Python numpy.isneginf() Examples

def scale_neg_1_to_1_with_zero_mean_log_abs_max(v):
	'''
	!!! not working
	'''
	df = pd.DataFrame({'v':v,
	                   'sign': (v > 0) * 2 - 1})
	df['lg'] = np.log(np.abs(v)) / np.log(1.96)
	df['exclude'] = (np.isinf(df.lg) | np.isneginf(df.lg))
	for mask in [(df['sign'] == -1) & (df['exclude'] == False),
	             (df['sign'] == 1) & (df['exclude'] == False)]:
		df[mask]['lg'] = df[mask]['lg'].max() - df[mask]['lg']
	df['lg'] *= df['sign']
	df['lg'] = df['lg'].fillna(0)
	print(df[df['exclude']]['lg'].values)
	#to_rescale = convention_df['lg'].reindex(v.index)
	df['to_out'] =  scale_neg_1_to_1_with_zero_mean_abs_max(df['lg'])
	print('right')
	print(df.sort_values(by='lg').iloc[:5])
	print(df.sort_values(by='lg').iloc[-5:])
	print('to_out')
	print(df.sort_values(by='to_out').iloc[:5])
	print(df.sort_values(by='to_out').iloc[-5:])
	print(len(df), len(df.dropna()))
	return df['to_out'] 

def isneginf(x, out=None):
    """
    Test element-wise for negative infinity, return result as sparse ``bool`` array.

    Parameters
    ----------
    x
        Input
    out, optional
        Output array

    Examples
    --------
    >>> import sparse
    >>> x = sparse.as_coo(np.array([-np.inf]))
    >>> sparse.isneginf(x).todense()
    array([ True])

    See Also
    --------
    numpy.isneginf : The NumPy equivalent
    """
    from .core import elemwise

    return elemwise(lambda x, out=None, dtype=None: np.isneginf(x, out=out), x, out=out) 

def load_covarep(truth_dict):
    for video_index in truth_dict:
        file_name = covarep_path + video_index + '.mat'
        fts = sio.loadmat(file_name)['features']
        #print fts.shape
        for seg_index in truth_dict[video_index]:
            for w in truth_dict[video_index][seg_index]['data']:
                start_frame = int(w['start_time_clip']*100)
                end_frame = int(w['end_time_clip']*100)
                ft = fts[start_frame:end_frame]
                if ft.shape[0] == 0:
                    avg_ft = np.zeros(ft.shape[1])
                else:
                    #print np.array(ft).shape
                    #print ft[0]
                    avg_ft = np.mean(ft,0)
                avg_ft[np.isnan(avg_ft)] = 0
                avg_ft[np.isneginf(avg_ft)] = 0
                w['covarep'] = avg_ft 

def fill_inf(arr, pos_value=0, neg_value=0, copy=True):
    """Replaces positive and negative infinity entries in an array with the
    provided values.

    Parameters
    ----------
    arr : np.array

    pos_value : float
        Fill value for np.inf

    neg_value : float
        Fill value for -np.inf

    copy : bool, optional
        If True, creates a copy of x, otherwise replaces values in-place.
        By default, True.

    """
    if copy:
        arr = arr.copy()
    arr[np.isposinf(arr)] = pos_value
    arr[np.isneginf(arr)] = neg_value
    return arr 

def PPMI_matrix(M):

	M = scale_sim_mat(M)
	nm_nodes = len(M)

	col_s = np.sum(M, axis=0).reshape(1,nm_nodes)
	row_s = np.sum(M, axis=1).reshape(nm_nodes,1)
	D = np.sum(col_s)
	rowcol_s = np.dot(row_s,col_s)
	PPMI = np.log(np.divide(D*M,rowcol_s))
	PPMI[np.isnan(PPMI)] = 0.0
	PPMI[np.isinf(PPMI)] = 0.0
	PPMI[np.isneginf(PPMI)] = 0.0
	PPMI[PPMI<0] = 0.0

	return PPMI 

def load_covarep(truth_dict):
    for video_index in truth_dict:
        file_name = covarep_path + video_index + '.mat'
        fts = sio.loadmat(file_name)['features']
        #print fts.shape
        for seg_index in truth_dict[video_index]:
            for w in truth_dict[video_index][seg_index]['data']:
                start_frame = int(w['start_time_clip']*100)
                end_frame = int(w['end_time_clip']*100)
                ft = fts[start_frame:end_frame]
                if ft.shape[0] == 0:
                    avg_ft = np.zeros(ft.shape[1])
                else:
                    #print np.array(ft).shape
                    #print ft[0]
                    avg_ft = np.mean(ft,0)
                avg_ft[np.isnan(avg_ft)] = 0
                avg_ft[np.isneginf(avg_ft)] = 0
                w['covarep'] = avg_ft 

def test_is_inf(self):
    if legacy_opset_pre_ver(10):
      raise unittest.SkipTest("ONNX version {} doesn't support IsInf.".format(
          defs.onnx_opset_version()))
    input = np.array([-1.2, np.nan, np.inf, 2.8, np.NINF, np.inf],
                     dtype=np.float32)
    expected_output = {
        "node_def": np.isinf(input),
        "node_def_neg_false": np.isposinf(input),
        "node_def_pos_false": np.isneginf(input)
    }
    node_defs = {
        "node_def":
            helper.make_node("IsInf", ["X"], ["Y"]),
        "node_def_neg_false":
            helper.make_node("IsInf", ["X"], ["Y"], detect_negative=0),
        "node_def_pos_false":
            helper.make_node("IsInf", ["X"], ["Y"], detect_positive=0)
    }
    for key in node_defs:
      output = run_node(node_defs[key], [input])
      np.testing.assert_equal(output["Y"], expected_output[key]) 

def jsonify_floats(json_object):
    """
    Traverses through the JSON object and converts non JSON-spec compliant
    floats(nan, -inf, inf) to their string representations.

    Parameters
    ----------
    json_object
        JSON object
    """
    if isinstance(json_object, dict):
        return {k: jsonify_floats(v) for k, v in json_object.items()}
    elif isinstance(json_object, list):
        return [jsonify_floats(item) for item in json_object]
    elif isinstance(json_object, float):
        if np.isnan(json_object):
            return "NaN"
        elif np.isposinf(json_object):
            return "Infinity"
        elif np.isneginf(json_object):
            return "-Infinity"
        return json_object
    return json_object 

def output(self, value, mask):
        if mask:
            return self._null_output
        if np.isfinite(value):
            if not np.isscalar(value):
                value = value.dtype.type(value)
            result = self._output_format.format(value)
            if result.startswith('array'):
                raise RuntimeError()
            if (self._output_format[2] == 'r' and
                result.endswith('.0')):
                result = result[:-2]
            return result
        elif np.isnan(value):
            return 'NaN'
        elif np.isposinf(value):
            return '+InF'
        elif np.isneginf(value):
            return '-InF'
        # Should never raise
        vo_raise(f"Invalid floating point value '{value}'") 

def is_invalid(arr):
    return any([f(arr).any() for f in [numpy.isinf, numpy.isnan, numpy.isneginf]]) 

def get_ppmi_matrix(self, mat):
        # Get Positive Pairwise Mutual Information(PPMI) matrix
        mat = self.random_surfing(mat)
        M = self.scale_matrix(mat)
        col_s = np.sum(M, axis=0).reshape(1, self.num_node)
        row_s = np.sum(M, axis=1).reshape(self.num_node, 1)
        D = np.sum(col_s)
        rowcol_s = np.dot(row_s, col_s)
        PPMI = np.log(np.divide(D * M, rowcol_s))
        
        PPMI[np.isnan(PPMI)] = 0.0
        PPMI[np.isinf(PPMI)] = 0.0
        PPMI[np.isneginf(PPMI)] = 0.0
        PPMI[PPMI < 0] = 0.0
        return PPMI 

def tolerant_eq(a, b):
    return (True
            if (numpy.isinf(a) and numpy.isinf(b))
            or (numpy.isneginf(a) and numpy.isneginf(b))
            or abs(a-b) < 1e-10
            else False) 

def _get_sum_starting_info(limits):
    if np.isinf(limits[0]) and np.isinf(limits[1]):
        raise ValueError("Cannot calculate a sum on infinite terms.")
    if np.isposinf(limits[0]):
        starting_value = limits[1]
        stop_value = limits[0]
        method = "increasing"
    elif np.isposinf(limits[1]):
        starting_value = limits[0]
        stop_value = limits[1]
        method = "increasing"
    elif np.isneginf(limits[0]):
        starting_value = limits[1]
        stop_value = limits[0]
        method = "decreasing"
    elif np.isneginf(limits[1]):
        starting_value = limits[0]
        stop_value = limits[1]
        method = "decreasing"
    elif np.greater(limits[1], limits[0]):
        starting_value = limits[0]
        stop_value = limits[1]
        method = "increasing"
    else:
        starting_value = limits[1]
        stop_value = limits[0]
        method = "increasing"
    return method, starting_value, stop_value 

def min(self):
        """
        Calculate minimum value that can be sampled in the quanitzed version of the distribution.

        :return: minimal value return from distribution.
        """
        hd_min = self.hd.min()
        if np.isneginf(hd_min):
            return hd_min
        return round(hd_min) 

def test_isneginf(self):
        self.check(np.isneginf) 

def get_mean_var(X, ignore_zeros=False, perc=None):
    data = X.data if issparse(X) else X
    mask_nans = np.isnan(data) | np.isinf(data) | np.isneginf(data)

    n_nonzeros = (X != 0).sum(0)
    n_counts = n_nonzeros if ignore_zeros else X.shape[0]

    if mask_nans.sum() > 0:
        if issparse(X):
            data[np.isnan(data) | np.isinf(data) | np.isneginf(data)] = 0
            n_nans = n_nonzeros - (X != 0).sum(0)
        else:
            X[mask_nans] = 0
            n_nans = mask_nans.sum(0)
        n_counts -= n_nans

    if perc is not None:
        if np.size(perc) < 2:
            perc = [perc, 100] if perc < 50 else [0, perc]
        lb, ub = np.percentile(data, perc)
        data = np.clip(data, lb, ub)

    if issparse(X):
        mean = (X.sum(0) / n_counts).A1
        mean_sq = (X.multiply(X).sum(0) / n_counts).A1
    else:
        mean = X.sum(0) / n_counts
        mean_sq = np.multiply(X, X).sum(0) / n_counts
    n_cells = np.clip(X.shape[0], 2, None)  # to avoid division by zero
    var = (mean_sq - mean ** 2) * (n_cells / (n_cells - 1))

    mean = np.nan_to_num(mean)
    var = np.nan_to_num(var)
    return mean, var 

def get_mean_var(X, ignore_zeros=False, perc=None):
    data = X.data if issparse(X) else X
    mask_nans = np.isnan(data) | np.isinf(data) | np.isneginf(data)

    n_nonzeros = (X != 0).sum(0)
    n_counts = n_nonzeros if ignore_zeros else X.shape[0]

    if mask_nans.sum() > 0:
        if issparse(X):
            data[np.isnan(data) | np.isinf(data) | np.isneginf(data)] = 0
            n_nans = n_nonzeros - (X != 0).sum(0)
        else:
            X[mask_nans] = 0
            n_nans = mask_nans.sum(0)
        n_counts -= n_nans

    if perc is not None:
        if np.size(perc) < 2:
            perc = [perc, 100] if perc < 50 else [0, perc]
        lb, ub = np.percentile(data, perc)
        data = np.clip(data, lb, ub)

    if issparse(X):
        mean = (X.sum(0) / n_counts).A1
        mean_sq = (X.multiply(X).sum(0) / n_counts).A1
    else:
        mean = X.sum(0) / n_counts
        mean_sq = np.multiply(X, X).sum(0) / n_counts
    n_cells = np.clip(X.shape[0], 2, None)  # to avoid division by zero
    var = (mean_sq - mean ** 2) * (n_cells / (n_cells - 1))

    mean = np.nan_to_num(mean)
    var = np.nan_to_num(var)
    return mean, var 

def set_logp_to_neg_inf(X, logp, bounds):
    """Set `logp` to negative infinity when `X` is outside the allowed bounds.

    # Arguments
        X: tensorflow.Tensor
            The variable to apply the bounds to
        logp: tensorflow.Tensor
            The log probability corrosponding to `X`
        bounds: list of `Region` objects
            The regions corrosponding to allowed regions of `X`

    # Returns
        logp: tensorflow.Tensor
            The newly bounded log probability
    """
    conditions = []
    for l, u in bounds:
        lower_is_neg_inf = not isinstance(l, tf.Tensor) and np.isneginf(l)
        upper_is_pos_inf = not isinstance(u, tf.Tensor) and np.isposinf(u)

        if not lower_is_neg_inf and upper_is_pos_inf:
            conditions.append(tf.greater(X, l))
        elif lower_is_neg_inf and not upper_is_pos_inf:
            conditions.append(tf.less(X, u))
        elif not (lower_is_neg_inf or upper_is_pos_inf):
            conditions.append(tf.logical_and(tf.greater(X, l), tf.less(X, u)))

    if len(conditions) > 0:
        is_inside_bounds = conditions[0]
        for condition in conditions[1:]:
            is_inside_bounds = tf.logical_or(is_inside_bounds, condition)

        logp = tf.select(
            is_inside_bounds,
            logp,
            tf.fill(tf.shape(X), config.dtype(-np.inf))
        )

    return logp 

def sanitize_array(array, replace_nan=0.0, replace_inf=0.0, replace_neg_inf=0.0, min_value=None, max_value=None):
    array[np.isneginf(array)] = replace_neg_inf
    array[np.isinf(array)] = replace_inf
    array[np.isnan(array)] = replace_nan

    if min_value is not None or max_value is not None:
        array = np.clip(array, min_value, max_value)

    return array 

def _fill_minmax(self, X: np.ndarray):
        """
        Fill inf/-inf values in features of the array based on their min & max values.
        Compounded by the ``power`` value so long as the result doesn't exceed the
        current array's dtype's max/min. Otherwise it will use those.
        """

        # For each feature fill inf/-inf with pre-calculate fill values
        for feature_idx, (posinf_fill, neginf_fill) in enumerate(
            zip(self._posinf_fill_values, self._neginf_fill_values)
        ):
            X[:, feature_idx][np.isposinf(X[:, feature_idx])] = posinf_fill
            X[:, feature_idx][np.isneginf(X[:, feature_idx])] = neginf_fill
        return X 

def test_coint_identical_series():
    nobs = 200
    scale_e = 1
    np.random.seed(123)
    y = scale_e * np.random.randn(nobs)
    warnings.simplefilter('always', ColinearityWarning)
    with warnings.catch_warnings(record=True) as w:
        c = coint(y, y, trend="c", maxlag=0, autolag=None)
    assert_equal(len(w), 1)
    assert_equal(c[1], 0.0)
    assert_(np.isneginf(c[0])) 

def test_coint_perfect_collinearity():
    # test uses nearly perfect collinearity
    nobs = 200
    scale_e = 1
    np.random.seed(123)
    x = scale_e * np.random.randn(nobs, 2)
    y = 1 + x.sum(axis=1) + 1e-7 * np.random.randn(nobs)
    warnings.simplefilter('always', ColinearityWarning)
    with warnings.catch_warnings(record=True) as w:
        c = coint(y, x, trend="c", maxlag=0, autolag=None)
    assert_equal(c[1], 0.0)
    assert_(np.isneginf(c[0])) 

def assert_almost_equal_inf(x, y, decimal=6, msg=None):
    x = np.atleast_1d(x)
    y = np.atleast_1d(y)
    assert_equal(np.isposinf(x), np.isposinf(y))
    assert_equal(np.isneginf(x), np.isneginf(y))
    assert_equal(np.isnan(x), np.isnan(y))
    assert_almost_equal(x[np.isfinite(x)], y[np.isfinite(y)]) 

def test_infimputer_fill_values():
    """
    InfImputer when fill values are provided
    """
    base_x = np.random.random((100, 10)).astype(np.float32)

    flat_view = base_x.ravel()

    pos_inf_idxs = [1, 2, 3, 4, 5]
    neg_inf_idxs = [6, 7, 8, 9, 10]

    flat_view[pos_inf_idxs] = np.inf
    flat_view[neg_inf_idxs] = -np.inf

    # Our base x should now be littered with pos/neg inf values
    assert np.isposinf(base_x).sum() > 0, "Expected some positive infinity values here"
    assert np.isneginf(base_x).sum() > 0, "Expected some negative infinity values here"

    imputer = InfImputer(inf_fill_value=9999.0, neg_inf_fill_value=-9999.0)
    X = imputer.fit_transform(base_x)
    np.equal(
        X.ravel()[[pos_inf_idxs]], np.array([9999.0, 9999.0, 9999.0, 9999.0, 9999.0])
    )
    np.equal(
        X.ravel()[[neg_inf_idxs]],
        np.array([-9999.0, -9999.0, -9999.0, -9999.0, -9999.0]),
    ) 

def transform(self, X: Union[pd.DataFrame, np.ndarray], y=None):

        # Ensure we're dealing with numpy array if it's a dataframe or similar
        X = X.values if hasattr(X, "values") else X

        # Apply specific fill values if provided.
        if self.inf_fill_value is not None:
            X[np.isposinf(X)] = self.inf_fill_value
        if self.neg_inf_fill_value is not None:
            X[np.isneginf(X)] = self.neg_inf_fill_value

        # May still be left over infs, if only one fill value was supplied for example
        if self.strategy is not None:
            return getattr(self, f"_fill_{self.strategy}")(X)
        return X 

def _fill_extremes(self, X: np.ndarray):
        """
        Fill negative and postive infs with their dtype's min/max values
        """
        X[np.isposinf(X)] = np.finfo(X.dtype).max
        X[np.isneginf(X)] = np.finfo(X.dtype).min
        return X 

def on_iteration_end(self, iter_cnt, data):
        loss_train = 0
        if data.get("y_train_predicted") is not None:
            loss_train = self.metric(
                data.get("y_train_true"), data.get("y_train_predicted")
            )
        loss_validation = self.metric(
            data.get("y_validation_true"), data.get("y_validation_predicted")
        )

        for loss in [loss_train, loss_validation]:
            if np.isnan(loss) or np.isinf(loss) or np.isneginf(loss):
                self.learner.stop_training = True
                log.info("Terminating learning, invalid loss value") 

def _has_infs(result):
    if isinstance(result, np.ndarray):
        if result.dtype == 'f8':
            return lib.has_infs_f8(result)
        elif result.dtype == 'f4':
            return lib.has_infs_f4(result)
        return False
    return np.isinf(result) or np.isneginf(result) 

def assert_all_close(tensor_1, tensor_2, rtol=1e-4, atol=1e-4):
    tensor_1 = to_numpy(tensor_1).astype(np.float64)
    tensor_2 = to_numpy(tensor_2).astype(np.float64)
    np.testing.assert_equal(np.isposinf(tensor_1),
                            np.isposinf(tensor_2))
    np.testing.assert_equal(np.isneginf(tensor_1),
                            np.isneginf(tensor_2))
    indices = np.isfinite(tensor_1)
    if indices.sum():
        tensor_1 = tensor_1[indices]
        tensor_2 = tensor_2[indices]
        err = np.max(np.abs(tensor_1 - tensor_2))
        err_msg = "Max abs error: {0:.3g}".format(err)
        np.testing.assert_allclose(tensor_1, tensor_2, rtol=rtol, atol=atol,
                                   err_msg=err_msg) 

def isneginf(x):
  return x == array_ops.full_like(x, -np.inf)