Domain adaptation helps transfer the knowledge gained from a labeled source\ndomain to an unlabeled target domain. During the past few years, different\ndomain adaptation techniques have been published. One common flaw of these\napproaches is that while they might work well on one input type, such as\nimages, their performance drops when applied to others, such as text or\ntime-series. In this paper, we introduce Proportional Progressive Pseudo\nLabeling (PPPL), a simple, yet effective technique that can be implemented in a\nfew lines of code to build a more general domain adaptation technique that can\nbe applied on several different input types. At the beginning of the training\nphase, PPPL progressively reduces target domain classification error, by\ntraining the model directly with pseudo-labeled target domain samples, while\nexcluding samples with more likely wrong pseudo-labels from the training set\nand also postponing training on such samples. Experiments on 6 different\ndatasets that include tasks such as anomaly detection, text sentiment analysis\nand image classification demonstrate that PPPL can beat other baselines and\ngeneralize better.\n