import numpy as np
import unittest

from caffe2.proto import caffe2_pb2
from caffe2.python import core, workspace, test_util, model_helper, brew, build


@unittest.skipIf(build.CAFFE2_NO_OPERATOR_SCHEMA,
                 'Built with CAFFE2_NO_OPERATOR_SCHEMA')
class TestShapeInference(test_util.TestCase):

    def testShapeInferenceSimpleFC(self):
        m = model_helper.ModelHelper(name="test_model")

        brew.fc(m, "data", "fc1", dim_in=96, dim_out=32)
        brew.fc(m, "fc1", "fc2", dim_in=32, dim_out=55)

        for b in [0, 64]:
            (shapes, types) = workspace.InferShapesAndTypes(
                [m.param_init_net, m.net],
                {'data': [b, 96]}
            )

            self.assertEquals(shapes['data'], [b, 96])
            self.assertEquals(shapes['fc1_w'], [32, 96])
            self.assertEquals(shapes['fc1_b'], [32])
            self.assertEquals(shapes['fc1'], [b, 32])
            self.assertEquals(shapes['fc2_w'], [55, 32])
            self.assertEquals(shapes['fc2_b'], [55])
            self.assertEquals(shapes['fc2'], [b, 55])

    def testFCAxis2(self):
        model = model_helper.ModelHelper(name="test_model")
        model.net.FC(["x", "w", "b"], ["y"], axis=2)
        workspace.FeedBlob("x", np.random.rand(4, 20, 36).astype(np.float32))
        workspace.FeedBlob("w", np.random.rand(36, 36).astype(np.float32))
        workspace.FeedBlob("b", np.random.rand(36,).astype(np.float32))
        self.InferTensorRunAndCompare(model)

    def testFCTransposed(self):
        model = model_helper.ModelHelper(name="test_model")
        model.net.FCTransposed(["x", "wt", "b"], ["y"])
        workspace.FeedBlob("x", np.random.rand(20, 36).astype(np.float32))
        workspace.FeedBlob("wt", np.random.rand(36, 48).astype(np.float32))
        workspace.FeedBlob("b", np.random.rand(48,).astype(np.float32))
        self.InferTensorRunAndCompare(model)

    def testShapeInferenceSlice(self):
        model = model_helper.ModelHelper(name="test_model")
        model.net.Slice(["x"], ["y"], starts=[0, 0, 0, 0], ends=[-1, -1, -3, -1])
        workspace.FeedBlob("x", np.random.rand(64, 1, 255, 384).astype(np.float32))

        slice_starts = np.array([0, 0, 0, 0]).astype(np.int32)
        slice_ends = np.array([-1, -1, -3, -1]).astype(np.int32)
        slice_starts = model.net.GivenTensorIntFill(
            [], shape=[4], values=slice_starts)
        slice_ends = model.net.GivenTensorIntFill(
            [], shape=[4], values=slice_ends)
        model.net.Slice(["x2", slice_starts, slice_ends], ["y2"])
        workspace.FeedBlob("x2", np.random.rand(64, 1, 255, 384).astype(np.float32))

        self.InferTensorRunAndCompare(model, ["y2"])

    def testShapeInferenceDistances(self):
        model = model_helper.ModelHelper(name="test_model")
        model.net.L1Distance(["x1", "y1"], "dl1_D1")
        model.net.SquaredL2Distance(["x1", "y1"], "dl2_D1")
        model.net.CosineSimilarity(["x1", "y1"], "dcos_D1")
        model.net.DotProduct(["x1", "y1"], "ddot_D1")
        model.net.DotProductWithPadding(["x1", "y1"], "ddotpad_D1")

        model.net.L1Distance(["x2", "y2"], "dl1_D2")
        model.net.SquaredL2Distance(["x2", "y2"], "dl2_D2")
        model.net.CosineSimilarity(["x2", "y2"], "dcos_D2")
        model.net.DotProduct(["x2", "y2"], "ddot_D2")
        model.net.DotProductWithPadding(["x2", "z2"], "ddotpad_D2")

        workspace.FeedBlob("x1", np.random.rand(10).astype(np.float32))
        workspace.FeedBlob("y1", np.random.rand(10).astype(np.float32))

        workspace.FeedBlob("x2", np.random.rand(10, 5).astype(np.float32))
        workspace.FeedBlob("y2", np.random.rand(10, 5).astype(np.float32))
        workspace.FeedBlob("z2", np.random.rand(10, 4).astype(np.float32))
        self.InferTensorRunAndCompare(model)

    def testShapeInferenceReduceBackFrontX(self):
        model = model_helper.ModelHelper(name="test_model")
        model.net.ReduceBackSum(["x"], ["x_back_sum"])
        model.net.ReduceBackMean(["x"], ["x_back_mean"])
        model.net.ReduceBackMax(["x"], ["x_back_max"])
        model.net.ReduceFrontSum(["x"], ["x_front_sum"])
        model.net.ReduceFrontMean(["x"], ["x_front_mean"])
        model.net.ReduceFrontMax(["x"], ["x_front_max"])

        workspace.FeedBlob("x", np.random.rand(10, 12, 18).astype(np.float32))
        self.InferTensorRunAndCompare(model)

    def testGather(self):
        model = model_helper.ModelHelper(name="test_model")
        model.net.Gather(["X", "idx"], "Y")
        workspace.FeedBlob("X", np.random.rand(100, 4, 5).astype(np.float32))
        workspace.FeedBlob("idx", np.array([[3, 18], [99, 4], [2, 5]]).astype(np.int32))
        self.InferTensorRunAndCompare(model)

    def testShapeInferenceConvNet(self):
        model = model_helper.ModelHelper(name="convtest")
        model.NHWC2NCHW("data", "data_nchw")
        brew.conv(model, "data_nchw", 'conv1', 3, 64,
                   weight_init=("MSRAFill", {}), kernel=7,
                   stride=2, pad=3, no_bias=0)
        brew.spatial_bn(model, 'conv1', 'conv1_spatbn_relu', 64, epsilon=1e-3, is_test=False)
        brew.relu(model, 'conv1_spatbn_relu', 'conv1_spatbn_relu')
        brew.max_pool(model, 'conv1_spatbn_relu', 'pool1', kernel=3, stride=2)
        brew.fc(model, 'pool1', 'fc', dim_in=(64 * 56 * 56), dim_out=100)
        brew.dropout(model, 'fc', 'fc_drop', is_test=False)
        model.Sigmoid('fc_drop', 'fc_sigm')
        brew.softmax(model, 'fc_sigm', 'softmax')
        model.LabelCrossEntropy(['softmax', 'label'], 'xent')
        loss = model.AveragedLoss('xent', 'loss')

        model.AddGradientOperators([loss])

        LR = model.param_init_net.ConstantFill(
            [], 'LR', shape=[1], value=0.1
        )

        for param in model.GetParams():
            param_grad = model.param_to_grad[param]
            param_momentum = model.param_init_net.ConstantFill(
                [param], param + '_momentum', value=0.0
            )
            model.net.MomentumSGDUpdate(
                [param_grad, param_momentum, LR, param],
                [param_grad, param_momentum, param],
            )

        workspace.FeedBlob(
            "data",
            np.random.rand(16, 227, 227, 3).astype(np.float32),
        )
        workspace.FeedBlob(
            "label",
            (100 * np.random.rand(16)).astype(np.int32),
        )
        workspace.FeedBlob(
            "label",
            (100 * np.random.rand(16)).astype(np.int32),
        )
        # Then do automatic comparison test: run the next once to
        # initialize everything
        self.InferTensorRunAndCompare(model)

    def testShapeInferenceTranspose(self):
        model = model_helper.ModelHelper(name="test_model")

        workspace.FeedBlob(
            "tensor",
            np.random.rand(4, 2, 3, 3, 5).astype(np.float32)
        )

        # Testing with axes undefined
        brew.transpose(
            model,
            ["tensor"],
            "transpose",
        )
        self.InferTensorRunAndCompare(model)

        # Testing with axes defined
        brew.transpose(
            model,
            ["tensor"],
            "transpose",
            axes=np.random.permutation(5)
        )

        return self.InferTensorRunAndCompare(model)

    def testShapeInferencePad(self):
        model = model_helper.ModelHelper(name="padtest")
        model.PadImage("data", 'padded', pad_t=100, pad_l=37, pad_b=28,
                       pad_r=20, mode="constant", order="NCHW")

        workspace.FeedBlob(
            "data",
            np.random.rand(16, 3, 228, 228).astype(np.float32),
        )

        self.InferTensorRunAndCompare(model)

    def testShapeInferenceTwoClass(self):
        model = model_helper.ModelHelper(name="twoclass")
        model.MakeTwoClass("v", "v2")
        workspace.FeedBlob("v", np.random.rand(32).astype(np.float32))
        self.InferTensorRunAndCompare(model)

    def testShapeInferencePadZero(self):
        model = model_helper.ModelHelper(name="padtest")
        model.PadImage("data", 'padded', pad=0, mode="constant",
                       order="NCHW")

        workspace.FeedBlob(
            "data",
            np.random.rand(16, 3, 228, 228).astype(np.float32),
        )

        self.InferTensorRunAndCompare(model)

    def testShapeInferenceMatMul(self):
        model = model_helper.ModelHelper(name="test_model")

        model.MatMul(["x", "y"], "MatMul")

        workspace.FeedBlob("x", np.random.rand(10, 5).astype(np.float32))
        workspace.FeedBlob("y", np.random.rand(5, 10).astype(np.float32))

        self.InferTensorRunAndCompare(model)

    def testShapeInferenceSoftmaxWithLoss(self):
        model = model_helper.ModelHelper(name="test_model")

        model.SoftmaxWithLoss(
            ["logits", "labels"],
            ["softmax", "loss"],
        )

        # 2D Shape of [batch_size, num_classes]
        workspace.FeedBlob(
            "logits",
            np.random.rand(4, 3).astype(np.float32),
        )

        # Shape of size batch_size with all values [0, num_classes)
        workspace.FeedBlob(
            "labels",
            np.random.randint(low=0, high=3, size=(4, 1)).astype(np.int32),
        )
        self.InferTensorRunAndCompare(model)

        # Testing with 1D labels arg
        workspace.FeedBlob(
            "logits",
            np.random.rand(4, 3).astype(np.float32),
        )

        workspace.FeedBlob(
            "labels",
            np.random.randint(low=0, high=3, size=4).astype(np.int32),
        )
        self.InferTensorRunAndCompare(model)

        # Testing with weight_tensor
        model.SoftmaxWithLoss(
            ["logits", "labels", "weight_tensor"],
            ["softmax", "loss"],
        )

        workspace.FeedBlob(
            "logits",
            np.random.rand(4, 3).astype(np.float32),
        )

        workspace.FeedBlob(
            "labels",
            np.random.randint(low=0, high=3, size=4).astype(np.int32),
        )

        workspace.FeedBlob(
            "weight_tensor",
            np.random.rand(4).astype(np.float32),
        )
        self.InferTensorRunAndCompare(model)

        # Test spatial model
        model = model_helper.ModelHelper(name="test_model")
        workspace.FeedBlob(
            "img",
            np.random.rand(32, 19, 33, 28).astype(np.float32)
        )
        workspace.FeedBlob(
            "img_labels",
            (np.random.rand(32, 33, 28) * 19).astype(np.int32)
        )
        model.SpatialSoftmaxWithLoss(
            ["img", "img_labels"],
            ["softmax_img", "loss"],
        )
        self.InferTensorRunAndCompare(model)

    def testShapeInferenceIm2Col(self):
        # Test with NCHW
        model = model_helper.ModelHelper(name="test_model")
        model.Im2Col("X", "Y", pad=1, kernel=4, dilation=2, stride=2,
                     order="NCHW")

        workspace.FeedBlob(
            "X",
            np.random.rand(16, 3, 228, 228).astype(np.float32),
        )

        self.InferTensorRunAndCompare(model)

        # Test with NHWC
        model = model_helper.ModelHelper(name="test_model")
        model.Im2Col("X", "Y", pad=1, kernel=4, dilation=2, stride=2,
                     order="NHWC")

        workspace.FeedBlob(
            "X",
            np.random.rand(16, 228, 228, 3).astype(np.float32),
        )

        self.InferTensorRunAndCompare(model)

        # Test with different width and height
        model = model_helper.ModelHelper(name="test_model")
        model.Im2Col("X", "Y", pad=1, kernel_h=8, kernel_w=4,
                     dilation=2, stride=2)

        workspace.FeedBlob(
            "X",
            np.random.rand(16, 3, 228, 114).astype(np.float32),
        )

        self.InferTensorRunAndCompare(model)

    def testShapeInferenceTile(self):
        m = model_helper.ModelHelper(name="test_model")

        workspace.FeedBlob(
            "tensor",
            np.random.rand(4, 2, 3, 3, 5).astype(np.float32)
        )

        # Testing with axes undefined
        for i in range(0, 4):
            m.net.Tile(
                "tensor", "tiled_tensor_{}".format(i), tiles=5, axis=i)
        self.InferTensorRunAndCompare(m)