diff --git a/dbms/src/AggregateFunctions/AggregateFunctionMLMethod.h b/dbms/src/AggregateFunctions/AggregateFunctionMLMethod.h
index 650665fc673..0c443a859f8 100644
--- a/dbms/src/AggregateFunctions/AggregateFunctionMLMethod.h
+++ b/dbms/src/AggregateFunctions/AggregateFunctionMLMethod.h
@@ -29,6 +29,104 @@ namespace ErrorCodes
     extern const int BAD_ARGUMENTS;
 }
 
+class IGradientComputer
+{
+public:
+    virtual ~IGradientComputer()
+    {}
+
+    void add(/* weights, */Float64 target, const IColumn ** columns, size_t row_num) final
+    {
+        ++cur_batch;
+        std::vector<Float64> cur_grad = compute_gradient(/*...*/);
+        for (size_t i = 0; i != batch_gradient.size(); ++i) {
+            batch_gradient[i] += cur_grad[i];
+        }
+    }
+
+    std::vector<Float64> get() final
+    {
+        std::vector<Float64> result(batch_gradient.size());
+        for (size_t i = 0; i != batch_gradient.size(); ++i) {
+            result[i] = batch_gradient[i] / cur_batch;
+            batch_gradient[i] = 0.0;
+        }
+        cur_batch = 0;
+        return result;
+    }
+
+protected:
+    virtual std::vector<Float64> compute_gradient(/* weights, */Float64 target, const IColumn ** columns, size_t row_num) = 0;
+
+private:
+    UInt32 cur_batch = 0;
+    std::vector<Float64> batch_gradient;  // gradient for bias lies in batch_gradient[batch_gradient.size() - 1]
+};
+
+class LinearRegression : public IGradientComputer
+{
+public:
+    virtual ~LinearRegression()
+    {}
+
+protected:
+    virtual std::vector<Float64> compute_gradient(/* weights, */Float64 target, const IColumn ** columns, size_t row_num)
+    {
+        // TODO
+    }
+};
+
+class IWeightsUpdater
+{
+public:
+    virtual ~IWeightsUpdater()
+    {}
+
+    virtual void update(/* weights, gradient */) = 0;
+};
+
+class GradientDescent : public IWeightsUpdater
+{
+public:
+    virtual ~GradientDescent()
+    {}
+
+    virtual void update(/* weights, gradient */) = 0 {
+        // TODO
+    }
+};
+
+struct LinearModelData
+{
+    LinearModelData()
+    {}
+
+    LinearModelData(Float64 learning_rate_, UInt32 param_num_, UInt32 batch_size_)
+    : learning_rate(learning_rate_), batch_size(batch_size_) {
+        weights.resize(param_num_, Float64{0.0});
+        batch_gradient.resize(param_num_ + 1, Float64{0.0});
+        cur_batch = 0;
+    }
+
+    std::vector<Float64> weights;
+    Float64 bias{0.0};
+    std::shared_ptr<IGradientComputer> gradient_computer;
+    std::shared_ptr<IWeightsUpdater> weights_updater;
+
+    void add(Float64 target, const IColumn ** columns, size_t row_num)
+    {
+        gradient_cumputer->add(target, columns, row_num);
+        if (cur_batch == batch_size)
+        {
+            cur_batch = 0;
+            weights_updater->update(/* weights */, gradient_computer->get());
+        }
+    }
+
+    void merge(const LinearModelData & rhs)
+    {
+    }
+};
 
 struct LinearRegressionData
 {