# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import unittest
from copy import deepcopy

import numpy as np
from op_test import (
    OpTest,
    get_device,
    get_device_place,
    get_devices,
    get_places,
    is_custom_device,
)

import paddle
from paddle import base
from paddle.framework import core

RTOL = 1e-06
ATOL = 1e-06


def radam_step(inputs, attributes, dtype="float32"):
    param = inputs['param']
    grad = inputs['grad']
    lr = inputs['learning_rate']

    # accumulators
    beta1_pow = inputs['beta1_pow']
    beta2_pow = inputs['beta2_pow']
    rho = inputs['rho']
    moment1 = inputs['moment1']
    moment2 = inputs['moment2']

    # attrs
    epsilon = attributes['epsilon']
    beta1 = attributes['beta1']
    beta2 = attributes['beta2']

    rho_inf = 2 / (1 - beta2) - 1

    beta1_pow *= beta1
    beta2_pow *= beta2

    rho = (rho * (beta2 - beta2_pow) + beta2_pow) / (1 - beta2_pow)

    moment1 = beta1 * moment1 + (1.0 - beta1) * grad
    moment2 = beta2 * moment2 + (1.0 - beta2) * grad * grad

    moment1_hat = moment1 / (1 - beta1_pow)

    rho_t = rho_inf - 2 * rho

    if rho_t.reshape(-1)[0] > 5:
        l_t = np.sqrt(1 - beta2_pow) / (np.sqrt(moment2) + epsilon)
        r_t = np.sqrt(
            ((rho_t - 4) * (rho_t - 2) * rho_inf)
            / ((rho_inf - 4) * (rho_inf - 2) * rho_t)
        )

        param = param - lr * moment1_hat * r_t * l_t
    else:
        param = param - lr * moment1_hat

    # get accumulators
    return (
        param.astype(dtype),
        beta1_pow.astype(dtype),
        beta2_pow.astype(dtype),
        rho.astype(dtype),
        moment1.astype(dtype),
        moment2.astype(dtype),
    )


def radam_wrapper(
    param,
    grad,
    lr,
    beta1_pow,
    beta2_pow,
    rho,
    moment1,
    moment2,
    master_param=None,
    beta1=0.9,
    beta2=0.999,
    epsilon=1e-8,
    multi_precision=False,
):
    _, _, _, _, _, _, _ = paddle._C_ops.radam_(
        param,
        grad,
        lr,
        beta1_pow,
        beta2_pow,
        rho,
        moment1,
        moment2,
        master_param,
        beta1,
        beta2,
        epsilon,
        multi_precision,
    )


class TestRAdamOp(OpTest):
    def _init_param(self):
        self.beta1 = 0.78
        self.beta2 = 0.915
        self.epsilon = 1e-8

    def setUp(self):
        '''Test RAdam Op with supplied attributes'''
        np.random.seed(2024)

        self.op_type = "radam"
        self.python_api = radam_wrapper
        self.python_out_sig = ['out']

        param = np.random.uniform(-1, 1, (102, 105)).astype("float32")
        grad = np.random.uniform(-1, 1, (102, 105)).astype("float32")
        learning_rate = np.array(0.003).astype("float32")

        self._init_param()

        # accumulators
        beta1_pow = (np.ones((102, 105)) * (self.beta1**3)).astype("float32")
        beta2_pow = (np.ones((102, 105)) * (self.beta2**3)).astype("float32")

        rho_inf = 2 / (1 - self.beta2) - 1
        rho = (np.ones((102, 105)) * self._init_rho(rho_inf)).astype("float32")

        moment1 = np.random.uniform(-1, 1, (102, 105)).astype("float32")
        # The second moment is positive
        moment2 = np.random.random((102, 105)).astype("float32")

        self.inputs = {
            "param": param,
            "grad": grad,
            "beta1_pow": beta1_pow,
            "beta2_pow": beta2_pow,
            "rho": rho,
            "moment1": moment1,
            "moment2": moment2,
            "learning_rate": learning_rate,
        }

        self.attrs = {
            "epsilon": self.epsilon,
            "beta1": self.beta1,
            "beta2": self.beta2,
        }

        (
            param_out,
            beta1_pow_out,
            beta2_pow_out,
            rho_out,
            moment1_out,
            moment2_out,
        ) = radam_step(
            deepcopy(self.inputs), deepcopy(self.attrs), dtype="float32"
        )

        self.outputs = {
            "param_out": param_out,
            "beta1_pow_out": beta1_pow_out,
            "beta2_pow_out": beta2_pow_out,
            "rho_out": rho_out,
            "moment1_out": moment1_out,
            "moment2_out": moment2_out,
        }

    def _init_rho(self, rho_inf):
        return np.array((rho_inf - 5) / 2 + 5.0).astype("float32")

    def test_check_output(self):
        self.check_output(check_pir=True, rtol=RTOL, atol=ATOL)


class TestRAdamOpWithDefault(TestRAdamOp):
    def _init_param(self):
        self.beta1 = 0.9
        self.beta2 = 0.999
        self.epsilon = 1.0e-8


class TestRAdamOpRhoSmall(TestRAdamOp):
    def _init_rho(self, rho_inf):
        return np.array((rho_inf - 5) / 2 - 5.0).astype("float32")


@unittest.skipIf(
    not (core.is_compiled_with_cuda() or is_custom_device()),
    "core is not compiled with CUDA",
)
class TestRAdamOpGPU(TestRAdamOp):
    def test_check_output(self):
        self.check_output_with_place(
            get_device_place(), check_pir=True, rtol=RTOL, atol=ATOL
        )


class TestRAdamOpGPURhoSmall(TestRAdamOpGPU):
    def _init_rho(self, rho_inf):
        return np.array((rho_inf - 5) / 2 - 5.0).astype("float32")


class TestRAdamOpMultipleSteps(TestRAdamOp):
    num_steps = 10

    def test_check_output(self):
        for _ in range(self.num_steps):
            (
                param_out,
                beta1_pow_out,
                beta2_pow_out,
                rho_out,
                moment1_out,
                moment2_out,
            ) = radam_step(
                deepcopy(self.inputs), deepcopy(self.attrs), dtype="float32"
            )

            self.outputs = {
                "param_out": param_out,
                "beta1_pow_out": beta1_pow_out,
                "beta2_pow_out": beta2_pow_out,
                "rho_out": rho_out,
                "moment1_out": moment1_out,
                "moment2_out": moment2_out,
            }
            # Verify output for this step
            self.check_output()

            # Output of this step becomes input for next step
            self.inputs['param'] = param_out
            self.inputs['beta1_pow'] = beta1_pow_out
            self.inputs['beta2_pow'] = beta2_pow_out
            self.inputs['rho'] = rho_out
            self.inputs['moment1'] = moment1_out
            self.inputs['moment2'] = moment2_out

            # Randomize gradient for next step
            self.inputs['grad'] = np.random.uniform(-1, 1, (102, 105)).astype(
                "float32"
            )


class TestRAdamAPI(unittest.TestCase):
    def test_radam_dygraph(self):
        paddle.disable_static()
        value = np.arange(26).reshape(2, 13).astype("float32")
        a = paddle.to_tensor(value)
        linear = paddle.nn.Linear(13, 5)
        radam = paddle.optimizer.RAdam(
            learning_rate=0.01,
            parameters=linear.parameters(),
            weight_decay=0.01,
        )

        for _ in range(2):
            out = linear(a)
            out.backward()
            radam.step()
            radam.clear_gradients()

    def test_radam_apply_gradients(self):
        paddle.disable_static()
        value = np.arange(26).reshape(2, 13).astype("float32")
        a = paddle.to_tensor(value)
        linear = paddle.nn.Linear(13, 5)
        radam = paddle.optimizer.RAdam(
            learning_rate=0.01,
            parameters=linear.parameters(),
            weight_decay=0.01,
        )

        for _ in range(2):
            out = linear(a)
            loss = paddle.mean(out)
            param_grads = radam.backward(loss)
            radam.apply_gradients(param_grads)
            radam.clear_gradients()

    def test_radam_static(self):
        paddle.enable_static()
        place = base.CPUPlace()
        shape = [2, 3, 8, 8]
        exe = base.Executor(place)
        train_prog = base.Program()
        startup = base.Program()
        with (
            base.program_guard(train_prog, startup),
            base.unique_name.guard(),
        ):
            data = paddle.static.data(name="data", shape=shape)
            hidden = paddle.static.nn.fc(x=data, size=10)
            loss = paddle.mean(hidden)

            beta1 = 0.85
            beta2 = 0.95
            opt = paddle.optimizer.RAdam(
                learning_rate=1e-5,
                beta1=beta1,
                beta2=beta2,
                weight_decay=0.01,
                epsilon=1e-8,
            )
            opt.minimize(loss)

        exe.run(startup)
        data_np = np.random.random(shape).astype("float32")
        rets = exe.run(train_prog, feed={"data": data_np}, fetch_list=[loss])
        assert rets[0] is not None
        paddle.disable_static()

    def test_pir_radam(self):
        with paddle.pir_utils.IrGuard():
            place = base.CPUPlace()
            shape = [2, 3, 8, 8]
            exe = base.Executor(place)
            train_prog = paddle.static.Program()
            startup = paddle.static.Program()
            with (
                paddle.static.program_guard(train_prog, startup),
                base.unique_name.guard(),
            ):
                data = paddle.static.data(name="data", shape=shape)
                hidden = paddle.static.nn.fc(x=data, size=10)
                loss = paddle.mean(hidden)

                beta1 = 0.85
                beta2 = 0.95
                opt = paddle.optimizer.RAdam(
                    learning_rate=1e-5,
                    beta1=beta1,
                    beta2=beta2,
                    weight_decay=0.01,
                    epsilon=1e-8,
                )
                opt.minimize(loss)

            exe.run(startup)
            data_np = np.random.random(shape).astype("float32")
            rets = exe.run(
                train_prog, feed={"data": data_np}, fetch_list=[loss]
            )
            assert rets[0] is not None

    def test_radam_invalid_input(self):
        paddle.disable_static()
        linear = paddle.nn.Linear(10, 10)
        with self.assertRaises(ValueError):
            _ = paddle.optimizer.RAdam(
                learning_rate=-1, parameters=linear.parameters()
            )
        with self.assertRaises(ValueError):
            _ = paddle.optimizer.RAdam(
                0.1, beta1=-1, parameters=linear.parameters()
            )
        with self.assertRaises(ValueError):
            _ = paddle.optimizer.RAdam(
                0.1, beta2=-1, parameters=linear.parameters()
            )
        with self.assertRaises(ValueError):
            _ = paddle.optimizer.RAdam(
                0.1, beta1=2.0, parameters=linear.parameters()
            )
        with self.assertRaises(ValueError):
            _ = paddle.optimizer.RAdam(
                0.1, beta2=2.0, parameters=linear.parameters()
            )
        with self.assertRaises(ValueError):
            _ = paddle.optimizer.RAdam(
                0.1, epsilon=-1, parameters=linear.parameters()
            )


class TestRAdamAPIWeightDecay(unittest.TestCase):
    def test_weight_decay_int(self):
        paddle.disable_static()
        value = np.arange(26).reshape(2, 13).astype("float32")
        a = paddle.to_tensor(value)
        linear = paddle.nn.Linear(13, 5)
        radam = paddle.optimizer.RAdam(
            learning_rate=0.01,
            parameters=linear.parameters(),
            weight_decay=1,
        )

        for _ in range(2):
            out = linear(a)
            out.backward()
            radam.step()
            radam.clear_gradients()


class TestRAdamAPIGroup(TestRAdamAPI):
    def test_radam_dygraph(self):
        paddle.disable_static()
        value = np.arange(26).reshape(2, 13).astype("float32")
        a = paddle.to_tensor(value)
        linear_1 = paddle.nn.Linear(13, 5)
        linear_2 = paddle.nn.Linear(5, 3)
        radam = paddle.optimizer.RAdam(
            learning_rate=0.01,
            parameters=[
                {'params': linear_1.parameters()},
                {'params': linear_2.parameters(), 'weight_decay': 0.001},
            ],
            weight_decay=0.01,
        )

        for _ in range(2):
            out = linear_1(a)
            out = linear_2(out)
            out.backward()
            radam.step()
            radam.clear_gradients()


class TestRAdamMultiPrecision(unittest.TestCase):
    def _test_radam_dygraph_place_amp(self, place, use_amp=False):
        paddle.disable_static()
        paddle.seed(10)
        paddle.set_device(place)

        input = paddle.randn((5, 5))

        model = paddle.nn.Linear(5, 5)

        optimizer = paddle.optimizer.RAdam(
            parameters=[
                {
                    'params': model.parameters(),
                    'weight_decay': 0.001,
                    'beta1': 0.1,
                    'beta2': 0.99,
                }
            ],
        )

        optimizer._multi_precision = use_amp

        for _ in range(2):
            if place == get_device() and use_amp:
                model = paddle.amp.decorate(models=model, level='O2')
                scaler = paddle.amp.GradScaler(init_loss_scaling=1024)

            if place == get_device() and use_amp:
                with paddle.amp.auto_cast(level='O2'):
                    output = model(input)
                    loss = paddle.mean(output)
                scaled = scaler.scale(loss)
                scaled.backward()
                scaler.step(optimizer)
                optimizer.clear_grad()
            else:
                output = model(input)
                loss = paddle.mean(output)
                loss.backward()
                optimizer.step()
                optimizer.clear_grad()

    def test_main(self):
        for place in get_devices():
            use_amp_list = [True, False]
            for use_amp in use_amp_list:
                self._test_radam_dygraph_place_amp(place, use_amp)


class TestNdamaxMultiPrecision2_0(unittest.TestCase):
    def dygraph_radam_mp(self, mp, use_amp):
        paddle.disable_static()
        paddle.seed(2024)
        paddle.set_device(get_device())
        input = paddle.randn((2, 2))
        model = paddle.nn.Linear(2, 2)
        optimizer = paddle.optimizer.RAdam(0.1, parameters=model.parameters())
        optimizer._multi_precision = mp
        if use_amp:
            model = paddle.amp.decorate(models=model, level='O2')
            scaler = paddle.amp.GradScaler(init_loss_scaling=1024)

        for idx in range(5):
            if use_amp:
                with paddle.amp.auto_cast(level='O2'):
                    output = model(input)
                    loss = paddle.mean(output)
                scaled = scaler.scale(loss)
                scaled.backward()
                scaler.minimize(optimizer, scaled)
                optimizer.clear_grad()
            else:
                output = model(input)
                loss = paddle.mean(output)
                loss.backward()
                optimizer.step()
                optimizer.clear_grad()

        return output, model.parameters()

    def static_radam_mp(self, mp, use_amp):
        paddle.seed(2024)
        paddle.enable_static()
        exe = paddle.static.Executor(get_device_place())
        train_program = paddle.static.Program()
        startup_program = paddle.static.Program()

        with paddle.static.program_guard(train_program, startup_program):
            linear = paddle.nn.Linear(2, 10)
            optimizer = paddle.optimizer.NAdam(
                learning_rate=0.1, parameters=linear.parameters()
            )
            if use_amp:
                data = paddle.static.data('X', [2, 2], 'float16')
                if paddle.framework.in_pir_mode():
                    linear, optimizer = paddle.amp.decorate(
                        models=linear,
                        optimizers=optimizer,
                        level='O2',
                        master_grad=False,
                    )
                else:
                    optimizer = paddle.static.amp.decorate(
                        optimizer,
                        init_loss_scaling=128.0,
                        use_dynamic_loss_scaling=True,
                        use_pure_fp16=True,
                        use_fp16_guard=False,
                    )
            else:
                data = paddle.static.data('X', [2, 2], 'float32')

            out = linear(data)
            loss = paddle.mean(out)
            optimizer.minimize(loss)

        exe.run(startup_program)

        np.random.seed(2024)
        if use_amp:
            optimizer.amp_init(
                place=get_device_place(), scope=paddle.static.global_scope()
            )
            x = np.random.random(size=(2, 2)).astype('float16')
        else:
            x = np.random.random(size=(2, 2)).astype('float32')
        out = []
        for idx in range(5):
            (loss_data,) = exe.run(
                train_program, feed={"X": x}, fetch_list=[loss]
            )
            out.append(loss_data)

        return out

    def test_main(self):
        if not (paddle.is_compiled_with_cuda() or is_custom_device()):
            return
        "Test dygraph mode"
        output1_dy, params1_dy = self.dygraph_radam_mp(use_amp=True, mp=True)
        output2_dy, params2_dy = self.dygraph_radam_mp(use_amp=False, mp=False)
        np.testing.assert_allclose(
            output1_dy.astype('float32').numpy(),
            output2_dy.astype('float32').numpy(),
            rtol=1e-05,
            atol=0.1,
        )
        for idx in range(len(params1_dy)):
            np.testing.assert_allclose(
                params1_dy[idx].astype('float32').numpy(),
                params2_dy[idx].astype('float32').numpy(),
                rtol=1e-05,
                atol=0.1,
            )
        "Test static mode"
        output1_st = self.static_radam_mp(use_amp=True, mp=True)
        output2_st = self.static_radam_mp(use_amp=False, mp=False)
        for idx in range(len(output1_st)):
            np.testing.assert_allclose(
                output1_st[idx].astype('float32'),
                output2_st[idx].astype('float32'),
                rtol=1e-05,
                atol=0.1,
            )


class TestRAdamGroupWithLR(TestRAdamAPI):
    def test_radam(self):
        paddle.disable_static()
        value = np.arange(26).reshape(2, 13).astype("float32")
        a = paddle.to_tensor(value)
        linear_1 = paddle.nn.Linear(13, 5)
        linear_2 = paddle.nn.Linear(5, 3)
        radam = paddle.optimizer.RAdam(
            learning_rate=paddle.optimizer.lr.PiecewiseDecay(
                boundaries=[3, 6], values=[0.1, 0.2, 0.3]
            ),
            parameters=[
                {
                    'params': linear_1.parameters(),
                    'learning_rate': 0.1,
                },
                {
                    'params': linear_2.parameters(),
                    'weight_decay': 0.001,
                },
            ],
            weight_decay=0.01,
        )

        for _ in range(2):
            out = linear_1(a)
            out = linear_2(out)
            out.backward()
            radam.step()
            radam.clear_gradients()


def main_test_func(place, dtype):
    paddle.enable_static()
    main = base.Program()
    startup = base.Program()
    with (
        base.program_guard(main, startup),
        base.scope_guard(base.Scope()),
    ):
        x = paddle.static.data(name='x', shape=[None, 13], dtype=dtype)
        y = paddle.static.data(name='y', shape=[None, 1], dtype=dtype)
        y_predict = paddle.static.nn.fc(x, size=1)
        cost = paddle.nn.functional.square_error_cost(input=y_predict, label=y)
        avg_cost = paddle.mean(cost)

        radam_optimizer = paddle.optimizer.RAdam(0.01)
        radam_optimizer.minimize(avg_cost)

        fetch_list = [avg_cost]
        train_reader = list(
            zip(
                np.random.rand(101, 13),
                np.random.randint(12, size=(101, 1)),
            )
        )
        feeder = base.DataFeeder(place=place, feed_list=[x, y])
        exe = base.Executor(place)
        exe.run(base.default_startup_program())
        for data in train_reader:
            exe.run(main, feed=feeder.feed([data]), fetch_list=fetch_list)

    paddle.disable_static()


class RAdamFp32Test(unittest.TestCase):
    def setUp(self):
        self.dtype = "float32"

    def test_main(self):
        for p in get_places():
            main_test_func(p, self.dtype)


class RAdamFp64Test(RAdamFp32Test):
    def setUp(self):
        self.dtype = 'float64'


if __name__ == "__main__":
    unittest.main()