(x, axis)

def argmin(x, axis=-1):
    """Returns the index of the minimum value along an axis.

    # Arguments
        x: Tensor or variable.
        axis: axis along which to perform the reduction.

    # Returns
        A tensor.
    """
    axis = _normalize_axis(axis, ndim(x))
    return tf.argmin(x, axis)

def match_to_dict_conv(image_as_patches, dictionary, include_counts=False):
    print 'match_to_dict_conv'
    [n,w,h,c] = dictionary.get_shape().as_list()
    #dict_as_filt = tf.transpose(tf.reshape(dictionary, [-1, w*h*c,1,1]))
    dict_as_filt = tf.transpose(tf.reshape(dictionary, [-1, w*h*c]))
    print dict_as_filt.get_shape()

    [n,w,h,c] = image_as_patches.get_shape().as_list()
    #image_flattened = tf.reshape(image_as_patches, [-1,1,1,w*h*c])
    image_flattened = tf.reshape(image_as_patches, [-1,w*h*c])
    print image_flattened.get_shape()

    #pair_dist = -2 * tf.reshape(tf.nn.conv2d(image_flattened, dict_as_filt, [1,1,1,1], 'SAME'), [n, -1])
    pair_dist = -2 * tf.matmul(image_flattened, dict_as_filt)
    print pair_dist.get_shape()

    single_dist = tf.reduce_sum(tf.square(dictionary),[1,2,3])
    distance = single_dist + pair_dist
    print distance.get_shape()

    min_loc = tf.argmin(distance,1)
    print min_loc.get_shape()

    if include_counts:
        y, _, count = tf.unique_with_counts(min_loc)
        return tf.gather(dictionary, min_loc), [y, count]
    else:
        return tf.gather(dictionary, min_loc)

def match_to_dict(image_as_patches, dictionary):
    patch_size = len(image_as_patches.get_shape())+1

    distance = tf.reduce_sum(tf.square(tf.expand_dims(image_as_patches,1) - tf.expand_dims(dictionary,0)), range(2,patch_size))
    min_loc = tf.argmin(distance, 1)
    return tf.gather(dictionary, min_loc)

def argmin(x, axis=-1, keepdims=False):
    return tf.argmin(x, axis=axis)

def assign_label(label, x, cluster_center):
    """ Assign Labels

    Input:
        x: embedding of size N x D
        label: cluster label of size N X 1
        K: number of clusters
        tf_eps: small constant

    Output:
        cluster_center: cluster center of size K x D
    """

    dist = pdist(x, cluster_center)
    return label.assign(tf.argmin(dist, 1))

def argmin(x, axis=-1):
    '''Returns the index of the minimum value
    along a tensor axis.
    '''
    if axis < 0:
        axis = axis % len(x.get_shape())
    return tf.argmin(x, axis)

def test_ArgMin(self):
        t = tf.argmin(self.random(3, 4, 2), 1)
        self.check(t)

def argmin(x, axis=-1):
    if axis < 0:
        axis = axis % len(x.get_shape())
    return tf.argmin(x, axis)

def kMeans(iterations, labelledSet, columnPrefix="cluster"):
    X = labelledSet.as_matrix()

    start_pos = tf.Variable(X[np.random.randint(X.shape[0], size=iterations),:], dtype=tf.float32)
    centroids = tf.Variable(start_pos.initialized_value(), "S", dtype=tf.float32)

    points           = tf.Variable(X, 'X', dtype=tf.float32)
    ones_like        = tf.ones((points.get_shape()[0], 1))
    prev_assignments = tf.Variable(tf.zeros((points.get_shape()[0], ), dtype=tf.int64))

    p1 = tf.matmul(
        tf.expand_dims(tf.reduce_sum(tf.square(points), 1), 1),
        tf.ones(shape=(1, iterations))
    )
    p2 = tf.transpose(tf.matmul(
        tf.reshape(tf.reduce_sum(tf.square(centroids), 1), shape=[-1, 1]),
        ones_like,
        transpose_b=True
    ))
    distance = tf.sqrt(tf.add(p1, p2) - 2 * tf.matmul(points, centroids, transpose_b=True))

    point_to_centroid_assignment = tf.argmin(distance, axis=1)

    total = tf.unsorted_segment_sum(points, point_to_centroid_assignment, iterations)
    count = tf.unsorted_segment_sum(ones_like, point_to_centroid_assignment, iterations)
    means = total / count

    is_continue = tf.reduce_any(tf.not_equal(point_to_centroid_assignment, prev_assignments))

    with tf.control_dependencies([is_continue]):
        loop = tf.group(centroids.assign(means), prev_assignments.assign(point_to_centroid_assignment))

    sess = tf.Session()
    sess.run(tf.global_variables_initializer())

    has_changed, cnt = True, 0
    while has_changed and cnt < 300:
        cnt += 1
        has_changed, _ = sess.run([is_continue, loop])

    res = sess.run(point_to_centroid_assignment)

    return pandas.DataFrame(res, columns=[columnPrefix + "_" + str(iterations)])

def _deepfoolx(model, x, epochs, eta, clip_min, clip_max, min_prob):
    y0 = tf.stop_gradient(model(x))
    y0 = tf.reshape(y0, [-1])
    k0 = tf.argmax(y0)

    ydim = y0.get_shape().as_list()[0]
    xdim = x.get_shape().as_list()[1:]
    xflat = _prod(xdim)

    def _cond(i, z):
        xadv = tf.clip_by_value(x + z*(1+eta), clip_min, clip_max)
        y = tf.reshape(model(xadv), [-1])
        p = tf.reduce_max(y)
        k = tf.argmax(y)
        return tf.logical_and(tf.less(i, epochs),
                              tf.logical_or(tf.equal(k0, k),
                                            tf.less(p, min_prob)))

    def _body(i, z):
        xadv = tf.clip_by_value(x + z*(1+eta), clip_min, clip_max)
        y = tf.reshape(model(xadv), [-1])

        gs = [tf.reshape(tf.gradients(y[i], xadv)[0], [-1])
              for i in range(ydim)]
        g = tf.stack(gs, axis=0)

        yk, yo = y[k0], tf.concat((y[:k0], y[(k0+1):]), axis=0)
        gk, go = g[k0], tf.concat((g[:k0], g[(k0+1):]), axis=0)

        yo.set_shape(ydim - 1)
        go.set_shape([ydim - 1, xflat])

        a = tf.abs(yo - yk)
        b = go - gk
        c = tf.norm(b, axis=1)
        score = a / c
        ind = tf.argmin(score)

        si, bi = score[ind], b[ind]
        dx = si * bi
        dx = tf.reshape(dx, [-1] + xdim)
        return i+1, z+dx

    _, noise = tf.while_loop(_cond, _body, [0, tf.zeros_like(x)],
                             name='_deepfoolx_impl', back_prop=False)
    return noise

def pit_mse_loss(s_x, s_y, pit_axis=1, perm_size=None, name='pit_loss'):
    '''
    Permutation invariant MSE loss, batched version

    Args:
        s_x: tensor
        s_y: tensor
        pit_axis: which axis permutations occur
        perm_size: size of permutation, infer from tensor shape by default
        name: string

    Returns:
        s_loss, v_perms, s_loss_sets_idx

        s_loss: scalar loss
        v_perms: constant int matrix of permutations
        s_perm_sets_idx: int matrix, indicating selected permutations

    '''
    x_shp = s_x.get_shape().as_list()
    ndim = len(x_shp)

    batch_size = x_shp[0]
    if batch_size is None:
        batch_size = hparams.BATCH_SIZE

    assert -ndim <= pit_axis < ndim
    pit_axis %= ndim
    assert pit_axis != 0
    reduce_axes = [
        i for i in range(1, ndim+1) if i not in [pit_axis, pit_axis+1]]
    with tf.variable_scope(name):
        v_perms = tf.constant(
            list(itertools.permutations(range(hparams.MAX_N_SIGNAL))),
            dtype=hparams.INTX)
        s_perms_onehot = tf.one_hot(
            v_perms, hparams.MAX_N_SIGNAL, dtype=hparams.FLOATX)

        s_x = tf.expand_dims(s_x, pit_axis+1)
        s_y = tf.expand_dims(s_y, pit_axis)
        if s_x.dtype.is_complex and s_y.dtype.is_complex:
            s_diff = s_x - s_y
            s_cross_loss = tf.reduce_mean(
                tf.square(tf.real(s_diff)) + tf.square(tf.imag(s_diff)),
                reduce_axes)
        else:
            s_cross_loss = tf.reduce_mean(
                tf.squared_difference(s_x, s_y), reduce_axes)
        s_loss_sets = tf.einsum(
            'bij,pij->bp', s_cross_loss, s_perms_onehot)
        s_loss_sets_idx = tf.argmin(s_loss_sets, axis=1)
        s_loss = tf.gather_nd(
            s_loss_sets,
            tf.stack([
                tf.range(hparams.BATCH_SIZE, dtype=tf.int64),
                s_loss_sets_idx], axis=1))
        s_loss = tf.reduce_mean(s_loss)
    return s_loss, v_perms, s_loss_sets_idx

def __call__(self, s_embed, s_src_pwr=None, s_mix_pwr=None, s_embed_flat=None):
        with tf.variable_scope(self.name):
            v_anchors = tf.get_variable(
                'anchors', [hparams.NUM_ANCHOR, hparams.EMBED_SIZE],
                initializer=tf.random_normal_initializer(
                    stddev=1.))

            # all combinations of anchors
            s_anchor_sets = ops.combinations(
                v_anchors, hparams.MAX_N_SIGNAL)

            # equation (6)
            s_anchor_assignment = tf.einsum(
                'btfe,pce->bptfc',
                s_embed, s_anchor_sets)
            s_anchor_assignment = tf.nn.softmax(s_anchor_assignment)

            # equation (7)
            s_attractor_sets = tf.einsum(
                'bptfc,btfe->bpce',
                s_anchor_assignment, s_embed)
            s_attractor_sets /= tf.expand_dims(
                tf.reduce_sum(s_anchor_assignment, axis=(2,3)), -1)

            # equation (8)
            s_in_set_similarities = tf.reduce_max(
                tf.matmul(
                    s_attractor_sets,
                    tf.transpose(s_attractor_sets, [0, 1, 3, 2])),
                axis=(-1, -2))

            # equation (9)
            s_subset_choice = tf.argmin(s_in_set_similarities, axis=1)
            s_subset_choice = tf.transpose(tf.stack([
                tf.range(hparams.BATCH_SIZE, dtype=tf.int64),
                s_subset_choice]))
            s_attractors = tf.gather_nd(s_attractor_sets, s_subset_choice)

        if hparams.DEBUG:
            self.debug_fetches = dict(
                asets=s_attractor_sets,
                anchors=v_anchors,
                subset_choice=s_subset_choice)

        return s_attractors

def _construct_sequence(batch):
        hidden, boxes = batch
        # initializing the state with features
        states = [hidden[0]]
        # TODO: make this dependent on the data
        # TODO: make it with scan ?
        for t in range(1, T):
          # find the matching boxes. TODO: try with the soft matching function
          if c.match_kind == 'boxes':
            dists = nnutil.cdist(boxes[t-1], boxes[t])
            idxs = tf.argmin(dists, 1, 'idxs')
            state_prev = tf.gather(states[t-1], idxs)
          elif c.match_kind == 'hidden':
            # TODO: actually it makes more sense to compare on states
            dists = nnutil.cdist(hidden[t-1], hidden[t])
            idxs = tf.argmin(dists, 1, 'idxs')
            state_prev = tf.gather(states[t-1], idxs)
          elif c.match_kind == 'hidden-soft':
            dists = nnutil.cdist(hidden[t-1], hidden[t])
            weights = slim.softmax(dists)
            state_prev = tf.matmul(weights, states[t-1])
          else:
            raise RuntimeError('Unknown match_kind: %s' % c.match_kind)

          def _construct_update(reuse):
            state = tf.concat(1, [state_prev, hidden[t]])
            # TODO: initialize jointly
            reset = slim.fully_connected(state, NFH, tf.nn.sigmoid,
                                         reuse=reuse,
                                         scope='reset')
            step = slim.fully_connected(state, NFH, tf.nn.sigmoid,
                                        reuse=reuse,
                                        scope='step')
            state_r = tf.concat(1, [reset * state_prev, hidden[t]])
            state_up = slim.fully_connected(state_r, NFH, tf.nn.tanh,
                                            reuse=reuse,
                                            scope='state_up')
            return state_up, step
          try:
            state_up, step = _construct_update(reuse=True)
          except ValueError:
            state_up, step = _construct_update(reuse=False)

          state = step * state_up + (1.0 - step) * state_prev
          states.append(state)
        return tf.pack(states)

def build_model_graph(self):
        self.filter_tensors = {}
        self.bias_tensors = {}
        # lots to decisions
        with tf.variable_scope(self.name) as self.scope:
            self.input_place_holder = tf.placeholder(tf.float32, shape=(None, self.params.window, self.params.ob_size * 4 + 2), name='input')
            curr_dimension = [tf.shape(self.input_place_holder)[0], self.params.window, self.params.ob_size * 4 + 2, 1]
            curr_layer = tf.reshape(self.input_place_holder, curr_dimension)
            for name, layer_params in sorted(self.layers.items()):
                print curr_dimension
                print curr_layer 
                if layer_params['type'] == 'conv':
                    n = 'conv_{}_filter_size_{}_stride_{}_num_{}'.format(name, layer_params['size'], layer_params['stride'], layer_params['num'])
                    s = [layer_params['size'], layer_params['size'], curr_dimension[3], layer_params['num']]
                    strides = [1, layer_params['stride'], layer_params['stride'], 1]
                    self.filter_tensors[name] = tf.Variable(tf.truncated_normal(s, stddev=0.0001), name=n)
                    self.bias_tensors[name] = tf.Variable(tf.truncated_normal(shape=[layer_params['num']], stddev=0.1), name=n + '_bias')
                    conv_output = tf.nn.conv2d(curr_layer, self.filter_tensors[name], strides, "VALID")
                    conv_bias = tf.nn.bias_add(conv_output, self.bias_tensors[name])
                    curr_layer = tf.nn.relu(conv_bias)
                    curr_dimension = compute_output_size(curr_dimension[0], curr_dimension[1], curr_dimension[2],layer_params['size'], layer_params['stride'], 0, layer_params['num'])
                if layer_params['type'] == 'pool':
                    if layer_params['pool_type'] == 'max':
                        s = [1, layer_params['size'], layer_params['size'], 1]
                        stride = [1, layer_params['stride'], layer_params['stride'], 1]
                        x = tf.nn.max_pool(curr_layer, s, stride, 'VALID')
                        curr_layer = x
                        curr_dimension = compute_pool_size(curr_dimension[0], curr_dimension[1], curr_dimension[2],layer_params['size'], layer_params['stride'], curr_dimension[3])
                    if layer_params['pool_type'] == 'avg':
                        s = [1, layer_params['size'], layer_params['size'], 1]
                        stride = [1, layer_params['stride'], layer_params['stride'], 1]
                        x = tf.nn.avg_pool(curr_layer, s, stride, 'VALID')
                        curr_layer = x
                        curr_dimension = compute_pool_size(curr_dimension[0], curr_dimension[1], curr_dimension[2],layer_params['size'], layer_params['stride'], curr_dimension[3])
                if layer_params['type'] == 'fc':
                    print 'hi'
            print curr_dimension
            print curr_layer
            if not self.advantage:
                final_s = [curr_dimension[1], curr_dimension[2], curr_dimension[3],self.params.actions]
                strides = [1,1,1,1]
                projection = tf.Variable(tf.truncated_normal(final_s, stddev=0.1), name="final_projection")
                bias = tf.Variable(tf.truncated_normal([self.params.actions], stddev=0.1), name="final_projection")
                self.outs = tf.nn.conv2d(curr_layer, projection, strides, 'VALID') + bias
                self.predictions = tf.squeeze(self.outs, squeeze_dims=[1, 2])
            else:
                self.advantage_stream, self.value_stream = tf.split(curr_layer, 2, 3)
                final_s_a = [curr_dimension[1], curr_dimension[2], curr_dimension[3]/2, self.params.actions]
                final_s_v = [curr_dimension[1], curr_dimension[2], curr_dimension[3]/2, 1]
                strides = [1,1,1,1]
                self.projection_a = tf.Variable(tf.truncated_normal(final_s_a, stddev=0.01), name="final_projection")
                self.projection_v = tf.Variable(tf.truncated_normal(final_s_v, stddev=0.01), name="final_projection")
                self.A = tf.squeeze(tf.nn.conv2d(self.advantage_stream, self.projection_a, strides, 'VALID'), squeeze_dims=[1,2])
                self.V = tf.squeeze(tf.nn.conv2d(self.value_stream, self.projection_v, strides, 'VALID'), squeeze_dims=[1,2])
                self.predictions = self.V + tf.subtract(self.A, tf.reduce_mean(self.A, axis=1, keep_dims=True))
            self.min_score = tf.reduce_min(self.predictions, axis=[1])
            self.min_action = tf.argmin(tf.squeeze(self.predictions), axis=0, name="arg_min")