Provided is a numerical controller capable of easily controlling a pressure without a pressure sensor. The numerical controller estimates a pressure based on at least one of a command value and a feedback value. A machine learning device for learning the pressure corresponding to the at least one of the command value and the feedback value is included. The machine learning device includes a state observation unit for observing the at least one of the command value and the feedback value as a state variable, a label data acquisition unit for acquiring label data indicating the pressure, and a learning unit for associating and learning the state variable with the label data.